U.S. flag

An official website of the United States government

The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

  • Publications
  • Account settings
  • Advanced Search
  • Journal List

Logo of plosone

Three novel bird strike likelihood modelling techniques: The case of Brisbane Airport, Australia

Robert Andrews

1 School of Information Systems, Faculty of Science, Queensland University of Technology, Brisbane, Queensland, Australia

Bayan Bevrani

Brigitte colin, moe t. wynn, arthur h. m. ter hofstede, jackson ring.

2 Wildlife Management Group, Brisbane Airport Corporation, Brisbane, Queensland, Australia

Associated Data

All relevant data are within the paper and its Supporting information files. A GitHub repository has also been created to make larger files available to reviewers and readers https://github.com/robertandrews59/ABCBirdstrike ”.

The risk posed by wildlife to air transportation is of great concern worldwide. In Australia alone, 17,336 bird-strike incidents and 401 animal-strike incidents were reported to the Air Transport Safety Board (ATSB) in the period 2010-2019. Moreover, when collisions do occur, the impact can be catastrophic (loss of life, loss of aircraft) and involve significant cost to the affected airline and airport operator (estimated at globally US$1.2 billion per year). On the other side of the coin, civil aviation, and airport operations have significantly affected bird populations. There has been an increasing number of bird strikes, generally fatal to individual birds involved, reported worldwide (annual average of 12,219 reported strikes between 2008-2015 being nearly double the annual average of 6,702 strikes reported 2001-2007) (ICAO, 2018). Airport operations including construction of airport infrastructure, frequent take-offs and landings, airport noise and lights, and wildlife hazard management practices aimed at reducing risk of birdstrike, e.g., spraying to remove weeds and invertebrates, drainage, and even direct killing of individual hazard species, may result in habitat fragmentation, population decline, and rare bird extinction adjacent to airports (Kelly T, 2006; Zhao B, 2019; Steele WK, 2021). Nevertheless, there remains an imperative to continually improve wildlife hazard management methods and strategies so as to reduce the risk to aircraft and to bird populations. Current approved wildlife risk assessment techniques in Australia are limited to ranking of identified hazard species, i.e., are ‘static’ and, as such, do not provide a day-to-day risk/collision likelihood. The purpose of this study is to move towards a dynamic, evidence-based risk assessment model of wildlife hazards at airports. Ideally, such a model should be sufficiently sensitive and responsive to changing environmental conditions to be able to inform both short and longer term risk mitigation decisions. Challenges include the identification and quantification of contributory risk factors, and the selection and configuration of modelling technique(s) that meet the aforementioned requirements. In this article we focus on likelihood of bird strike and introduce three distinct, but complementary, assessment techniques, i.e., A lgebraic, B ayesian, and C lustering (ABC) for measuring the likelihood of bird strike in the face of constantly changing environmental conditions. The ABC techniques are evaluated using environment and wildlife observations routinely collected by the Brisbane Airport Corporation ( BAC ) wildlife hazard management team. Results indicate that each of the techniques meet the requirements of providing dynamic, realistic collision risks in the face of changing environmental conditions.

Introduction

The risk posed by wildlife to air transportation is of great concern worldwide. Wildlife Hazard Management (WHM) is a mandatory procedure in every significant international and national airport [ 1 ]. A hazard is defined as “the condition or circumstance that could lead to damage or destruction of an aircraft, or to loss of life as a result of aircraft operations” [ 2 ]. Statistics and data on wildlife collisions with aircraft give some indication of the scale of the problem. In Australia alone, 17,336 bird-strikes occurrences and 401 animal-strike incidents were reported to the Australian Transport Safety Bureau between 2010 and 2019 [ 3 ] and 17,360 strikes reported to the US Federal Aviation Authority (FAA) in a single year (2019) [ 4 ]. Wildlife strikes can have catastrophic consequences. The FAA report more than 301 people killed and over 298 aircraft destroyed globally as a consequence of wildlife strikes since 1988 [ 5 ]. It is estimated that wildlife strikes incur cost to the civil aviation industry in the US of approximately $625 million per year [ 6 ] and globally US$1.2 billion per year [ 7 ]. On the other side of the coin, civil aviation, and airport operations have significantly affected bird populations. There has been an increasing number of bird strikes, generally fatal to individual birds involved, reported worldwide (annual average of 12,219 reported strikes between 2008–2015 being nearly double the annual average of 6,702 strikes reported 2001–2007) [ 8 ]. Airport operations including construction of airport infrastructure, frequent take-offs and landings, airport noise and lights, and wildlife hazard management practices aimed at reducing risk of birdstrike, e.g., spraying to remove weeds and invertebrates, drainage, and even direct killing of individual hazard species, may result in habitat fragmentation, population decline, and rare bird extinction adjacent to airports [ 9 – 11 ]. Nevertheless, there remains an imperative to continually improve wildlife hazard management methods and strategies so as to reduce the risk to aircraft and to bird populations.

Certification or accreditation of an airport with the regulatory authority is contingent on the airport implementing a Wildlife Hazard Management Plan (WHMP) for the management of wildlife hazards. In Australia, compliance with the Civil Aviation Safety Authority (CASA’s) Advisory Circular AC 139–26(0) Wildlife Hazard Management at Aerodromes [ 12 ] requires of airport operators, that when assessing wildlife hazard risk, “individual species should be identified and prioritised in order of risk” [ 12 ]. AC 139–26(0) also points out the airport operator’s requirement to provide advice to airmen at varying levels of immediacy including (i) standing caution for a bird or animal hazard that poses a constant risk, (ii) periodic when there is a significantly increased risk, for a relatively constrained period of time, posed by a hazard species, and (iii) immediate advice of risk from the control tower to warn approaching and departing aircraft of, say, birds (or bats), detected in the flight path.

Essential components of a WHMP are risk assessment techniques or tools that (i) comply with regulations and, (ii) are sensitive enough to properly align risk with changing conditions around the airfield. Such a management plan (risk mitigation strategy) will incorporate reactive elements (such as providing warnings to pilots as they approach or depart from the airport regarding the presence of one or more hazard species) and proactive elements (such as altering habitat in the vicinity of the airport with a view to reducing hazard species’ populations over time).

Various wildlife hazard risk assessment approaches have been proposed/implemented around the world. Common to all, is the aim of reducing the likelihood of birdstrike, however differences in approaches characterised by their (i) sphere of application (local airport to (inter-)national), (ii) forecast horizon (static to real-time), and (iii) realisation (as being technique/algorithm, a model, or an implemented system). Here we take the definitions of technique (theoretical, mathematically based structure), model (framework of techniques and algorithms to describe/predict real word situation), and system (interacting components including models, sensors, communications networks to utlise model outputs) from [ 13 ]. The following models/systems are at large spatial scale, i.e., larger scale than individual airport. The models/systems differ also in their forecast horizon. For example, real-time birdstrike warning systems include the Avian Hazard Advisory System (AHAS) [ 14 ] in the United States, the Dutch Radar Observation of Bird Intensities system (Dutch ROBIN system— https://www.robinradar.com/ ) in the Netherlands. In Europe, the FlySafe Bird Avoidance Model [ 15 ] provides near real-time information and forecast on large scale bird movement in the air space of The Netherlands, Germany and Belgium. Longer forecast horizon model/systems include the United States and North America Bird Avoidance Model (USBAM) [ 16 ], the German birdstrike risk forecast model [ 13 ], and the Swiss/Dutch Dynamic Bird Migration Model [ 13 , 17 ]. The USBAM is a statistical model, while the German birdstrike risk forecast model is a conceptual model, and the Swiss/Dutch dynamic bird migration model is a simulation model. Purely static, wildlife species risk ranking techniques/algorithms include Allan’s formula [ 11 ], Paton’s approach [ 18 ], and Carter’s formula [ 19 ] These techniques are intended for application at individual airports.

Little attention has been paid to applying machine learning to aspects of birdstrike risk. Rosa et al. [ 20 ] applied 6 different machine learning techniques to data collected by marine radar in Portugal. This study concluded that all techniques were able to distinguish between birds and stationary objects, but distinguishing between different species of birds was less successful. Recently, Nimmagadda et al. [ 21 ] applied decision tree and naive Bayesian approaches to the problem of predicting whether an airline crash has occurred due to a birdstrike. Verma et al. [ 22 ] provides a literature review of various techniques for prediction of general aviation accidents including, among others, Bayesian Networks, Artificial Neural Networks, data mining, and ensemble approaches. None of these techniques were specific to predicting likelihood of wildlife collision on/near an airfield.

Risk assessment techniques/methods generally integrate the likelihood of occurrence of a risk event with the severity of an event (or any other risk-related factors) to give an overall risk rating. Examples of such techniques can be found in [ 19 , 23 , 24 ].

Allan’s formula [ 7 ], endorsed by the Australian Airports Association for use in Australian airports [ 25 ], assesses hazard species’ risk by combining the probability of the species being involved in a strike event (determined using only historical strike data) with the estimated severity of a strike (calculated from the average mass of an adult member of the hazard (bird) species adjusted for flocking behaviour). The principal function of Allan’s approach is to identify and rank hazard species. While being easy to implement in practice, Allan’s approach has limitations as a day-to-day indicator of risk due to wildlife including (i) any species has a zero probability of involvement in a collision (and hence a low risk) until a strike is actually recorded, and (ii) the risk is largely static since it is derived from a rolling window of the past 5 years’ observations, i.e. it does not take immediate conditions or short-term changes or trends into account.

Similar to Allan’s formula, Paton’s approach [ 18 ] is data-driven and uses a severity/likelihood matrix to assign a risk rating to a species (with however, significant differences to Allan’s formula in the way in which the severity and likelihood categories are calculated). In particular, the likelihood indicators used in Paton’s approach acknowledge that conditions on the airfield vary over time (e.g. number of birds on the airfield), making it sensitive to changing environmental conditions and potentially useful as the basis for a dynamic risk model.

It should be noted that the principal use of both Allan’s approach and Paton’s approach is in identifying and ranking hazard species. As such, both Allan’s formula and Paton’s approach are accepted by Australia’s Civil Aviation Safety Authority (CASA). Neither method however, is suitable for providing a risk assessment directly related to current conditions at the airport .

Further, static risk ranking approaches do not take into account seasonal or environmental factors, and hence, cannot be used to (i) reason about likely future states on the airfield, or (ii) to plan mitigation measures in anticipation of some likely future states. By contrast, the techniques developed in this study do provide evidence-based, dynamic, collision likelihood assessments, and can be used to inform wildlife hazard managers of likely, required, mitigation activities. Here we note that risk assessment also includes taking into account a severity measure. This aspect was outside the scope of this paper and we concern ourselves with only collision likelihood.

Materials and methods

Study area and data collection.

To develop these methods, a case study was conducted at the Brisbane International Airport, Australia located at the geographic coordinates 27°23’07.58” S, 153°07’13.48” E and an elevation of -3 m below mean sea level. It is located adjacent to the Brisbane coastline in a temperate (mesothermal) climate zone, defined as humid subtropical climate (Cfa) and temperate oceanic climate (Cfb) according to the Koeppen climate classification [ 26 ] system. The climate is characterised by hot and warm summers, without a dry season, although the average precipitation is higher in summer months than in winter months. Average temperatures range between 8°C in the coldest month (July) to 27°C in summer months. Within a 3km radius of the Brisbane airport natural features and wildlife attractants include the port of Brisbane, an oil refinery, refuse dumping ground, waste transfer station, a major golf club, big cemetery, wetlands reserve, waste water treatment plant, the Kedron Brook Floodway and several smaller brooks and creeks.

The (routinely collected) environmental, geographical and wildlife data used in this paper was provided by Brisbane Airport Corporation (BAC) in Australia and covers the period 1 May 2017 to 30 June 2019. The selection of variables was made in collaboration with our industry partner (BAC) and incorporated the expert knowledge of BAC’s wildlife management team. The data set comprised six main data aspects: (i) environmental data such as daily rainfall and temperature, (ii) existing infrastructure features such as runways, taxi ways, terminal buildings, etc., (iii) a proximity weighting factor for zones on the airfield (relative to the distance of the zone from areas of high-speed aircraft movement such as runways where strikes are most frequently reported), (iv) details of reported bird strikes during the 761 days (1 May 2017 to 30 June 2019), (v) daily counts of three bird species nominated as hazard species of interest by BAC, and (vi) details of mitigation (harassment) strategies conducted to reduce the number of birds adjacent to aircraft movement corridors.

Wildlife observations are carried out daily between 06:00 and 09:00 by the WHM team at BAC. The count is conducted by a single team member, (usually) traversing the airport on a standard route. The airport is divided into 11 zones. The daily count data records the date and time of wildlife observations, the species observed and the number of individuals of the species, as well as the location on the airport (zone) where the species was observed. Environmental variables such as rainfall, wind speed, wind direction, cloud height and cover are also recorded. Wildlife ‘harassment’ activities are also recorded (harassment is moving hazard species away from runways and taxi-ways). The date, time, and method (e.g. siren, pyrotechnics, etc.) of harassment activities are recorded along with the species and number of individuals harassed. Lastly, details of strikes are recorded including date and time, species and number involved, location (zone), aircraft, damage to aircraft, and whether the strike can be confirmed (by remains on the ground or evidence on the aircraft). S1 Table summarises the strike information for the three hazard species involved in the study during the period of the study.

BAC nominated the three hazard species it considered as being high risk species for inclusion in the study. The Cattle Egret ( Bubulcus ibis ) and Nankeen Kestrel ( Falco cenchroides ) were included due to their comparatively frequent involvement in collisions. The Straw-necked Ibis ( Threskiornis spinicollis ) was included due to the large numbers of this species (flocks of thousands of individuals) frequently observed on the airfield. The data set was a rectangular array—761 rows for each of three bird species, comprising 80 explanatory variables as a mix of continuous real values, categorical attributes and ordinal data.

Discussions with the BAC WHM team added some context to some relevant behaviours of the study hazard species.

  • Nankeen Kestrels are small, non-migratory, raptors (preying on small mammals and insects) which breed July-Nov, and exhibit a strong seasonal pattern in their mode of hunting [ 27 ], alternating between hover-hunting (riding thermals, particularly over concrete runways and taxiways where aircraft are moving) and still-hunting (launching from a perch). Nankeen Kestrels are the most frequently struck bird species at Brisbane airport, with strikes throughout the year.
  • Cattle Egrets are migratory and have high numbers from November till February (southern summer), with breeding occurring October till January with markedly fewer individuals observed in March–October (southern winter). Cattle Egrets feed on grasshoppers and other insects, foraging on open grassland.
  • The Straw-necked Ibis population on/around the airport is the most volatile and inconsistent of the three hazard species, and while numerous, are rarely involved in strike events. These fluctuations cannot be explained by specific criteria such as migratory behaviour, seasonality, or hunting behaviour. Instead, the Straw-necked Ibis population appears to be strongly linked to food availability (leading to birds being attracted to the airport precincts, and increased breeding).

The main objective of this study was to investigate a set of data-driven modelling techniques, from the point of view of generating dynamic collision likelihood scores. It was decided to test three (3) separate modelling approaches based, in part, on what has been described in the literature, and in part on novelty of approach, i.e. not described in the bird-strike literature but which has application in related data-driven analyses. The techniques included (i) an algebraic approach based on function approximation, (ii) a probabilistic approach using Bayesian networks, and (iii) an unsupervised machine learning approach utilising K-means clustering. Note that supervised machine learning techniques for modelling likelihood of bird strikes were not considered due to the small number of actual strikes recorded against each hazard species. Table 1 provides a summary of the essential features of each selected modelling approach.

The algebraic (function approximation) approach was to use variables used in Allan’s and Paton’s approaches, and variables identified in the literature as being indicative of collision risk, to develop an annual risk profile for each of the selected hazard species. The Bayesian network approach was intended to provide a likelihood rating of a strike involving a given hazard species under a given set of environmental conditions. K-means clustering and associated cluster analysis was selected with a view to ascertaining whether it was possible to (i) separate (daily) observations into classes, (ii) to then characterise classes (and hence, species-day observation units) as being high or low risk, and (iii) use cluster analysis as a means of feature reduction/selection (identifying the most indicative variables of the 80 daily measurements).

Algebraic modelling

This approach is based on the following observations and assumptions.

  • Collisions between birds and aircraft only happen when they are both in the same (air)space at the same time.
  • While being rare in absolute terms (i.e. fewer than 10 strikes per 10,000 flights), strikes happen frequently enough for their occurrence to not be completely random.
  • There exist some combination(s) of (environmental) factors that lead to birds and aeroplanes being in the same space at the same time.
  • Some factors are so-called ‘lead’ indicators of likelihood of collision. For instance, season is associated with certain usual weather conditions (e.g. rainfall, temperature, hours of daylight, etc.) that influence ‘intermediate’ indicators such as grass coverage (which in turns influences insect population, etc.). Both lead and intermediate indicators impact on ‘immediate’ likelihood indicators such as numbers of birds on/near runways (see Fig 3).
  • These factors and the relationships between them can be discovered through data analysis of historical environmental (observation, weather, harassment and strike) data.

Let E be the set of all environmental factors and e 1 , e 2 , e 3 , … e n ∈ E be individual environmental factors. Let H be the set of hazard species and E h ⊆ E be the set of environmental factors relevant to a given hazard species h ∈ H . Then, there exists some function f : E × H → likelihood that maps environmental factors to likelihood of a collision. Further, for h 1 , h 2 ∈ H , it is likely that E h 1 ≠ E h 2 . That is, different environmental factors are relevant to different hazard species.

Paton, when describing his risk assessment model [ 18 ], outlined various likelihood indicators including (i) relative abundance, (ii) frequency of occurrence and or area of occurrence. Carter, in his study on risk assessment and prioritisation of wildlife hazards [ 19 ] includes ten likelihood indicators including (i) overall population of the wildlife species (in total number of individuals), (ii) location of the species with respect to flight operations, and (iii) number of reported strikes involving the species. Importantly, in [ 19 ], the author also mentions the ability to actually influence the species through wildlife control (harassment) as being relevant to likelihood of involvement in a collision. Lastly, analysis of our own K-means clustering results reveal the significance of the number of individuals belonging to hazard species observed and their proximity to runways (see Results…K-Means Clustering).

In this study, it was decided to include (i) a seasonality measure, (ii) abundance which we define as the number of hazard species counted in the daily count as well as the number of hazard species involved in harassment operations, (iii) overall count of wildlife and, (iv) location of observed hazard species with respect to flight operations (proximity to runways) as key variables to include as model inputs.

The seasonality measure, based on modular arithmetic, was derived by converting (linear) calendar days to solar days (polar co-ordinates) as follows. For some calendar date d and marker date m representing solar day 0, we define daydiff(d, m) as number of days offset of d from m . Let s be the seasonality index as:

Such an approach means that, if 1-Jan is used as the marker date, days at opposite ends of the calendar year (e.g. in January and December) are seasonally close, while dates in January and June are seasonally distant.

The airfield is divided into eleven wildlife management zones. A location indicator for each zone was derived based on the proximity of the zone from the active runways and approach/take-off pathways. Let Z be the set of zones on the airfield, and P be the set of all proximity measures. Let Z p ⊆ Z denotes the set of zones which have proximity measure p where p ∈ P .

The abundance of a hazard species was calculated on a daily basis as:

  • calculate the count of the individuals belonging to hazard species h ∈ H on any day d in zone z ∈ Z is sum of the individuals of the hazard species observed during the daily count plus the individuals of the hazard species harassed by zone; c o u n t h z = o b s e r v e d h z + h a r a s s e d h z (1)
  • calculate the count of the individuals belonging to hazard species h ∈ H per proximity area p ∈ P is sum of the count per zone having the same proximity value; c o u n t h p = ∑ z ∈ Z p c o u n t h z (2)
  • abundance is the sum of 1/proximity * count per proximity area a b u n d a n c e h = ∑ q ∈ P c o u n t h q q (3)

The effect of multiplying by 1/proximity is to give more weight to hazard species that are closer to runways and approach/take-off flight paths. For any hazard species h ∈ H , zone z ∈ Z , and proximity p ∈ P :

We note that (i) non-detection of individuals from the hazard species (either through not being observed in one or any zone during the daily count, or not being involved in any harassment activities on a given day) will result in c o u n t h z being 0 for each such zone on the airfield, and will be reflected in the c o u n t h p and abundance h values for the day, and (ii) overall wildlife abundance on the airfield may be calculated similarly as for a single species.

Paton [ 18 ] points to population and proximity of hazard species to runways, and seasonality as factors influencing likelihood of collision. In our approach, we have a single measure, abundance, which combines population and proximity, as well as a seasonality measure. Plotting (seasonality, abundance) pairs across the year gives an abundance distribution and allows us to fit a curve, expected abundance , for some hazard species h as e h = f ( s , a h ). In [ 28 , 29 ] techniques are described for constructive function approximation using Gaussian kernels and sigmoid functions respectively.

The graph of a Gaussian is a characteristic symmetric ‘bell curve’ shape in which the parameter a i represents the height of the curve’s peak, b i is the position of the center of the peak, and c i controls the width of the ‘bell’. Such a curve is useful in modelling peaks in population over time.

The graph of a sigmoid function has a characteristic ‘S’ shape. The difference of two sigmoids may be used to cut off a continuous range of the domain to form a ‘bump’ where the difference between the two sigmoid functions is close to the maximum value of the function over this range, and near the minimum value elsewhere [ 30 ]. Such a function is useful in modelling a population that is relatively stable over a period of time.

Difference between two sigmoids resulting in a ‘bump’

where the parameter a i represents the height of the bump (maximum value), b i is the centre of the bump, c i controls the width of the bump, and the parameter k i = controls the steepness of the bump (the range over which the function saturates from minimum to maximum value).

A kernel mix of a set of m gaussian kernels and n sigmoid kernels allows us to model both periods of the year where populations rise to a peak, and periods of the year where populations remain largely constant. Thus, we have the expected abundance function for some hazard species h at seasonality s as

We then define l h d ( s h d , a b u n d a n c e h d , e h d ) → { r | r ∈ { l o w , e l e v a t e d } } as the collision likelihood for some hazard species h on day d . We instantiate l h d through logistic regression to model likelihood of a strike involving the respective hazard species on any given day. The trained model was then analysed to determine (i) the importance of the individual inputs, and (ii) the threshold(s) of the attributes that contributed positively to the predicted likelihood. The modelling was done in Python using the sklearn library for the LogisticRegression model and coefficient importance, and the imblearn library for the SMOTE() method for oversampling of minority class (to deal with unbalanced dataset as strikes were observed only infrequently).

In its simplest form:

where a h d ∈ { s h d , a b u n d a n c e h d , e h d ) } represents the most significant attribute, and t a represents the threshold for a h d found through logistic regression modelling.

Bayesian networks

Bayesian Networks are a method for graphical representation and probabilistic calculation in uncertain and complex scenarios [ 31 ]. Bayesian Networks require (i) a set of random variables, (ii) the conditional relationships that exist between them, and (iii) their probability distributions. A Bayesian network is a directed acyclic graph (DAG) such that each node represents a random variable and has associated probabilistic information. Conditional relationships between variables (nodes) are represented by arcs (edges) joining a node to other nodes.

Thus, Bayesian Networks should have the capability of accounting for all the conditional relationships and uncertainties (such as changes in the number of and types of birds, weather, aircraft movements, runway usage, etc.) as they relate to the likelihood of bird strike making them a good approach for risk assessment. Bayesian approaches have been mentioned in relation to climate and environmental factors influencing forest management [ 32 ], for imputing missing values in a data set used for analysis of engine failure following bird strike [ 33 ], and for estimating structural damage to aircraft following a bird strike [ 34 ]. To date, however, Bayesian Networks have not been applied to assess factors influencing the likelihood of wildlife hazard strike occurrence at airports. To address this gap, we show how Bayesian Networks can be utilised in the context of bird strike likelihood.

Network construction utilised the variables shown in Table 2 . Note that as most algorithms for Bayesian Networks do not use continuous valued variables, these were discretised into a maximum of three states (‘low’, ‘medium’ and ‘high’) with ranges and cut-offs based on existing literature in the field, analysis of data, and expert input from the BAC WHM team. The prior knowledge on likelihood of strike is measured and derived from the recorded data. In [ 18 ] number of observed, location, and proximity to the runway were the variables that were identified as important factors that impact on the likelihood of a strike. As in our algebraic modelling approach, use location and distance from runway to derive a proximity weighting. Hence, we derived and calculated the likelihood using observed data as:

The Bayesian Networks were developed in GeNIe ( https://www.bayesfusion.com/genie/ ), by integrating all the parameters identified as having direct or/and indirect effect on the occurrence of strikes (see Table 2 ). Fig 1a–1c respectively show the design configuration of BNM for Nankeen Kestrel, Straw-necked Ibis, and Cattle Egret respectively.

An external file that holds a picture, illustration, etc.
Object name is pone.0277794.g001.jpg

(a) BNM design for Nankeen Kestrel, (b) BNM design for Straw-necked Ibis, (c) BNM design for Cattle Egret.

Models were validated using K-fold cross validation (K = 2). Then, the Expectation Maximization (EM) method was used to discover maximum-likelihood estimates for all Conditional Probability Table (CPTs) and for refitting the case file data to the final model while minimizing negative log likelihood. Parameters of the model are re-learned each fold, while the structure of the model remained fixed.

A Receiver Operating Characteristic (ROC) curve can efficiently express the quality of a Bayesian Network model and demonstrate “the theoretical limits of accuracy of the model on one plot” [ 35 ]. The ROC curve plots the true positive rate (Sensitivity) against the false positive rate (100-Specificity) considering multiple cut-off points of a parameter. ROC curves are able to express the quality of a model effectively. The ROC curve showcases an insight into the trade-off with selecting any states of a variable/node. The diagonal line on the plot illustrates a baseline ROC curve. In other words, a ROC curve above this diagonal line is a classifier that demonstrates that the model works effectively. The Area Under the Curve (AUC) is an easy way to identify the quality of the model with a number shown in each ROC diagram. Model quality was assessed by generating ROC curves for each BNM.

(K-Means) clustering

The rationale for applying K-means clustering [ 36 ] as a data exploration technique was (i) to determine whether airfield days could be clustered into day ‘types’, i.e. grouped by similarity, (ii) to identify the principal data attributes used to differentiate between day ‘types’, and (iii) to see whether particular day types are associated with bird strikes.

Accordingly, we took the following steps.

  • Applied unsupervised learning (clustering) to the BAC wildlife observational data to build a ‘codebook’ of day types. Note that details of strikes were not included in this phase.
  • Validated the models using internal and external validation.
  • Extracted indicative variables using principal component analysis.
  • Used bootstrapping and comparision of collision probabilities at 95% confidence intervals to determine prediction accuracy, i.e., if some day types were actually associated with strikes.

Unsupervised K-means partitions n observations into K clusters such that each observation is assigned to only one cluster, and observations in the same cluster are similar to each other as much as possible, and observations in different clusters are distinctly different. The K centroids serve as prototypes of the respective clusters to which they belong.

Internal validation of clustering may be assessed using silhouette plots. A silhouette plot [ 37 ] measures how well each individual observation is assigned to its respective cluster. The y-axis of a silhouette plot shows all the individual data points and the x-axis gives an indication if the data point is farthest away from neighbouring clusters (+1), or is on the decision boundary between two neighbouring clusters (0), or might have been assigned to the wrong cluster (-1). Also, the width the individual silhouette cluster plots and the average silhouette coefficient can give insight if the optimal number of clusters was determined correctly.

Principal Component Analysis (PCA) [ 38 ] gives insight into patterns and relationships that exist among the variables, and helps in reducing the complexity of high-dimensional and highly correlated data, while retaining as much of the variance in the dataset as possible. Keeping only the first two principal components finds the two-dimensional plane through the high-dimensional data set in which the data is most spread out. So if the data contains clusters these too may be most spread out, and therefore most visible to be plotted out in a two-dimensional diagram. As a final anlaysis, we applied PCA to identify variables significant in forming clusters.

Thus, in our approach, each cluster will represent a day ‘type’ and the principal components, identified by PCA, will give insights into the data attributes that differentiate between day types.

Lastly, by calculating the probability of collision on any given day with the probabilitity of collision given that the day belonged to a particular cluster and comparing (95%) confidence intervals for overlap, we determine whether day ‘types’ are associated with collisions for a particular hazard species. To deal with the low frequency of collisions recorded in the observational data, we use bootstrapping to derive the confidence intervals.

To optimise the data for presentation to the K-means algorithm, we applied data standardisation [ 39 , 40 ] to address issues of different data ranges, units of measure, and variance apparent in the variables.

Following clustering, internal and external validation measures were applied to determine the goodness of the clustering solution. Internal cluster validation measures reflect on the compactness, connectedness, and the separation of the cluster partitions, and can be determined by the silhouette coefficient (with range [-1..1] [ 37 ]) and the Dunn index (with range [0..∞) [ 41 ]). The goal is to maximise both the silhouette coefficient the Dunn index [ 42 ]. If the data set contains compact and well-separated clusters, the diameter of the clusters is expected to be small and the distance between the clusters is expected to be large (silhouette coefficient will be close to 1 and the Dunn index will be large). That is, we want the average distance within cluster to be as small as possible; and the average distance between clusters to be as large as possible.

The computational environment was the R statistical modelling software version 4.2.0 [ 43 ]. The get_clust_tendency() function was used to determine the Hopkins statistic [ 44 ] as an indicator of suitability of the data for clustering. The returned (average) value of 0.1, being close to zero, indicated the data was suitable for clustering. The fviz_nbclust() function, using each of the three methods provided for estimating the optimal number of clusters, gave recommended number of clusters, for each species, as two, i.e. K = 2. The kmeans() function was used as the implementation of K-means clustering. The clvalid() function was used to return internal and external measures of validity. The PCA() and get_pca_var() functions were used to return principal components.

The expected abundance curves for the three hazard species are shown in Fig 2 . Fig 2a ) clearly reflects the migratory behaviour of the Cattle Egret. The solar day was zeroed at 1-May meaning values <0 represent the period from November to May. This is the period during which Cattle Egrets begin to arrive in Brisbane (November) and leave Brisbane (late April), with the population peaking in February. This is also the period where the most strikes (orange markers) involving Cattle Egrets were observed. Fig 2b shows the abundance values against solar day (seasonality) for the Nankeen Kestrel. In this case, the solar day (seasonality) value was zeroed at 1-Nov meaning values <0 represent the period from May to November (cooler months of the year with shorter day lengths). Fig 2c shows the abundance values against solar day (seasonality) for the Straw-necked Ibis.

An external file that holds a picture, illustration, etc.
Object name is pone.0277794.g002.jpg

Blue markers indicate abundance on a particular day, orange markers shows days where at least one individual of the relevant hazard species was involved in a strike occurrence, and grey markers give the expected abundance on any day. (a) Cattle Egret abundance, seasonality, and expected abundance, (b) Nankeen Kestrel abundance, seasonality and expected abundance, (c) Straw-necked Ibis abundance, seasonality, and expected abundance.

Table 3 shows the results of logistic modelling used to determine the most indicative attribute, and associated threshold value to be used in generating collision likelihood for each hazard species.

Note that the general shape of the expected abundance curve for the Nankeen Kestrel is similar to that of the Cattle Egret, i.e., shows distinct seasonality. Unlike Cattle Egrets, Nankeen Kestrels are involved in collisions throughout the year. We found we got better predictive accuracy by splitting the observational data into two distinct sets at the change in seasonality. Using these attributes and thresholds, the following collision likelihoods may be derived (see Eqs 7 – 10 ).

The kernel mix parameters for each of the hazard species abundance models are given in S3 Table .

For the Nankeen Kestrel, Straw-necked Ibis and Cattle Egret hazard species, the accuracy of the predicted likelihood is 0.9, 0.96 and 0.85 respectively. The ROC curves for the states of Likelihood (High) are shown in Fig 3 .

An external file that holds a picture, illustration, etc.
Object name is pone.0277794.g003.jpg

(a) Nankeen Kestrel data set, (b) Straw-necked Ibis data set, (c) Cattle Egret data set.

Figs ​ Figs4 4 – 6 respectively show the BNM’s after training for Cattle Egret, Nankeen Kestrel, and Straw-necked Ibis respectively. Each figure highlights the ‘influence strength’ of connected nodes (shown as arrow width and direction), and includes the conditional probabilities and variables associated with high and low likelihood of collision involving the respective hazard species. For instance, in Fig 4 (i) Season strongly influences Seasonal Behaviour (the Cattle Egret is a migratory species and breeds in summer), and (ii) the likelihood of collision is high when the number of Cattle Egrets Observed is high, with highest collision likelihood when the Population per Zone is High, the number of Cattle Egrets Observed is High, the overall Population of hazard species is Medium, and the number of Aircraft Movements is High.

An external file that holds a picture, illustration, etc.
Object name is pone.0277794.g004.jpg

Node influence strength is given by the width of connecting arcs.

An external file that holds a picture, illustration, etc.
Object name is pone.0277794.g006.jpg

Conditional probabilities for all nodes, from each of the trained networks are available from our github repository ( https://github.com/robertandrews59/ABCBirdstrike ).

K-means clustering

Clustering results are visualised in the 2-dimensional plots in Fig 7 with cluster validity measures shown in Table 4 .

An external file that holds a picture, illustration, etc.
Object name is pone.0277794.g007.jpg

The axes for these plots are the principal components of the multi-dimensional vectors used as input to the K-means algorithm. (a) Cluster results for Cattle Egret where k = 2. Cluster 1 includes 187 days, cluster 2 includes 574 days. (b) Cluster results for Straw-necked Ibis where k = 2. Cluster 1 includes 191 days, cluster 2 includes 570 days. (c) Cluster results for Nankeen Kestrel where k = 2. Cluster 1 includes 130 days, cluster 2 includes 631 days.

Internal validation was by way of silhouette plots. Fig 8 is a silhouette plot for the hazard species Cattle Egret. The 761 observations form two (2) clusters with the average silhouette coefficient 0.33. (The higher the average silhouette coefficient, the more clearly clustering has segmented the data.) Cluster 2 incorporates the majority of the data points (574 out of 761 data points) with 85 points in cluster 1 having a negative silhouette width (avg. -0.11).

An external file that holds a picture, illustration, etc.
Object name is pone.0277794.g008.jpg

Figs ​ Figs9 9 and ​ and10 10 show silhouette plots for the Straw-necked Ibis and Nankeen Kestrel hazard species, again with each plot including the 761 observations of the respective hazard species’ data set. For the Straw-necked Ibis, cluster 2 incorporates the majority of the data (570 out of 761 data points) with 83 of the 191 points in cluster 1 having a having a negative silhouette width. For the Nankeen Kestrel, cluster 2 incorporates the majority of the data (631 out of 761 data points) with 57 of the 130 points in cluster 1 having a negative silhouette width.

An external file that holds a picture, illustration, etc.
Object name is pone.0277794.g009.jpg

Analysis of the distribution of bird strikes by species across clusters is shown in Table 5 . Analysis is hampered by the low number of strikes observed over the two year period of the study. However, K-means clustering shows that:

  • for Cattle Egrets, essentially all strikes occur in day type represented by cluster 1;
  • for Straw-necked Ibis, strikes are recorded in both day type clusters; and
  • for Nankeen Kestrels, strikes are recorded in both day types, with the majority of strikes occurring in day type represented by cluster 2.

Examination of strike data in S1 Table shows:

  • all but one of the strikes involving Cattle Egrets occurred in summer months (Nov-Feb), with a single strike in a winter month;
  • of the three strikes recorded for Straw-necked Ibis, two were in summer months and one was in a winter month;
  • of the 20 strikes recorded for Nankeen Kestrels, 13 occurred in the months between April and October (cooler months) while 7 occurred in the period December to March (warmer months).

Table 5 also shows, for each of the hazard species, the overall probability of a strike on any given day, the probability of a strike given the day falls into a given cluster, and 95% CI for each of the probabilities. Significance is determined by overlap of the confidence intervals. This analysis indicates that for the hazard species considered, strikes are not associated with day ‘types’ (as represented by the clusters) in that, apart form one instance, the confidence intervals are not distinct, i.e., they overlap.

One of the uses of clustering was to identify variables that contribute to likelihood. Fig 11a–11c respectively show the PCA analysis for Cattle Egret, Straw-necked Ibis, and Nankeen Kestrel respectively highlighting the contributions of individual variables to PC1 and PC2 BNM for each of the three hazard species.

An external file that holds a picture, illustration, etc.
Object name is pone.0277794.g011.jpg

For any variable, its contribution to overall variance is given by the distance from the origin (vector length) and is colour-coded (blue indicates a lower contribution, and red the highest). (a) Plot of contribution of variables to PC1 and PC2 for Cattle Egret. (b) Plot of contribution of variables to PC1 and PC2 for Straw-necked Ibis. (c) Plot of contribution of variables to PC1 and PC2 for Nankeen Kestrel.

S2 Table lists the two principal components responsible for the largest variation for each of the hazard species. Here, the attributes responsible for the most variation across the three hazard species include attributes related to overall population on the airfield (counted and harassed), population in the zones closest to the runways (counted and harassed), a seasonality indicator (solar day).

Our intent in conducting this work was to develop some techniques to support wildlife hazard risk assessment at airports, that went beyond only hazard ranking, to provide a more dynamic, i.e. sensitive to changing environmental factors, method of providing a collision likelihood. As previously noted, risk assessment also includes taking into account a severity measure. This aspect was outside the scope of this paper, and we concerned ourselves with only collision likelihood and the factors able to be discerned from routinely collected observational data that contribute most to collision likelihood.

A strength of the Algebraic approach is that it clearly visualised the seasonal nature of the abundance of the hazard species included in this study. In particular, the migratory behaviour of the Cattle Egret is evident as is the hunting behaviour of Nankeen Kestrels across the year. For the Cattle Egret, peak abundance corresponds with the period where the most strikes involving Cattle Egrets were observed. For the Nankeen Kestrel we hypothesise that the seasonal abundance is related to the bird’s hunting behaviour and the fixed time of the day at which the daily count is conducted, rather than actual changes in the number of birds on the airfield. In summer, the ground over the entire airfield would be warm enough, from late-morning, to generate rising air columns and hence support hover-hunting over most of the airport, particularly over concrete runways and taxi-ways, i.e., spaces where aircraft are departing and arriving. However, in winter months, when the air is cooler, Nankeen Kestrels revert to perch-hunting in the morning (corresponding to the time when the daily count is being conducted) with hover hunting occurring later in the day when the concrete runways and taxi-ways would be warmer than other parts of the airfield, and would thus generate rising air towards the middle of the day. As the daily count is conducted at the same time of day, such behaviour would affect the number of Nankeen Kestrels observed observed during the daily count.

Expert opinion from the BAC WHM indicated that the population of Nankeen Kestrels on the airfield remained constant throughout the year, yet the abundance figures (derived from the daily count) show fewer birds counted during the warmer period of the year, with more strikes occurring in this period. We ascribe this to the timing of the daily count and the different hunting behaviours exhibited by Nankeen Kestrels at different times of the year. That is, the birds were on/around the airport, but weren’t actively hunting or flying at the time the daily count was conducted, i.e., they were perching out of sight and hence were not observed during the daily count. A point worth noting is that, for 7 of the 20 strike events involving Nankeen Kestrels recorded during the study period, there were no Nankeen Kestrels recorded in the daily count or included in harassment activities conducted by the Wildlife Hazard Management team. Although Straw-necked Ibis are among the most abundant bird species on the airport precincts, they are involved in relatively few strikes (3 strike occurrences over the two year period of our study). The BAC WHM report that the species exhibits flocking behaviour, and that it is relatively easy to move the birds to parts of the airport away from runways and taxi-ways. So, although there are large numbers (often several hundreds) of birds on the airport on any given day, they are rarely in close proximity to areas of aircraft movement.

The expected abundance curve however, gives an indication of the expected number of the relevant hazard species at various times of the year and may act as supplement to the daily count of actual abundance. When overlaid with strike information, it provides an easy way to derive rules that quantify collision likelihood. A significant contribution of this approach is the notion of ‘seasonality’ instantiated in this study as ‘solar day’. Seasonality is useful in concentrating routine, daily observations into the seasonality window to make patterns of behaviour more apparent. In this study, a calendar year was used as the seasonality window. This clearly highlighted migratory patterns of the Cattle Egret, and to some extent, the changing hunting behaviour of Nankeen Kestrels at different times of the year. We note that the notion of seasonality may be applied over other time windows. For instance, seasonality could be applied to a daily 24 hour period to highlight diurnal/nocturnal behaviour patterns. Bayesian Network modelling makes explicit the causal relationships between variables associated with the likelihood of bird strike. Bayesian Networks allow, given, a set of environmental conditions, to reason forward to arrive at a collision likelihood. An additional feature of the Bayesian Network approach is the ability to reason backward from a specified condition (such as high likelihood of collision) to examine the sets of pre-conditions (causal factors) that give rise to the elevated collision likelihood. As such, Bayesian Networks could support decision-making in response to heightened immediate risk, or guide longer-term risk reduction strategies, by identifying currently significant risk factors.

Bayesian Networks, in contrast to the simplistic Risk = Severity × Likelihood , showcase the integration of all the factors that impact on both elements of risk function (i.e. Severity and likelihood). Bayesian Networks are also able to quantify uncertainty and allow the integration of known probability values associated with any nodes within the networks (prior to an adaptive learning phase). Further, Bayesian Networks provide a visual mechanism to record and test subjective probabilities, an important role in circumstances where there is not much data.

The K-means clustering approach was used to determine whether it was possible to develop a set of ‘day types’ and then determine if particular ‘day types’ were associated with strikes. For each of the hazard species considered in this study, (i) clustering generated a set of prototypical ‘day types’, (ii) strike occurrences (days on which a strike occurred) were localised within particular clusters (‘day types’). Principal Component Analysis revealed that numbers of hazard species (counted), their proximity to runways, and seasonality contributed strongly to clustering of days. We note that these attributes were shown, through logistic regression modelling in the Algebraic approach, to be predictive of elevated collision likelihood. We also note that the seasonality value perhaps subsumes other environmental indicators such as rainfall and temperature as, in Brisbane, these values have distinct ranges associated with the time of year. However, analysis of collision likelihoods for most clusters showed that there was little support for notion of day ‘types’ determined by clustering being indicative of strikes. We feel that more extensive investigation is required before completely discounting this notion.

As for practical application of the ABC approach to collision likelihood, we feel that the techniques are complementary and should be applied in parallel, and if any one of the techniques predicts a heightened likelihood of collision, actions should be taken. As discussed earlier, the techniques give indications of the factors leading to heightened collision likelihood. These could be used to inform mitigation actions.

The methods described in this paper utilise species-at-a-time modelling. We do not see this as a limitation as it is naive to assume that collisions (strikes) and methods of mitigating strikes are generally applicable to all hazard species and all airports. Our approach provides understanding at the species level, thus allowing for tailored collision likelihood modelling, and potentially tailored mitigation actions. Future work could, however, include generalising the approach to model ‘species types’. That is, hazard species could be characterised by a set of features, and risk models constructed for each hazard species type. This would have the advantage of better utilising observational data and strike data as data relating to each species belonging to a particular type would be aggregated to model the type.

The techniques described in this paper can be characterised as having local sphere of application, intermediate forecast horizon, and, at this stage, algorithmic/mode realisation (although the effort required to systematise the individual ABC algorithms would not be great). As such, the ABC approach fills an obvious gap and form a middle ground between purely static risk assessment methods such those described in [ 11 , 18 ], and immediate indicators such as avian radar making them suitable for application in a wide variety of airports around the world. Neither static, hazard species risk ranking approaches, nor immediate methods, take into account seasonal or environmental factors, hence they cannot be used to (i) reason about likely future states on the airfield, or (ii) to plan mitigation measures in anticipation of some likely future state (a further requirement of airport operators according to CASA and described in [ 12 ]). A very obvious example is that of migratory behaviour (as captured in our expected abundance model) or the temporal offset between a rain event and abundance of a given hazard species. That is a) rain, leads to b) grass growing, which provides c) increased food supply for insects and cover for small, ground dwelling animals, which act as d) attractants for insectivorous birds and birds which hunt ground dwelling animals. Such causal chains are nicely captured by the Bayesian Network approach, and can be used at the airport to plan mitigation activities such as mowing, draining of standing water, and removal of perching opportunities for birds across the airport.

In this paper we have described three techniques (the ABC approach—Algebraic, Bayesian, Clustering) useful in understanding causal factors, and in assessing the likelihood of bird strike at an airport. The techniques were evaluated with routinely collected environment and hazard species data at Brisbane airport.

Contributions of this work to the body of work dealing with aircraft collisions with (avian) hazard species at airports (i) include the definition of attributes for ‘seasonality’ and ‘proximity count’ (to deal with risk associated with hazard species at varying distances from aircraft movement corridors), and (ii) the development of three novel approaches for assessing likelihood of collision.

Each of the three methods improve on simple ‘ranking’ of hazard species, the minimum required risk assessment for airport accreditation, by providing dynamic risk assessment based on the state of environmental variables. Further, the variables used by the models are those routinely collected by airport Wildlife Hazard Management teams thus implementing the techniques at other airports should not add any burden to data collection, or require changes in data collection practices. Areas for future research around data include (i) how frequently should the historical environmental and collision data be updated in order to keep the models current, i.e., reflective of the actual collision likelihood at the airport, and (ii) what ‘window’ of historical data is optimal for modelling.

Of particular importance is the identification of variables that contribute strongly to collision likelihood. The Bayesian Network Models are well suited for ‘what if’ analysis. That is, these models allow for a set of environmental conditions to be provided to the model with the associated risk profile being generated, thus allowing for both diagnostic and prognostic use of the models.

An avenue for future work is to discover means to group ‘similar’ hazard species and apply the ABC approach to the hazard species group thus making the approach more general. Lastly, the ABC approach fits between purely static risk assessment approaches and purely immediate (e.g. avian radar) in terms of responsiveness to changing conditions on the airfield. The ABC approach also supports planning of mitigation activities through the models providing a ‘look ahead’ view allowing wildlife hazard management teams to anticipate future risk states. Lastly, this work was developmental. As such, software used was chosen that directly supported the models and provided appropriate analysis and visualisation capabilities. Future development may include development of an integrated platform making all the techniques available in the one tool. Diffusion theory [ 45 ] is concerned with the uptake and spread of innovations. It tells us, among others, that technology adoption is driven by not only technological factors, but adoption pathways, time, and the social system around the innovation. In this context, Realisation of the ABC approach as a tool for use in practice has both technological and awareness/acceptance considerations. The technological side will require, among other things, the implementation of each technique in an integrated platform, thus making all the techniques available in the one tool. Such a tool would also require options to allow wildlife control teams to configure routinely collected data for presentation to the models, and to define hazard species relevant to the airport at which the approach was to be deployed. Awareness will require disseminating information about the approach to the industry through for instance wildlife hazard working groups. Acceptance will require a quantum of early adopter airports and their subsequent communication of positive feedback to other airports, leading to more widespread adoption and ultimately approval or even enforcement by from regulatory bodies.

Supporting information

Dates of strike occurrences are shown together with the number of individuals involved in the occurrence.

Acknowledgments

We gratefully acknowledge Brisbane Airport Corporation for making members of its Wildlife Hazard Management team available to the research team.

Funding Statement

The grant funding was used for (partial) salaries of the following authors (as QUT employees): BB, BC, RA. The funder provided support in the form of salaries for author JR (as an employee of BAC). The specific roles of these authors are articulated in the ‘author contributions’ section. The funder did not have any role in the analysis, or preparation of manuscript. The funder was involved in validating the study design, and made historical data collected by its Wildlife Hazard Management team available to the researchers. The funder was also involved in the decision to publish.

Data Availability

  • PLoS One. 2022; 17(12): e0277794.

Decision Letter 0

21 Apr 2022

PONE-D-21-25171The ABC of Bird Strikes at BAC: Modelling bird strike likelihood at Brisbane AirportPLOS ONE

Dear Dr. Andrews,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

Please submit your revised manuscript by Jun 05 2022 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at  gro.solp@enosolp . When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

  • A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.
  • A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.
  • An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols . Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols .

We look forward to receiving your revised manuscript.

Kind regards,

Juan Manuel Pérez-García, PhD

Academic Editor

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at 

https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and 

https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2. Thank you for stating the following in the Financial Disclosure section: "Authors receiving funding: MW

Grant: Direct Funding

Funder: Brisbane Airport Corporation

Website: https://www.bne.com.au/corporate

Funder reviewed the manuscript and agreed to publish"

We note that one or more of the authors have an affiliation to the commercial funders of this research study : Brisbane Airport Corporation

a. Please provide an amended Funding Statement declaring this commercial affiliation, as well as a statement regarding the Role of Funders in your study. If the funding organization did not play a role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript and only provided financial support in the form of authors' salaries and/or research materials, please review your statements relating to the author contributions, and ensure you have specifically and accurately indicated the role(s) that these authors had in your study. You can update author roles in the Author Contributions section of the online submission form.

Please also include the following statement within your amended Funding Statement. 

“The funder provided support in the form of salaries for authors [insert relevant initials], but did not have any additional role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. The specific roles of these authors are articulated in the ‘author contributions’ section.”

If your commercial affiliation did play a role in your study, please state and explain this role within your updated Funding Statement. 

b. Please also provide an updated Competing Interests Statement declaring this commercial affiliation along with any other relevant declarations relating to employment, consultancy, patents, products in development, or marketed products, etc.  

Within your Competing Interests Statement, please confirm that this commercial affiliation does not alter your adherence to all PLOS ONE policies on sharing data and materials by including the following statement: "This does not alter our adherence to  PLOS ONE policies on sharing data and materials.” (as detailed online in our guide for authors http://journals.plos.org/plosone/s/competing-interests ). If this adherence statement is not accurate and  there are restrictions on sharing of data and/or materials, please state these. Please note that we cannot proceed with consideration of your article until this information has been declared.

Please include both an updated Funding Statement and Competing Interests Statement in your cover letter. We will change the online submission form on your behalf.

3. Please upload a new copy of Figure 7 as the detail is not clear. Please follow the link for more information: https://blogs.plos.org/plos/2019/06/looking-good-tips-for-creating-your-plos-figures-graphics/ " https://blogs.plos.org/plos/2019/06/looking-good-tips-for-creating-your-plos-figures-graphics/

4. Please include captions for your Supporting Information files at the end of your manuscript, and update any in-text citations to match accordingly. Please see our Supporting Information guidelines for more information: http://journals.plos.org/plosone/s/supporting-information. 

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Partly

Reviewer #2: Partly

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: No

Reviewer #2: I Don't Know

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: No

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: PONE-D-21-25171

The MS presents an original approach of determining the dynamic likelihood of bird collisions with aircrafts that has potential application in reducing such incidents to save both human and avian lives. The concept of the MS is interesting, however, its presentation needs major improvement, particularly in terms of explaining/clarifying how the analytical methods were used in this specific problem and how the results were interpreted. Currently, there is large ambiguity and lack of flow in the methods and results that limit the reproduction of this approach. While the ABC approach of dynamic collision risk modeling is elegant, the statistical treatment around it is not robust in the current MS, and the interpretations of whether some ‘days’ have higher collision risk are subjective and not based on rigorous statistical examination. My detailed comments are attached below:

1. The Introduction nowhere acknowledges that bird strike can be a problem for bird populations as well, given the scale of incidents and mortality

2. L 51-70 contains details that are very specific to the case and not relevant for general readers. This section can be substantially revised and reduced keeping in mind the general readership.

3. L 159-184 details of three hazard bird species can be reduced by retaining only the most relevant information.

4. L 164-170 Authors postulate hypotheses on factors influencing the space use by Nankeen kestrel, but do not return to this point explicitly in the results and discussion section. Results and inferences pertaining to this hypothesis can be added in Discussion as an explanation for seasonal changes in the species’ space use/numbers.

5. Methods overlap with Results and need substantial revision to improve the flow and clarity. The statistical analysis around the ABC approach can be improved by including objective examination of whether some ‘days’ have higher collision risk (see suggestion below in point 7). Theoretical concepts behind the three methods – Algebraic modelling, Bayesian networks and K-means cluster analysis are explained well, but are sometimes superfluous as these techniques are widely used, and these sections can be reduced by referring to relevant literature wherever applicable, to keep the text tight, to the point, and focused on the application of the technique. More details on how these techniques have been applied such as variables and analytical treatments need to be mentioned explicitly, so that the approach can be clearly understood and replicated. Some analysis that these techniques depend on such as PCA, ROC, and validation techniques for Cluster analysis are not mentioned in Methods and appear abruptly in Results, making it difficult to follow how the methods were applied.

6. L 234 – 320 Algebraic modelling approach has been explained in sufficient detail and clarity to allow replication. However, the content is verbose and can be revised for brevity. For instance, repetition of the mathematical notation in L274 – 276 is redundant, and L283-301 explanations of Gaussian and Sigmoid functions can be reduced.

7. L 304 – 309 The choice of expected abundance thresholds in classifying collision likelihoods as ‘low’, ‘medium’, ‘high’ is subjective and arbitrary. No statistical treatment has been presented that relates expected abundance to air strike probability. An objective way would be to predict collision probability using binomial GLM (logistic regression) by modeling air strike (1/0) on daily expected abundance of hazard species (eh), and to subsequently classify abundance thresholds based on stipulated cut-offs of predicted collision probability.

8. L 310 Bayesian networks have been used to examine causal relationships between factors leading to collisions. Which factors / variables have been examined in this study are not mentioned in Methods.

9. L 316 In Bayesian networks, a node (factor) has a conditional probability distribution table for each parent node (causal factors) that explains the causal relationships. However, these tables are missing from the Results. Only the networks have been shown in Results which tells readers that X leads to Y but not how X influences Y that is vital for this MS – to understand how seasonal environmental changes can lead to conditions promoting collisions. I return to this point in the Results.

10. L 317-333 Content is verbose and can be revised for brevity

11. In (K-means) Clustering, L 338 Although the objective was ‘to see whether particular day types are associated with bird strikes’, the following section on cluster analysis does not explain how this objective is answered from the knowledge of ‘similar days/conditions’ groups.

12. L 339-350 & 370-378 Given that cluster analysis is a very widely used technique, parts of the section can be reduced by citing relevant literature.

13. L 351-359 Data standardization is a widely used technique and can be explained in a single sentence with relevant citation.

Much of the Results are actually details of methods or interpretation of results that should be shifted to Methods and Discussion sections, respectively. Results should provide adequate reporting of the output statistics of each analysis (see detailed suggestions below).

14. Figures lack legends, axis labels and have been presented without diligence that make them difficult to fully comprehend. There are too many figures, and figures of similar type (Fig 1-3, Fig 4-6, Fig 8-10, Fig 11-12, Fig 13-15) can be grouped with species names as labels / icons. Other recommendations specific to each figure are as follows:

15. Fig 1-3: Shift y-axis to the left and remove chart titles.

16. Fig 4-6: These figures show what factors influence collision likelihood. However, the tables/graphs accompanying child nodes seem to be unconditional data distributions that do not allow readers to interpret the nature of the relationships between nodes. Authors can think of presenting the conditional probability tables of child nodes as contingency tables on edges/arrows, or some other concise, interpretable ways, so that readers (including managers) can understand how some environmental changes can promote factors that ultimately lead to higher collision risk.

17. L 381-389 should be removed and included in the figure legend

18. L 390-392 & L 405 & L 415 Collision likelihood based on Algebraic Modeling need to be revised based on the suggestion in point 7.

19. L 397-403 should be included in Discussion and removed from here.

20. L 420-430 details on Bayesian network analysis should be included in Methods and removed from here.

21. L 438 – 448 description of ROC should be included in Methods and removed from here.

22. L 450 – 462 details of Cluster analysis should be shifted from here to Methods

23. L 457 is confusing, as it mentions that PCA can be used as a variable for Algebraic Modeling; however, this variable was not used in Algebraic Modeling and was not referred there. There are several such ambiguous, half-explained statements throughout the MS that makes the methods difficult to follow and replicate.

24. L 463 – 475 details of Cluster analysis validation should be shifted from here to Methods.

25. L 496 Table 4 showing the number of strikes as a function of cluster type is non-informative as we do not know how many days are there in each cluster, to be able to infer if the frequency of strikes was more in one cluster than other. Also cluster difference in collision probability has not been tested statistically, hence the inferences are not reliable. Authors should revise this section (and the corresponding Methods) by including a statistical test of difference in collision probability between day clusters, similar to my suggestion in point 7.

26. L 497 – 511 details of Principle Component Analysis should be shifted from here to Methods. PCA methods should clearly list the variables included and analytical details (data standardization, correlation vs covariance matrix use etc.) in Methods. PCA results should include the variance explained and variable loadings of components in Results.

27. Cluster analysis results should include cluster centroid descriptions in terms of variable means for each hazard species and number of days in each cluster in Results so that readers (including managers) can understand how days are grouped based on conditions and which condition set increases the risk of collision.

Discussion & conclusion

Discussion is weak and largely reiterates the advantages of the approach developed in the MS. It can be strengthened by: a) discussing the findings of applying this approach to the current study in terms of which dynamic factors increased collision risk of a hazard species, and interpreting these results from ecological perspective (see comments in Results section); b) referring to literature on how bird strikes are mitigated across the world beyond simple categorization of hazard species by their risk/severity; and c) how the approach developed in the MS can help advance these current approaches of bird strike mitigation, thereby expanding the scope of the work.

L 556 ‘PCA revealed that number of hazard species … were strong indicators of strikes’ : It is not clear how authors jump to this inference, as Methods do not clarify how daily collision frequency was related to Principle Component(s), and Results do not report any such statistic. The PCA figures 13-15 only show loadings and relative contributions of variables on components, not which component influences collision risk and how. Again refer to my suggestion in point 7 and revise the analytical approach in the MS accordingly.

Conclusion reiterates the approaches used, much of which have already been covered. Instead, it can be reduced to a few important concluding statements on the application of these techniques in reducing bird strike problems.

Reviewer #2: The paper presents three ways to inform collision risk from wild birds. This is an important study that has real world applications. The authors make a case for their study by referencing other risk assessment frameworks and approaches. However being a study that describes novel methods for risk assessment, a solid, readable and fluid methods section is indispensable to the manuscript, which is currently lacking.

There is scope for clarifying description of methods. In particular, the descriptions are presently difficult to follow as the important terms used in the calculations are not ordered. One has to refer to previous pages to understand. The conceptual parts of the 3 methods could be included as they apply to the problem being presented, or skipped altogether. Methods and results sections are intermixed in many ways.

The introduction and discussion are relatively well drafted, however the same clarity does not exist in the methods and results sections.

An accuracy test of the three methods, using predicted collision risk and actual data of collision could be a valuable addition.

Authors may check for the correct use of wildlife ecology terminology viz., population, species, abundance etc. ‘Populations’ are mostly estimates, and not absolute numbers. Moreover non-detection of birds has not been included in the methods, while it has been acknowledged in results (lines 402-403).

1. Not clear if how variation in bird numbers in Paton’s approach is taken into account, and how this approach does not account for seasonal/environmental changes. Authors may elaborate on this.

3. Study hazard species are those that have a history of collision with aircrafts. However the authors have stated in the introduction that it is important to include species that may not have a history but may still be potential hazards

4. For context, it may be useful to define the main food source (for reference in line 225), habitat requirements/nesting ecology of the hazard species.

5. Para beginning at line 237: for clarity, I suggest the authors provide some context of the references provided here (eg. Carter in his study on risk assessment and prioritisation of wildlife hazards, the author used ….).

6. Line 243 – 245: does this clustering refer to the k-means clustering? The meaning is unclear. Also a reference is probably missing at the end of the sentence.

7. Line 347: number of hazard species should be written as ‘number of individuals belonging to hazard species’.

8. Lines 267 and beyond: the term ‘population’ doesn’t seem appropriate here. Authors could use ‘count per zone’ or ‘individuals per zone’. Note that ‘sum of hazard species .. ‘ and number of individuals of the species (I think the authors mean this) are two separate things.

9. Lines 266 – 271: the language of this explanation is not clear.

10. Equation after line 274: q has not been defined. Do the authors mean p?

11. Line 353-359: standardisation is a standard practice, authors need not explain this.

12. What was done in references 23 and 24 need to be defined here. What is the average 0.1 value, and how is the 0.5 value set as threshold?

13. Lines 364 – 366: describe how.

14. Lines 367 – 369: use this to substantiate on what analyses were done using these packages in r, and which functions were used for which purpose (instead of not defining how average values and recommended number of clusters were derived – ref to comments above).

15. Lines 420 – 429: methodological details which are incorrectly placed in the results section.

6. PLOS authors have the option to publish the peer review history of their article ( what does this mean? ). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy .

Reviewer #1:  Yes:  SUTIRTHA DUTTA

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool,  https://pacev2.apexcovantage.com/ . PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at  gro.solp@serugif . Please note that Supporting Information files do not need this step.

Author response to Decision Letter 0

10 Jul 2022

We have provided detailed responses to each comment and suggestion for improvement provided by all of the editor and two reviewers in the Response to Reviewers letter submitted as a file in this upload.

Submitted filename: 20220708 Response to Reviewers Letter.pdf

Decision Letter 1

PONE-D-21-25171R1The ABC of Bird Strikes at BAC: Modelling bird strike likelihood at Brisbane AirportPLOS ONE

Please submit your revised manuscript by Oct 22 2022 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at  gro.solp@enosolp . When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice.

Additional Editor Comments:

I suggest that you take into consideration the recommendations of both reviewers. In particular reviewer 3 brings two very important issues. The first is the differentiation between "likelihood" and "risk", you should specify what you performed (in my opinion you did likelihood, but not risk). If so, correct in the title and in some parts of the text. Another option is to detail in methods that likelihood was used as a proxy for risk, but it should be adequately justified with references and explaining how you accounting for severity.  

Another critical point indicated by Reviewer 3 is that at no point is it evaluated which of the 3 techniques is better or if it is the combination of them. I consider that including some of this could greatly improve the conclusion of the paper. 

Moreover, I include two additional recommendations. The title is uninformative and could even be misleading due to the presence of so many abbreviations. I understand that a play on words may have been created, but for non-English speaking audiences it is not clear. I suggest avoiding the inclusion of abbreviations (e.g. ABC or BAC) or local names without their country. A suggestion may be “Evaluation of risk assessment techniques to model bird strikes: the case of Brisbane Airport in Australia”. Authors should include the meaning of the abbreviation ABC in the abstract (e.g. in line 39). 

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #2: All comments have been addressed

Reviewer #3: (No Response)

2. Is the manuscript technically sound, and do the data support the conclusions?

Reviewer #2: Yes

Reviewer #3: Partly

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #3: Yes

4. Have the authors made all data underlying the findings in their manuscript fully available?

5. Is the manuscript presented in an intelligible fashion and written in standard English?

6. Review Comments to the Author

Reviewer #2: The manuscript is considerably improved, and the authors have addressed all previous comments/suggestions. I recommend few additional suggestions for enhancing the readability and understanding of the general idea of the paper. I would also like to commend the authors on carrying out this important study.

General comments:

- Authors could add a mention of how frequently would the models be updated with emerging information (recent collision information, local weather changes, novel hazard species)?

- Curious why all three methods were carried out in different software, and not one. Authors could add a line justifying this.

- Figure and table titles require editing (typos) and work (addition of information to make it comprehensive by itself, and not dependent on text).

Specific comments:

Line 27: and rare bird extinction

Line 29: and bird populations.

Introduction:

Line 161: raptors (“preying on?” small mammals and insects)

Line 177-179: is this a repetition of line 150-153?

Line 232: “includes ten likelihood includes ten indicators including” – not clear

Table 2, page 13: Type column for Zone’s population and All population: continuous (instead of continues)

Fig 1, page 14: typos in title (Straw-necked ibis – at two places)

Line 336: demonstrates

Line 390: Connect explanation of silhouette coefficient and Dunn index with subsequent lines (392-395), in terms of which index/coefficient is a measure of what. Moreover these terms do not appear in the results.

Table 3: Title above table, title needs to be more explanatory (instead of explanation in text)

Line 422: not clear why the split was necessary

Line 429-432: two sentences with repetitive meaning

Line 434: an explanation of one the figures would add value to this section.

k-means clustering; lines 443-453: repetition of methods. Authors could add a few lines “…we found that….”

Lines 454-456: text better suited/more appropriate for figure title

Table 4: title needs revision; cluster tendency is a new term not been mentioned before (in methods)

Figures 8 and 9: reason for placing figures for species as such (1:2)? Figure titles need to explain coefficients etc mentioned in the plot.

Line 460: is average silhouette score the same as silhouette coefficient (mentioned in line 390)? Authors should use the same term throughout the manuscript.

Line 568: they are rarely in close to proximity

Line 572: when overlaid with strike information, <it abundance="" expected=""> provides

Line 600: <categorise group=""> days into clusters (?)</categorise></it>

Reviewer #3: Risk assessment procedures, as crucial as they are, do have their limitations in practice as identified by the authors. Hence, I appreciate their approach of investigating new methods to improve the data availability for wildlife controllers in the field.

The authors clearly state in the text that they focus on likelihood, excluding the severity aspect. This is perfectly fine, however, by doing so, they do not create a risk but a likelihood model. This is indicated in the title, but throughout the text, there is still some mentioning of their models as “risk assessment models”. Please correct.

It is interesting to see the three presented approaches as well as their results. What I am missing in the overall discussion is, how they can best be used in practice Lines 43-44 in the abstract state that each of the techniques meet the requirements – so would one be enough? Which one best? Or would a combination be wise after all?

In the conclusion you mention that the ABC approach is useful etc. But what exactly is the approach? Looking at each of the individual models? Combining their outcome? I am missing the big picture here, please enlighten me and the other readers in the discussion or the conclusions section.

In addition, since the authors seem to strive to actually support risk assessment procedures in practice, I am missing an outlook on where they will take the models from here to actually get them to the airports and make them useable for the wildlife control teams.

Comments on the text

• Source 9 (US Wildlife strikes 1990-2015) is outdated. The current version is 1990-2021, please update

• There is a very hard content-related break between the intro and the methods section (l 112-118). Help the reader by ending the introduction /starting the introduction with a brief reiteration of what you are going to present next. E.g. “to develop these models, a case study was performed at Brisbane airport”

• L 138: the text indicates that you selected three out of multiple hazardous species. Is this the case or are these THE three (most) hazardous species? In case it was three out of multiple, please state why you selected those. If it was the three, please clarify

• L 182-183: superfluous, starting the methods with what is written in l 184 is perfectly sufficient

• Table 1: What about limitations? Are these hidden in the requirements? If so, I would relabel the “requirements” descriptor. If not, limitations should be added for the sake of completeness

• L 279: I am missing a source here as well as equation numbers on this page

• Gaußian and Sigmoid functions come out of the blue. Please briefly state what they are and why they are relevant after l 279

• L 315-317: This is a strong statement. Either you have to provide a source or some reasoning why you believe this to back it.

• Equation before Table 2 – how can a likelihood be a function of a risk level? This does not make sense at first sight. Please elaborate and give an explanation what the risk level includes and how it is measured.

• Figures in general (except F8): Text of labels / in bubbles is extremely small and thus almost illegible. Please increase the font sizes. You might want to change color-coding to different shades of grey for readers who print the paper in grey scale (such as myself)

• L 442: please make this a full sentence

• 514-524: Would this information not be more suitable in the intro? And please add some sources for the individual systems/models. Even though in the business for quite a while, I have never heard of the Swiss/Dutch Bird Migration Model. Is this really its name? Please add a reference.

• 532-540: This is interesting but in large parts a repetition of the results – please make that paragraph more concise

• I realize that you discuss the results of the models in different level of details. Why is this? I would recommend to revisit this section and investigate the options to bring more balance into it. In addition, I would expect some mapping to literature and/or observations from BAC or other airports. Do the models reflect reality well?

• L 558-559: the bird’s weren’t active when? On the day when the considered strike occurred? Please clarify

• L 572 it provides (‘it’ is missing)

7. PLOS authors have the option to publish the peer review history of their article ( what does this mean? ). If published, this will include your full peer review and any attached files.

Reviewer #2:  Yes:  Akanksha Saxena

Reviewer #3: No

Author response to Decision Letter 1

10 Oct 2022

We have included our detailed responses to reviewers' comments as an uploaded file.

Submitted filename: 20221022 Response to Reviewers Letter.docx

Decision Letter 2

PONE-D-21-25171R2

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/ , click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at gro.solp@gnillibrohtua .

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact gro.solp@sserpeno .

Additional Editor Comments (optional):

Acceptance letter

28 Nov 2022

Dear Dr. Andrews:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact gro.solp@sserpeno .

If we can help with anything else, please email us at gro.solp@enosolp .

Thank you for submitting your work to PLOS ONE and supporting open access.

PLOS ONE Editorial Office Staff

on behalf of

Dr. Juan Manuel Pérez-García

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here .

Loading metrics

Open Access

Peer-reviewed

Research Article

Three novel bird strike likelihood modelling techniques: The case of Brisbane Airport, Australia

Roles Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Writing – original draft

* E-mail: [email protected]

Affiliation School of Information Systems, Faculty of Science, Queensland University of Technology, Brisbane, Queensland, Australia

ORCID logo

Roles Data curation, Formal analysis, Investigation, Writing – original draft

Roles Data curation, Formal analysis, Validation, Writing – original draft

Roles Conceptualization, Funding acquisition, Project administration, Supervision, Writing – review & editing

Roles Conceptualization, Funding acquisition, Writing – review & editing

Roles Conceptualization, Writing – review & editing

Affiliation Wildlife Management Group, Brisbane Airport Corporation, Brisbane, Queensland, Australia

  • Robert Andrews, 
  • Bayan Bevrani, 
  • Brigitte Colin, 
  • Moe T. Wynn, 
  • Arthur H. M. ter Hofstede, 
  • Jackson Ring

PLOS

  • Published: December 8, 2022
  • https://doi.org/10.1371/journal.pone.0277794
  • Peer Review
  • Reader Comments

Table 1

The risk posed by wildlife to air transportation is of great concern worldwide. In Australia alone, 17,336 bird-strike incidents and 401 animal-strike incidents were reported to the Air Transport Safety Board (ATSB) in the period 2010-2019. Moreover, when collisions do occur, the impact can be catastrophic (loss of life, loss of aircraft) and involve significant cost to the affected airline and airport operator (estimated at globally US$1.2 billion per year). On the other side of the coin, civil aviation, and airport operations have significantly affected bird populations. There has been an increasing number of bird strikes, generally fatal to individual birds involved, reported worldwide (annual average of 12,219 reported strikes between 2008-2015 being nearly double the annual average of 6,702 strikes reported 2001-2007) (ICAO, 2018). Airport operations including construction of airport infrastructure, frequent take-offs and landings, airport noise and lights, and wildlife hazard management practices aimed at reducing risk of birdstrike, e.g., spraying to remove weeds and invertebrates, drainage, and even direct killing of individual hazard species, may result in habitat fragmentation, population decline, and rare bird extinction adjacent to airports (Kelly T, 2006; Zhao B, 2019; Steele WK, 2021). Nevertheless, there remains an imperative to continually improve wildlife hazard management methods and strategies so as to reduce the risk to aircraft and to bird populations. Current approved wildlife risk assessment techniques in Australia are limited to ranking of identified hazard species, i.e., are ‘static’ and, as such, do not provide a day-to-day risk/collision likelihood. The purpose of this study is to move towards a dynamic, evidence-based risk assessment model of wildlife hazards at airports. Ideally, such a model should be sufficiently sensitive and responsive to changing environmental conditions to be able to inform both short and longer term risk mitigation decisions. Challenges include the identification and quantification of contributory risk factors, and the selection and configuration of modelling technique(s) that meet the aforementioned requirements. In this article we focus on likelihood of bird strike and introduce three distinct, but complementary, assessment techniques, i.e., A lgebraic, B ayesian, and C lustering (ABC) for measuring the likelihood of bird strike in the face of constantly changing environmental conditions. The ABC techniques are evaluated using environment and wildlife observations routinely collected by the Brisbane Airport Corporation ( BAC ) wildlife hazard management team. Results indicate that each of the techniques meet the requirements of providing dynamic, realistic collision risks in the face of changing environmental conditions.

Citation: Andrews R, Bevrani B, Colin B, Wynn MT, ter Hofstede AHM, Ring J (2022) Three novel bird strike likelihood modelling techniques: The case of Brisbane Airport, Australia. PLoS ONE 17(12): e0277794. https://doi.org/10.1371/journal.pone.0277794

Editor: Juan Manuel Pérez-García, Universidad Miguel Hernandez de Elche, SPAIN

Received: August 10, 2021; Accepted: November 3, 2022; Published: December 8, 2022

Copyright: © 2022 Andrews et al. This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: All relevant data are within the paper and its Supporting information files. A GitHub repository has also been created to make larger files available to reviewers and readers https://github.com/robertandrews59/ABCBirdstrike ”.

Funding: The grant funding was used for (partial) salaries of the following authors (as QUT employees): BB, BC, RA. The funder provided support in the form of salaries for author JR (as an employee of BAC). The specific roles of these authors are articulated in the ‘author contributions’ section. The funder did not have any role in the analysis, or preparation of manuscript. The funder was involved in validating the study design, and made historical data collected by its Wildlife Hazard Management team available to the researchers. The funder was also involved in the decision to publish.

Competing interests: The authors have read the journal’s policy and have the following competing interest: JR was, at the time of writing, an employee of Brisbane Airport Corporation. This does not alter our adherence to PLOS ONE policies on sharing data and materials. There are no patents, products in development or marketed products associated with this research to declare.

Introduction

The risk posed by wildlife to air transportation is of great concern worldwide. Wildlife Hazard Management (WHM) is a mandatory procedure in every significant international and national airport [ 1 ]. A hazard is defined as “the condition or circumstance that could lead to damage or destruction of an aircraft, or to loss of life as a result of aircraft operations” [ 2 ]. Statistics and data on wildlife collisions with aircraft give some indication of the scale of the problem. In Australia alone, 17,336 bird-strikes occurrences and 401 animal-strike incidents were reported to the Australian Transport Safety Bureau between 2010 and 2019 [ 3 ] and 17,360 strikes reported to the US Federal Aviation Authority (FAA) in a single year (2019) [ 4 ]. Wildlife strikes can have catastrophic consequences. The FAA report more than 301 people killed and over 298 aircraft destroyed globally as a consequence of wildlife strikes since 1988 [ 5 ]. It is estimated that wildlife strikes incur cost to the civil aviation industry in the US of approximately $625 million per year [ 6 ] and globally US$1.2 billion per year [ 7 ]. On the other side of the coin, civil aviation, and airport operations have significantly affected bird populations. There has been an increasing number of bird strikes, generally fatal to individual birds involved, reported worldwide (annual average of 12,219 reported strikes between 2008–2015 being nearly double the annual average of 6,702 strikes reported 2001–2007) [ 8 ]. Airport operations including construction of airport infrastructure, frequent take-offs and landings, airport noise and lights, and wildlife hazard management practices aimed at reducing risk of birdstrike, e.g., spraying to remove weeds and invertebrates, drainage, and even direct killing of individual hazard species, may result in habitat fragmentation, population decline, and rare bird extinction adjacent to airports [ 9 – 11 ]. Nevertheless, there remains an imperative to continually improve wildlife hazard management methods and strategies so as to reduce the risk to aircraft and to bird populations.

Certification or accreditation of an airport with the regulatory authority is contingent on the airport implementing a Wildlife Hazard Management Plan (WHMP) for the management of wildlife hazards. In Australia, compliance with the Civil Aviation Safety Authority (CASA’s) Advisory Circular AC 139–26(0) Wildlife Hazard Management at Aerodromes [ 12 ] requires of airport operators, that when assessing wildlife hazard risk, “individual species should be identified and prioritised in order of risk” [ 12 ]. AC 139–26(0) also points out the airport operator’s requirement to provide advice to airmen at varying levels of immediacy including (i) standing caution for a bird or animal hazard that poses a constant risk, (ii) periodic when there is a significantly increased risk, for a relatively constrained period of time, posed by a hazard species, and (iii) immediate advice of risk from the control tower to warn approaching and departing aircraft of, say, birds (or bats), detected in the flight path.

Essential components of a WHMP are risk assessment techniques or tools that (i) comply with regulations and, (ii) are sensitive enough to properly align risk with changing conditions around the airfield. Such a management plan (risk mitigation strategy) will incorporate reactive elements (such as providing warnings to pilots as they approach or depart from the airport regarding the presence of one or more hazard species) and proactive elements (such as altering habitat in the vicinity of the airport with a view to reducing hazard species’ populations over time).

Various wildlife hazard risk assessment approaches have been proposed/implemented around the world. Common to all, is the aim of reducing the likelihood of birdstrike, however differences in approaches characterised by their (i) sphere of application (local airport to (inter-)national), (ii) forecast horizon (static to real-time), and (iii) realisation (as being technique/algorithm, a model, or an implemented system). Here we take the definitions of technique (theoretical, mathematically based structure), model (framework of techniques and algorithms to describe/predict real word situation), and system (interacting components including models, sensors, communications networks to utlise model outputs) from [ 13 ]. The following models/systems are at large spatial scale, i.e., larger scale than individual airport. The models/systems differ also in their forecast horizon. For example, real-time birdstrike warning systems include the Avian Hazard Advisory System (AHAS) [ 14 ] in the United States, the Dutch Radar Observation of Bird Intensities system (Dutch ROBIN system— https://www.robinradar.com/ ) in the Netherlands. In Europe, the FlySafe Bird Avoidance Model [ 15 ] provides near real-time information and forecast on large scale bird movement in the air space of The Netherlands, Germany and Belgium. Longer forecast horizon model/systems include the United States and North America Bird Avoidance Model (USBAM) [ 16 ], the German birdstrike risk forecast model [ 13 ], and the Swiss/Dutch Dynamic Bird Migration Model [ 13 , 17 ]. The USBAM is a statistical model, while the German birdstrike risk forecast model is a conceptual model, and the Swiss/Dutch dynamic bird migration model is a simulation model. Purely static, wildlife species risk ranking techniques/algorithms include Allan’s formula [ 11 ], Paton’s approach [ 18 ], and Carter’s formula [ 19 ] These techniques are intended for application at individual airports.

Little attention has been paid to applying machine learning to aspects of birdstrike risk. Rosa et al. [ 20 ] applied 6 different machine learning techniques to data collected by marine radar in Portugal. This study concluded that all techniques were able to distinguish between birds and stationary objects, but distinguishing between different species of birds was less successful. Recently, Nimmagadda et al. [ 21 ] applied decision tree and naive Bayesian approaches to the problem of predicting whether an airline crash has occurred due to a birdstrike. Verma et al. [ 22 ] provides a literature review of various techniques for prediction of general aviation accidents including, among others, Bayesian Networks, Artificial Neural Networks, data mining, and ensemble approaches. None of these techniques were specific to predicting likelihood of wildlife collision on/near an airfield.

Risk assessment techniques/methods generally integrate the likelihood of occurrence of a risk event with the severity of an event (or any other risk-related factors) to give an overall risk rating. Examples of such techniques can be found in [ 19 , 23 , 24 ].

Allan’s formula [ 7 ], endorsed by the Australian Airports Association for use in Australian airports [ 25 ], assesses hazard species’ risk by combining the probability of the species being involved in a strike event (determined using only historical strike data) with the estimated severity of a strike (calculated from the average mass of an adult member of the hazard (bird) species adjusted for flocking behaviour). The principal function of Allan’s approach is to identify and rank hazard species. While being easy to implement in practice, Allan’s approach has limitations as a day-to-day indicator of risk due to wildlife including (i) any species has a zero probability of involvement in a collision (and hence a low risk) until a strike is actually recorded, and (ii) the risk is largely static since it is derived from a rolling window of the past 5 years’ observations, i.e. it does not take immediate conditions or short-term changes or trends into account.

Similar to Allan’s formula, Paton’s approach [ 18 ] is data-driven and uses a severity/likelihood matrix to assign a risk rating to a species (with however, significant differences to Allan’s formula in the way in which the severity and likelihood categories are calculated). In particular, the likelihood indicators used in Paton’s approach acknowledge that conditions on the airfield vary over time (e.g. number of birds on the airfield), making it sensitive to changing environmental conditions and potentially useful as the basis for a dynamic risk model.

It should be noted that the principal use of both Allan’s approach and Paton’s approach is in identifying and ranking hazard species. As such, both Allan’s formula and Paton’s approach are accepted by Australia’s Civil Aviation Safety Authority (CASA). Neither method however, is suitable for providing a risk assessment directly related to current conditions at the airport .

Further, static risk ranking approaches do not take into account seasonal or environmental factors, and hence, cannot be used to (i) reason about likely future states on the airfield, or (ii) to plan mitigation measures in anticipation of some likely future states. By contrast, the techniques developed in this study do provide evidence-based, dynamic, collision likelihood assessments, and can be used to inform wildlife hazard managers of likely, required, mitigation activities. Here we note that risk assessment also includes taking into account a severity measure. This aspect was outside the scope of this paper and we concern ourselves with only collision likelihood.

Materials and methods

Study area and data collection.

To develop these methods, a case study was conducted at the Brisbane International Airport, Australia located at the geographic coordinates 27°23’07.58” S, 153°07’13.48” E and an elevation of -3 m below mean sea level. It is located adjacent to the Brisbane coastline in a temperate (mesothermal) climate zone, defined as humid subtropical climate (Cfa) and temperate oceanic climate (Cfb) according to the Koeppen climate classification [ 26 ] system. The climate is characterised by hot and warm summers, without a dry season, although the average precipitation is higher in summer months than in winter months. Average temperatures range between 8°C in the coldest month (July) to 27°C in summer months. Within a 3km radius of the Brisbane airport natural features and wildlife attractants include the port of Brisbane, an oil refinery, refuse dumping ground, waste transfer station, a major golf club, big cemetery, wetlands reserve, waste water treatment plant, the Kedron Brook Floodway and several smaller brooks and creeks.

The (routinely collected) environmental, geographical and wildlife data used in this paper was provided by Brisbane Airport Corporation (BAC) in Australia and covers the period 1 May 2017 to 30 June 2019. The selection of variables was made in collaboration with our industry partner (BAC) and incorporated the expert knowledge of BAC’s wildlife management team. The data set comprised six main data aspects: (i) environmental data such as daily rainfall and temperature, (ii) existing infrastructure features such as runways, taxi ways, terminal buildings, etc., (iii) a proximity weighting factor for zones on the airfield (relative to the distance of the zone from areas of high-speed aircraft movement such as runways where strikes are most frequently reported), (iv) details of reported bird strikes during the 761 days (1 May 2017 to 30 June 2019), (v) daily counts of three bird species nominated as hazard species of interest by BAC, and (vi) details of mitigation (harassment) strategies conducted to reduce the number of birds adjacent to aircraft movement corridors.

Wildlife observations are carried out daily between 06:00 and 09:00 by the WHM team at BAC. The count is conducted by a single team member, (usually) traversing the airport on a standard route. The airport is divided into 11 zones. The daily count data records the date and time of wildlife observations, the species observed and the number of individuals of the species, as well as the location on the airport (zone) where the species was observed. Environmental variables such as rainfall, wind speed, wind direction, cloud height and cover are also recorded. Wildlife ‘harassment’ activities are also recorded (harassment is moving hazard species away from runways and taxi-ways). The date, time, and method (e.g. siren, pyrotechnics, etc.) of harassment activities are recorded along with the species and number of individuals harassed. Lastly, details of strikes are recorded including date and time, species and number involved, location (zone), aircraft, damage to aircraft, and whether the strike can be confirmed (by remains on the ground or evidence on the aircraft). S1 Table summarises the strike information for the three hazard species involved in the study during the period of the study.

BAC nominated the three hazard species it considered as being high risk species for inclusion in the study. The Cattle Egret ( Bubulcus ibis ) and Nankeen Kestrel ( Falco cenchroides ) were included due to their comparatively frequent involvement in collisions. The Straw-necked Ibis ( Threskiornis spinicollis ) was included due to the large numbers of this species (flocks of thousands of individuals) frequently observed on the airfield. The data set was a rectangular array—761 rows for each of three bird species, comprising 80 explanatory variables as a mix of continuous real values, categorical attributes and ordinal data.

Discussions with the BAC WHM team added some context to some relevant behaviours of the study hazard species.

  • Nankeen Kestrels are small, non-migratory, raptors (preying on small mammals and insects) which breed July-Nov, and exhibit a strong seasonal pattern in their mode of hunting [ 27 ], alternating between hover-hunting (riding thermals, particularly over concrete runways and taxiways where aircraft are moving) and still-hunting (launching from a perch). Nankeen Kestrels are the most frequently struck bird species at Brisbane airport, with strikes throughout the year.
  • Cattle Egrets are migratory and have high numbers from November till February (southern summer), with breeding occurring October till January with markedly fewer individuals observed in March–October (southern winter). Cattle Egrets feed on grasshoppers and other insects, foraging on open grassland.
  • The Straw-necked Ibis population on/around the airport is the most volatile and inconsistent of the three hazard species, and while numerous, are rarely involved in strike events. These fluctuations cannot be explained by specific criteria such as migratory behaviour, seasonality, or hunting behaviour. Instead, the Straw-necked Ibis population appears to be strongly linked to food availability (leading to birds being attracted to the airport precincts, and increased breeding).

The main objective of this study was to investigate a set of data-driven modelling techniques, from the point of view of generating dynamic collision likelihood scores. It was decided to test three (3) separate modelling approaches based, in part, on what has been described in the literature, and in part on novelty of approach, i.e. not described in the bird-strike literature but which has application in related data-driven analyses. The techniques included (i) an algebraic approach based on function approximation, (ii) a probabilistic approach using Bayesian networks, and (iii) an unsupervised machine learning approach utilising K-means clustering. Note that supervised machine learning techniques for modelling likelihood of bird strikes were not considered due to the small number of actual strikes recorded against each hazard species. Table 1 provides a summary of the essential features of each selected modelling approach.

thumbnail

  • PPT PowerPoint slide
  • PNG larger image
  • TIFF original image

https://doi.org/10.1371/journal.pone.0277794.t001

The algebraic (function approximation) approach was to use variables used in Allan’s and Paton’s approaches, and variables identified in the literature as being indicative of collision risk, to develop an annual risk profile for each of the selected hazard species. The Bayesian network approach was intended to provide a likelihood rating of a strike involving a given hazard species under a given set of environmental conditions. K-means clustering and associated cluster analysis was selected with a view to ascertaining whether it was possible to (i) separate (daily) observations into classes, (ii) to then characterise classes (and hence, species-day observation units) as being high or low risk, and (iii) use cluster analysis as a means of feature reduction/selection (identifying the most indicative variables of the 80 daily measurements).

Algebraic modelling

This approach is based on the following observations and assumptions.

  • Collisions between birds and aircraft only happen when they are both in the same (air)space at the same time.
  • While being rare in absolute terms (i.e. fewer than 10 strikes per 10,000 flights), strikes happen frequently enough for their occurrence to not be completely random.
  • There exist some combination(s) of (environmental) factors that lead to birds and aeroplanes being in the same space at the same time.
  • Some factors are so-called ‘lead’ indicators of likelihood of collision. For instance, season is associated with certain usual weather conditions (e.g. rainfall, temperature, hours of daylight, etc.) that influence ‘intermediate’ indicators such as grass coverage (which in turns influences insect population, etc.). Both lead and intermediate indicators impact on ‘immediate’ likelihood indicators such as numbers of birds on/near runways (see Fig 3).
  • These factors and the relationships between them can be discovered through data analysis of historical environmental (observation, weather, harassment and strike) data.

write a term paper on bird strike

Paton, when describing his risk assessment model [ 18 ], outlined various likelihood indicators including (i) relative abundance, (ii) frequency of occurrence and or area of occurrence. Carter, in his study on risk assessment and prioritisation of wildlife hazards [ 19 ] includes ten likelihood indicators including (i) overall population of the wildlife species (in total number of individuals), (ii) location of the species with respect to flight operations, and (iii) number of reported strikes involving the species. Importantly, in [ 19 ], the author also mentions the ability to actually influence the species through wildlife control (harassment) as being relevant to likelihood of involvement in a collision. Lastly, analysis of our own K-means clustering results reveal the significance of the number of individuals belonging to hazard species observed and their proximity to runways (see Results…K-Means Clustering).

In this study, it was decided to include (i) a seasonality measure, (ii) abundance which we define as the number of hazard species counted in the daily count as well as the number of hazard species involved in harassment operations, (iii) overall count of wildlife and, (iv) location of observed hazard species with respect to flight operations (proximity to runways) as key variables to include as model inputs.

write a term paper on bird strike

Such an approach means that, if 1-Jan is used as the marker date, days at opposite ends of the calendar year (e.g. in January and December) are seasonally close, while dates in January and June are seasonally distant.

The airfield is divided into eleven wildlife management zones. A location indicator for each zone was derived based on the proximity of the zone from the active runways and approach/take-off pathways. Let Z be the set of zones on the airfield, and P be the set of all proximity measures. Let Z p ⊆ Z denotes the set of zones which have proximity measure p where p ∈ P .

The abundance of a hazard species was calculated on a daily basis as:

write a term paper on bird strike

The effect of multiplying by 1/proximity is to give more weight to hazard species that are closer to runways and approach/take-off flight paths. For any hazard species h ∈ H , zone z ∈ Z , and proximity p ∈ P :

write a term paper on bird strike

Paton [ 18 ] points to population and proximity of hazard species to runways, and seasonality as factors influencing likelihood of collision. In our approach, we have a single measure, abundance, which combines population and proximity, as well as a seasonality measure. Plotting (seasonality, abundance) pairs across the year gives an abundance distribution and allows us to fit a curve, expected abundance , for some hazard species h as e h = f ( s , a h ). In [ 28 , 29 ] techniques are described for constructive function approximation using Gaussian kernels and sigmoid functions respectively.

The graph of a Gaussian is a characteristic symmetric ‘bell curve’ shape in which the parameter a i represents the height of the curve’s peak, b i is the position of the center of the peak, and c i controls the width of the ‘bell’. Such a curve is useful in modelling peaks in population over time.

write a term paper on bird strike

The graph of a sigmoid function has a characteristic ‘S’ shape. The difference of two sigmoids may be used to cut off a continuous range of the domain to form a ‘bump’ where the difference between the two sigmoid functions is close to the maximum value of the function over this range, and near the minimum value elsewhere [ 30 ]. Such a function is useful in modelling a population that is relatively stable over a period of time.

write a term paper on bird strike

Bayesian networks

Bayesian Networks are a method for graphical representation and probabilistic calculation in uncertain and complex scenarios [ 31 ]. Bayesian Networks require (i) a set of random variables, (ii) the conditional relationships that exist between them, and (iii) their probability distributions. A Bayesian network is a directed acyclic graph (DAG) such that each node represents a random variable and has associated probabilistic information. Conditional relationships between variables (nodes) are represented by arcs (edges) joining a node to other nodes.

Thus, Bayesian Networks should have the capability of accounting for all the conditional relationships and uncertainties (such as changes in the number of and types of birds, weather, aircraft movements, runway usage, etc.) as they relate to the likelihood of bird strike making them a good approach for risk assessment. Bayesian approaches have been mentioned in relation to climate and environmental factors influencing forest management [ 32 ], for imputing missing values in a data set used for analysis of engine failure following bird strike [ 33 ], and for estimating structural damage to aircraft following a bird strike [ 34 ]. To date, however, Bayesian Networks have not been applied to assess factors influencing the likelihood of wildlife hazard strike occurrence at airports. To address this gap, we show how Bayesian Networks can be utilised in the context of bird strike likelihood.

write a term paper on bird strike

https://doi.org/10.1371/journal.pone.0277794.t002

The Bayesian Networks were developed in GeNIe ( https://www.bayesfusion.com/genie/ ), by integrating all the parameters identified as having direct or/and indirect effect on the occurrence of strikes (see Table 2 ). Fig 1a–1c respectively show the design configuration of BNM for Nankeen Kestrel, Straw-necked Ibis, and Cattle Egret respectively.

thumbnail

(a) BNM design for Nankeen Kestrel, (b) BNM design for Straw-necked Ibis, (c) BNM design for Cattle Egret.

https://doi.org/10.1371/journal.pone.0277794.g001

Models were validated using K-fold cross validation (K = 2). Then, the Expectation Maximization (EM) method was used to discover maximum-likelihood estimates for all Conditional Probability Table (CPTs) and for refitting the case file data to the final model while minimizing negative log likelihood. Parameters of the model are re-learned each fold, while the structure of the model remained fixed.

A Receiver Operating Characteristic (ROC) curve can efficiently express the quality of a Bayesian Network model and demonstrate “the theoretical limits of accuracy of the model on one plot” [ 35 ]. The ROC curve plots the true positive rate (Sensitivity) against the false positive rate (100-Specificity) considering multiple cut-off points of a parameter. ROC curves are able to express the quality of a model effectively. The ROC curve showcases an insight into the trade-off with selecting any states of a variable/node. The diagonal line on the plot illustrates a baseline ROC curve. In other words, a ROC curve above this diagonal line is a classifier that demonstrates that the model works effectively. The Area Under the Curve (AUC) is an easy way to identify the quality of the model with a number shown in each ROC diagram. Model quality was assessed by generating ROC curves for each BNM.

(K-Means) clustering.

The rationale for applying K-means clustering [ 36 ] as a data exploration technique was (i) to determine whether airfield days could be clustered into day ‘types’, i.e. grouped by similarity, (ii) to identify the principal data attributes used to differentiate between day ‘types’, and (iii) to see whether particular day types are associated with bird strikes.

Accordingly, we took the following steps.

  • Applied unsupervised learning (clustering) to the BAC wildlife observational data to build a ‘codebook’ of day types. Note that details of strikes were not included in this phase.
  • Validated the models using internal and external validation.
  • Extracted indicative variables using principal component analysis.
  • Used bootstrapping and comparision of collision probabilities at 95% confidence intervals to determine prediction accuracy, i.e., if some day types were actually associated with strikes.

Unsupervised K-means partitions n observations into K clusters such that each observation is assigned to only one cluster, and observations in the same cluster are similar to each other as much as possible, and observations in different clusters are distinctly different. The K centroids serve as prototypes of the respective clusters to which they belong.

Internal validation of clustering may be assessed using silhouette plots. A silhouette plot [ 37 ] measures how well each individual observation is assigned to its respective cluster. The y-axis of a silhouette plot shows all the individual data points and the x-axis gives an indication if the data point is farthest away from neighbouring clusters (+1), or is on the decision boundary between two neighbouring clusters (0), or might have been assigned to the wrong cluster (-1). Also, the width the individual silhouette cluster plots and the average silhouette coefficient can give insight if the optimal number of clusters was determined correctly.

Principal Component Analysis (PCA) [ 38 ] gives insight into patterns and relationships that exist among the variables, and helps in reducing the complexity of high-dimensional and highly correlated data, while retaining as much of the variance in the dataset as possible. Keeping only the first two principal components finds the two-dimensional plane through the high-dimensional data set in which the data is most spread out. So if the data contains clusters these too may be most spread out, and therefore most visible to be plotted out in a two-dimensional diagram. As a final anlaysis, we applied PCA to identify variables significant in forming clusters.

Thus, in our approach, each cluster will represent a day ‘type’ and the principal components, identified by PCA, will give insights into the data attributes that differentiate between day types.

Lastly, by calculating the probability of collision on any given day with the probabilitity of collision given that the day belonged to a particular cluster and comparing (95%) confidence intervals for overlap, we determine whether day ‘types’ are associated with collisions for a particular hazard species. To deal with the low frequency of collisions recorded in the observational data, we use bootstrapping to derive the confidence intervals.

To optimise the data for presentation to the K-means algorithm, we applied data standardisation [ 39 , 40 ] to address issues of different data ranges, units of measure, and variance apparent in the variables.

Following clustering, internal and external validation measures were applied to determine the goodness of the clustering solution. Internal cluster validation measures reflect on the compactness, connectedness, and the separation of the cluster partitions, and can be determined by the silhouette coefficient (with range [-1..1] [ 37 ]) and the Dunn index (with range [0..∞) [ 41 ]). The goal is to maximise both the silhouette coefficient the Dunn index [ 42 ]. If the data set contains compact and well-separated clusters, the diameter of the clusters is expected to be small and the distance between the clusters is expected to be large (silhouette coefficient will be close to 1 and the Dunn index will be large). That is, we want the average distance within cluster to be as small as possible; and the average distance between clusters to be as large as possible.

The computational environment was the R statistical modelling software version 4.2.0 [ 43 ]. The get_clust_tendency() function was used to determine the Hopkins statistic [ 44 ] as an indicator of suitability of the data for clustering. The returned (average) value of 0.1, being close to zero, indicated the data was suitable for clustering. The fviz_nbclust() function, using each of the three methods provided for estimating the optimal number of clusters, gave recommended number of clusters, for each species, as two, i.e. K = 2. The kmeans() function was used as the implementation of K-means clustering. The clvalid() function was used to return internal and external measures of validity. The PCA() and get_pca_var() functions were used to return principal components.

The expected abundance curves for the three hazard species are shown in Fig 2 . Fig 2a ) clearly reflects the migratory behaviour of the Cattle Egret. The solar day was zeroed at 1-May meaning values <0 represent the period from November to May. This is the period during which Cattle Egrets begin to arrive in Brisbane (November) and leave Brisbane (late April), with the population peaking in February. This is also the period where the most strikes (orange markers) involving Cattle Egrets were observed. Fig 2b shows the abundance values against solar day (seasonality) for the Nankeen Kestrel. In this case, the solar day (seasonality) value was zeroed at 1-Nov meaning values <0 represent the period from May to November (cooler months of the year with shorter day lengths). Fig 2c shows the abundance values against solar day (seasonality) for the Straw-necked Ibis.

thumbnail

Blue markers indicate abundance on a particular day, orange markers shows days where at least one individual of the relevant hazard species was involved in a strike occurrence, and grey markers give the expected abundance on any day. (a) Cattle Egret abundance, seasonality, and expected abundance, (b) Nankeen Kestrel abundance, seasonality and expected abundance, (c) Straw-necked Ibis abundance, seasonality, and expected abundance.

https://doi.org/10.1371/journal.pone.0277794.g002

Table 3 shows the results of logistic modelling used to determine the most indicative attribute, and associated threshold value to be used in generating collision likelihood for each hazard species.

thumbnail

https://doi.org/10.1371/journal.pone.0277794.t003

write a term paper on bird strike

The kernel mix parameters for each of the hazard species abundance models are given in S3 Table .

For the Nankeen Kestrel, Straw-necked Ibis and Cattle Egret hazard species, the accuracy of the predicted likelihood is 0.9, 0.96 and 0.85 respectively. The ROC curves for the states of Likelihood (High) are shown in Fig 3 .

thumbnail

(a) Nankeen Kestrel data set, (b) Straw-necked Ibis data set, (c) Cattle Egret data set.

https://doi.org/10.1371/journal.pone.0277794.g003

Figs 4 – 6 respectively show the BNM’s after training for Cattle Egret, Nankeen Kestrel, and Straw-necked Ibis respectively. Each figure highlights the ‘influence strength’ of connected nodes (shown as arrow width and direction), and includes the conditional probabilities and variables associated with high and low likelihood of collision involving the respective hazard species. For instance, in Fig 4 (i) Season strongly influences Seasonal Behaviour (the Cattle Egret is a migratory species and breeds in summer), and (ii) the likelihood of collision is high when the number of Cattle Egrets Observed is high, with highest collision likelihood when the Population per Zone is High, the number of Cattle Egrets Observed is High, the overall Population of hazard species is Medium, and the number of Aircraft Movements is High.

thumbnail

Node influence strength is given by the width of connecting arcs.

https://doi.org/10.1371/journal.pone.0277794.g004

thumbnail

https://doi.org/10.1371/journal.pone.0277794.g005

thumbnail

https://doi.org/10.1371/journal.pone.0277794.g006

Conditional probabilities for all nodes, from each of the trained networks are available from our github repository ( https://github.com/robertandrews59/ABCBirdstrike ).

K-means clustering

Clustering results are visualised in the 2-dimensional plots in Fig 7 with cluster validity measures shown in Table 4 .

thumbnail

https://doi.org/10.1371/journal.pone.0277794.t004

thumbnail

The axes for these plots are the principal components of the multi-dimensional vectors used as input to the K-means algorithm. (a) Cluster results for Cattle Egret where k = 2. Cluster 1 includes 187 days, cluster 2 includes 574 days. (b) Cluster results for Straw-necked Ibis where k = 2. Cluster 1 includes 191 days, cluster 2 includes 570 days. (c) Cluster results for Nankeen Kestrel where k = 2. Cluster 1 includes 130 days, cluster 2 includes 631 days.

https://doi.org/10.1371/journal.pone.0277794.g007

Internal validation was by way of silhouette plots. Fig 8 is a silhouette plot for the hazard species Cattle Egret. The 761 observations form two (2) clusters with the average silhouette coefficient 0.33. (The higher the average silhouette coefficient, the more clearly clustering has segmented the data.) Cluster 2 incorporates the majority of the data points (574 out of 761 data points) with 85 points in cluster 1 having a negative silhouette width (avg. -0.11).

thumbnail

https://doi.org/10.1371/journal.pone.0277794.g008

Figs 9 and 10 show silhouette plots for the Straw-necked Ibis and Nankeen Kestrel hazard species, again with each plot including the 761 observations of the respective hazard species’ data set. For the Straw-necked Ibis, cluster 2 incorporates the majority of the data (570 out of 761 data points) with 83 of the 191 points in cluster 1 having a having a negative silhouette width. For the Nankeen Kestrel, cluster 2 incorporates the majority of the data (631 out of 761 data points) with 57 of the 130 points in cluster 1 having a negative silhouette width.

thumbnail

https://doi.org/10.1371/journal.pone.0277794.g009

thumbnail

https://doi.org/10.1371/journal.pone.0277794.g010

Analysis of the distribution of bird strikes by species across clusters is shown in Table 5 . Analysis is hampered by the low number of strikes observed over the two year period of the study. However, K-means clustering shows that:

  • for Cattle Egrets, essentially all strikes occur in day type represented by cluster 1;
  • for Straw-necked Ibis, strikes are recorded in both day type clusters; and
  • for Nankeen Kestrels, strikes are recorded in both day types, with the majority of strikes occurring in day type represented by cluster 2.

thumbnail

https://doi.org/10.1371/journal.pone.0277794.t005

Examination of strike data in S1 Table shows:

  • all but one of the strikes involving Cattle Egrets occurred in summer months (Nov-Feb), with a single strike in a winter month;
  • of the three strikes recorded for Straw-necked Ibis, two were in summer months and one was in a winter month;
  • of the 20 strikes recorded for Nankeen Kestrels, 13 occurred in the months between April and October (cooler months) while 7 occurred in the period December to March (warmer months).

Table 5 also shows, for each of the hazard species, the overall probability of a strike on any given day, the probability of a strike given the day falls into a given cluster, and 95% CI for each of the probabilities. Significance is determined by overlap of the confidence intervals. This analysis indicates that for the hazard species considered, strikes are not associated with day ‘types’ (as represented by the clusters) in that, apart form one instance, the confidence intervals are not distinct, i.e., they overlap.

One of the uses of clustering was to identify variables that contribute to likelihood. Fig 11a–11c respectively show the PCA analysis for Cattle Egret, Straw-necked Ibis, and Nankeen Kestrel respectively highlighting the contributions of individual variables to PC1 and PC2 BNM for each of the three hazard species.

thumbnail

For any variable, its contribution to overall variance is given by the distance from the origin (vector length) and is colour-coded (blue indicates a lower contribution, and red the highest). (a) Plot of contribution of variables to PC1 and PC2 for Cattle Egret. (b) Plot of contribution of variables to PC1 and PC2 for Straw-necked Ibis. (c) Plot of contribution of variables to PC1 and PC2 for Nankeen Kestrel.

https://doi.org/10.1371/journal.pone.0277794.g011

S2 Table lists the two principal components responsible for the largest variation for each of the hazard species. Here, the attributes responsible for the most variation across the three hazard species include attributes related to overall population on the airfield (counted and harassed), population in the zones closest to the runways (counted and harassed), a seasonality indicator (solar day).

Our intent in conducting this work was to develop some techniques to support wildlife hazard risk assessment at airports, that went beyond only hazard ranking, to provide a more dynamic, i.e. sensitive to changing environmental factors, method of providing a collision likelihood. As previously noted, risk assessment also includes taking into account a severity measure. This aspect was outside the scope of this paper, and we concerned ourselves with only collision likelihood and the factors able to be discerned from routinely collected observational data that contribute most to collision likelihood.

A strength of the Algebraic approach is that it clearly visualised the seasonal nature of the abundance of the hazard species included in this study. In particular, the migratory behaviour of the Cattle Egret is evident as is the hunting behaviour of Nankeen Kestrels across the year. For the Cattle Egret, peak abundance corresponds with the period where the most strikes involving Cattle Egrets were observed. For the Nankeen Kestrel we hypothesise that the seasonal abundance is related to the bird’s hunting behaviour and the fixed time of the day at which the daily count is conducted, rather than actual changes in the number of birds on the airfield. In summer, the ground over the entire airfield would be warm enough, from late-morning, to generate rising air columns and hence support hover-hunting over most of the airport, particularly over concrete runways and taxi-ways, i.e., spaces where aircraft are departing and arriving. However, in winter months, when the air is cooler, Nankeen Kestrels revert to perch-hunting in the morning (corresponding to the time when the daily count is being conducted) with hover hunting occurring later in the day when the concrete runways and taxi-ways would be warmer than other parts of the airfield, and would thus generate rising air towards the middle of the day. As the daily count is conducted at the same time of day, such behaviour would affect the number of Nankeen Kestrels observed observed during the daily count.

Expert opinion from the BAC WHM indicated that the population of Nankeen Kestrels on the airfield remained constant throughout the year, yet the abundance figures (derived from the daily count) show fewer birds counted during the warmer period of the year, with more strikes occurring in this period. We ascribe this to the timing of the daily count and the different hunting behaviours exhibited by Nankeen Kestrels at different times of the year. That is, the birds were on/around the airport, but weren’t actively hunting or flying at the time the daily count was conducted, i.e., they were perching out of sight and hence were not observed during the daily count. A point worth noting is that, for 7 of the 20 strike events involving Nankeen Kestrels recorded during the study period, there were no Nankeen Kestrels recorded in the daily count or included in harassment activities conducted by the Wildlife Hazard Management team. Although Straw-necked Ibis are among the most abundant bird species on the airport precincts, they are involved in relatively few strikes (3 strike occurrences over the two year period of our study). The BAC WHM report that the species exhibits flocking behaviour, and that it is relatively easy to move the birds to parts of the airport away from runways and taxi-ways. So, although there are large numbers (often several hundreds) of birds on the airport on any given day, they are rarely in close proximity to areas of aircraft movement.

The expected abundance curve however, gives an indication of the expected number of the relevant hazard species at various times of the year and may act as supplement to the daily count of actual abundance. When overlaid with strike information, it provides an easy way to derive rules that quantify collision likelihood. A significant contribution of this approach is the notion of ‘seasonality’ instantiated in this study as ‘solar day’. Seasonality is useful in concentrating routine, daily observations into the seasonality window to make patterns of behaviour more apparent. In this study, a calendar year was used as the seasonality window. This clearly highlighted migratory patterns of the Cattle Egret, and to some extent, the changing hunting behaviour of Nankeen Kestrels at different times of the year. We note that the notion of seasonality may be applied over other time windows. For instance, seasonality could be applied to a daily 24 hour period to highlight diurnal/nocturnal behaviour patterns. Bayesian Network modelling makes explicit the causal relationships between variables associated with the likelihood of bird strike. Bayesian Networks allow, given, a set of environmental conditions, to reason forward to arrive at a collision likelihood. An additional feature of the Bayesian Network approach is the ability to reason backward from a specified condition (such as high likelihood of collision) to examine the sets of pre-conditions (causal factors) that give rise to the elevated collision likelihood. As such, Bayesian Networks could support decision-making in response to heightened immediate risk, or guide longer-term risk reduction strategies, by identifying currently significant risk factors.

Bayesian Networks, in contrast to the simplistic Risk = Severity × Likelihood , showcase the integration of all the factors that impact on both elements of risk function (i.e. Severity and likelihood). Bayesian Networks are also able to quantify uncertainty and allow the integration of known probability values associated with any nodes within the networks (prior to an adaptive learning phase). Further, Bayesian Networks provide a visual mechanism to record and test subjective probabilities, an important role in circumstances where there is not much data.

The K-means clustering approach was used to determine whether it was possible to develop a set of ‘day types’ and then determine if particular ‘day types’ were associated with strikes. For each of the hazard species considered in this study, (i) clustering generated a set of prototypical ‘day types’, (ii) strike occurrences (days on which a strike occurred) were localised within particular clusters (‘day types’). Principal Component Analysis revealed that numbers of hazard species (counted), their proximity to runways, and seasonality contributed strongly to clustering of days. We note that these attributes were shown, through logistic regression modelling in the Algebraic approach, to be predictive of elevated collision likelihood. We also note that the seasonality value perhaps subsumes other environmental indicators such as rainfall and temperature as, in Brisbane, these values have distinct ranges associated with the time of year. However, analysis of collision likelihoods for most clusters showed that there was little support for notion of day ‘types’ determined by clustering being indicative of strikes. We feel that more extensive investigation is required before completely discounting this notion.

As for practical application of the ABC approach to collision likelihood, we feel that the techniques are complementary and should be applied in parallel, and if any one of the techniques predicts a heightened likelihood of collision, actions should be taken. As discussed earlier, the techniques give indications of the factors leading to heightened collision likelihood. These could be used to inform mitigation actions.

The methods described in this paper utilise species-at-a-time modelling. We do not see this as a limitation as it is naive to assume that collisions (strikes) and methods of mitigating strikes are generally applicable to all hazard species and all airports. Our approach provides understanding at the species level, thus allowing for tailored collision likelihood modelling, and potentially tailored mitigation actions. Future work could, however, include generalising the approach to model ‘species types’. That is, hazard species could be characterised by a set of features, and risk models constructed for each hazard species type. This would have the advantage of better utilising observational data and strike data as data relating to each species belonging to a particular type would be aggregated to model the type.

The techniques described in this paper can be characterised as having local sphere of application, intermediate forecast horizon, and, at this stage, algorithmic/mode realisation (although the effort required to systematise the individual ABC algorithms would not be great). As such, the ABC approach fills an obvious gap and form a middle ground between purely static risk assessment methods such those described in [ 11 , 18 ], and immediate indicators such as avian radar making them suitable for application in a wide variety of airports around the world. Neither static, hazard species risk ranking approaches, nor immediate methods, take into account seasonal or environmental factors, hence they cannot be used to (i) reason about likely future states on the airfield, or (ii) to plan mitigation measures in anticipation of some likely future state (a further requirement of airport operators according to CASA and described in [ 12 ]). A very obvious example is that of migratory behaviour (as captured in our expected abundance model) or the temporal offset between a rain event and abundance of a given hazard species. That is a) rain, leads to b) grass growing, which provides c) increased food supply for insects and cover for small, ground dwelling animals, which act as d) attractants for insectivorous birds and birds which hunt ground dwelling animals. Such causal chains are nicely captured by the Bayesian Network approach, and can be used at the airport to plan mitigation activities such as mowing, draining of standing water, and removal of perching opportunities for birds across the airport.

In this paper we have described three techniques (the ABC approach—Algebraic, Bayesian, Clustering) useful in understanding causal factors, and in assessing the likelihood of bird strike at an airport. The techniques were evaluated with routinely collected environment and hazard species data at Brisbane airport.

Contributions of this work to the body of work dealing with aircraft collisions with (avian) hazard species at airports (i) include the definition of attributes for ‘seasonality’ and ‘proximity count’ (to deal with risk associated with hazard species at varying distances from aircraft movement corridors), and (ii) the development of three novel approaches for assessing likelihood of collision.

Each of the three methods improve on simple ‘ranking’ of hazard species, the minimum required risk assessment for airport accreditation, by providing dynamic risk assessment based on the state of environmental variables. Further, the variables used by the models are those routinely collected by airport Wildlife Hazard Management teams thus implementing the techniques at other airports should not add any burden to data collection, or require changes in data collection practices. Areas for future research around data include (i) how frequently should the historical environmental and collision data be updated in order to keep the models current, i.e., reflective of the actual collision likelihood at the airport, and (ii) what ‘window’ of historical data is optimal for modelling.

Of particular importance is the identification of variables that contribute strongly to collision likelihood. The Bayesian Network Models are well suited for ‘what if’ analysis. That is, these models allow for a set of environmental conditions to be provided to the model with the associated risk profile being generated, thus allowing for both diagnostic and prognostic use of the models.

An avenue for future work is to discover means to group ‘similar’ hazard species and apply the ABC approach to the hazard species group thus making the approach more general. Lastly, the ABC approach fits between purely static risk assessment approaches and purely immediate (e.g. avian radar) in terms of responsiveness to changing conditions on the airfield. The ABC approach also supports planning of mitigation activities through the models providing a ‘look ahead’ view allowing wildlife hazard management teams to anticipate future risk states. Lastly, this work was developmental. As such, software used was chosen that directly supported the models and provided appropriate analysis and visualisation capabilities. Future development may include development of an integrated platform making all the techniques available in the one tool. Diffusion theory [ 45 ] is concerned with the uptake and spread of innovations. It tells us, among others, that technology adoption is driven by not only technological factors, but adoption pathways, time, and the social system around the innovation. In this context, Realisation of the ABC approach as a tool for use in practice has both technological and awareness/acceptance considerations. The technological side will require, among other things, the implementation of each technique in an integrated platform, thus making all the techniques available in the one tool. Such a tool would also require options to allow wildlife control teams to configure routinely collected data for presentation to the models, and to define hazard species relevant to the airport at which the approach was to be deployed. Awareness will require disseminating information about the approach to the industry through for instance wildlife hazard working groups. Acceptance will require a quantum of early adopter airports and their subsequent communication of positive feedback to other airports, leading to more widespread adoption and ultimately approval or even enforcement by from regulatory bodies.

Supporting information

https://doi.org/10.1371/journal.pone.0277794.s001

S1 Table. Summary of the strike information for the three hazard species involved in the study during the period of the study.

Dates of strike occurrences are shown together with the number of individuals involved in the occurrence.

https://doi.org/10.1371/journal.pone.0277794.s002

S2 Table. Complete listing of the two principal components, including variable loadings, responsible for the greatest variation as used in clustering models.

https://doi.org/10.1371/journal.pone.0277794.s003

S3 Table. Configuration parameters for the gaussian and sigmoid functions used in Algebraic models for each of the hazard species, and the ‘day 0’ value for each of the hazard species.

https://doi.org/10.1371/journal.pone.0277794.s004

Acknowledgments

We gratefully acknowledge Brisbane Airport Corporation for making members of its Wildlife Hazard Management team available to the research team.

  • View Article
  • Google Scholar
  • 3. Australian Transport Security Bureau (ATSB). Aviation Occurrence Statistics: 2010–2019. Australian Transport Safety Bureau; 2020. Available from: https://www.atsb.gov.au/media/5778822/ar-2020-047_final.pdf [last accessed 2021-07-12].
  • 4. Federal Aviation Authority. FAA Wildlife Strike Database; [last accessed: 2021-08-02]. https://wildlife.faa.gov/search .
  • 5. Dolbeer R, Begier M, Miller P, Weller J, AL A. Wildlife strikes to civil aircraft in the United States 1990–2021, Federal Aviation Administration. National Wildlife Strike Database. 2022;.
  • 6. DeVault TL, Blackwell BF, Belant JL, Begier MJ. Wildlife at Airports. USDA, APHIS, WS National Wildlife Research Center, Fort Collins, Colorado; 2017. 19p.
  • PubMed/NCBI
  • 8. International Civil Aviation Organisation. ICAO Safety Report 2018 Edition; [last accessed: 2022-06-26]. https://www.icao.int/safety/Documents/ICAO_SR_2018_30082018.pdf .
  • 9. Kelly T, Allan J. Ecological effects of aviation. In: The Ecology of Transportation: Managing Mobility for the Environment. Springer; 2006. p. 5–24.
  • 12. Civil Aviation Safety Authority (CASA). Wildlife Hazard Management at Aerodromes. Advisory Circular (AC 139-26(0)). 2011; p. 139–26(0).
  • 14. Kelly TA, Merritt R, Donalds TJM, White RL. The Avian Hazard Advisory System. In: 1999 Bird Strike Committee-USA/Canada, First Joint Annual Meeting, Vancouver, BC; 1999. p. 20.
  • 16. DeFusco R, Hovan M, Harper J, Heppard K. North American bird strike advisory system. In: Bird Strike North America Conference; 2005. Available from: http://digitalcommons.unl.edu/birdstrike2009/8/ .
  • 18. Paton DC. Bird risk assessment model for airports and aerodromes. Australia: The University of Adelaide; 2010. Available from: https://canadianbirdstrike.ca/wp-content/uploads/2018/02/Paton_2010.pdf [last accessed 2021-07-12].
  • 19. Carter NB. All Birds Are Not Created Equal: Risk Assessment and Prioritization of Wildlife Hazards at Airfields. In: 2001 Bird Strike Committee-USA/Canada, Third Joint Annual Meeting, Calgary, Alberta; 2001. p. 8. Available from: https://digitalcommons.unl.edu/birdstrike2001/8/ [last accessed 2021-07-12].
  • 21. Nimmagadda S, Sivakumar S, Kumar N, Haritha D. Predicting Airline Crash due to Birds Strike Using Machine Learning. In: 2020 7th International Conference on Smart Structures and Systems (ICSSS). IEEE; 2020. p. 1–4.
  • 22. Verma S, Kumar P, et al. A Comparative Overview of Accident Forecasting Approaches for Aviation Safety. In: Journal of Physics: Conference Series. vol. 1767:1. IOP Publishing; 2021. p. 012015.
  • 31. Fenton N, Neil M. Risk assessment and decision analysis with Bayesian networks. CRC Press; 2012.
  • 34. Dİkbayir HS, Bülbül Hİ. Estimating the Effect of Structural Damage on the Flight by Using Machine Learning. In: 2018 17th IEEE International Conference on Machine Learning and Applications (ICMLA). IEEE; 2018. p. 1333–1337.
  • 35. BayesFusion LLC. GeNIe Modeler User Manual; 2018. Available from: https://support.bayesfusion.com/docs/GeNIe.pdf [last accessed 2021-07-12].
  • 39. Larose DT, Larose CD. Discovering knowledge in data: An introduction to data mining. John Wiley & Sons; 2014.
  • 40. Allen M, Cervo D. Data Quality Management. In: Allen M, Cervo D, editors. Multi-Domain Master Data Management. Boston: Morgan Kaufmann; 2015. p. 131–160.
  • 43. R Core Team. R: A Language and Environment for Statistical Computing; 2013. Available from: http://www.R-project.org/ .
  • 45. Rogers EM. Diffusion of Innovations. New York:Free Press; 1962.

Academia.edu no longer supports Internet Explorer.

To browse Academia.edu and the wider internet faster and more securely, please take a few seconds to  upgrade your browser .

Enter the email address you signed up with and we'll email you a reset link.

  • We're Hiring!
  • Help Center

paper cover thumbnail

Bird Strike to Aircrafts An Assessment of Changing Bird Populations at Select Indian Airfields

Profile image of S Srinidhi

2021, Defence Life Science Journal

Bird Strikes (BS) are a significant threat to flight safety and a serious economic concern in the aviation industry. Variation of population and activity of different birds over an airfield leads to variation in their vulnerability for Bird Strike as well. In this study, an attempt was made to document the monthly variation of bird activity over three Indian airfields situated in different bio-geographical provinces in the year 2019-20. A significant activity of Black Kites (including the sub-species Black-eared Kite namely Milvus migrans govinda and Milvus migrans lineatus) and Lapwing (Vanellus indicus) were studied to understand their annual cycle as well as long term changes in their activity over airfields (over 30 years). Agra recorded an increase of 10.3 times in the activity of Black Kites in forty years. Black Kite data in correlation with the past information on Vultures indicate that the Black Kites are taking over the ecological niche of Vultures. Sirsa recorded an incre...

Related Papers

Roohi Kanwal Zoology

Roosting in birds is a common phenomenon in Avian Biology. Roosting behavior includes the mode, timing, duration, distance, ecological and seasonal patterns which are topics of primary consideration for the naturalists and avian ecologists. One of the aspects of the bird roosting is its hazardous effects on the airfields. In Pakistan a number of accidents have occurred every year as a result of bird collisions with aircraft. This is of concern for the aircrafts. Some accidents have caused damage to civilian aircrafts as well in Pakista and the present study is undertaken with special reference to three aerodromes in Karachi. Reporting of the wildlife collisions with aircraft is recorded all over the world.. According to the literature, collision between the birds and aircrafts are known to cause substantial loss in the avian industry in terms of damage and delay every year. The existing information reveals that the changes in the bird population around airfield should be noted. The relationship between the bird abundance and strike frequency is complex and the changes in bird number coincide with changes in strike frequency. In the present study 47 species of birds have been recorded with their seasonal occurance data which pose a great threat to aircrafts during Flight, Landing and Take Off. 44 species of flora were recorded which play an important role in providing nesting and roosting areas for avifauna. Globally, bird strikes have been of great safety concern. It is expected that proposed research would advance knowledge about the roosting of birds and methodologies to turn the airport into safe areas.

write a term paper on bird strike

International Journal of Scientific Research in Biological Sciences

Yossi Leshem

In order to serve the flight safety community and mufti-disciplinary research projects a global database on bird movements and bird strikes in military and civilian flight is being developed at the International Center for the Study of Migration at Latrun. The information system will include biological and statistical data from various sources that have been collected through the years such

Satheesan S M

Journal of Wildlife Management

Richard Dolbeer

Environmental monitoring and assessment

Goutam Saha

Although Black Kites (Milvus migrans govinda) serve as major scavenging raptor in most of the urban areas, scientific studies on this important ecosystem service provider are almost non-existent in Indian context. The present study was carried out in a metropolis in eastern India to find out the factors influencing relative abundance and roosting site selection of Black Kites. Separate generalized linear models (GLMs) were performed considering encounter rate and roosting Black Kite abundance as response variables. The study conclusively indicated that encounter rates of Black Kites were significantly influenced by the presence of garbage dumps in its vicinity. Numbers of Black Kites were also higher in the roosting sites situated closer to garbage dumps and open spaces. In addition, expected counts of Black Kites significantly increased in roosting sites situated away from buildings and water bodies. However, built-up area and tree cover around the roosting sites had no influence o...

Manoiu Valentina-Mariana , Stefan Gheorghe , Adrian Tiscovschi , Răzvan-Mădălin Spiridon

This paper tackles a never-before-approached topic in Romania, as it attempts to empirically analyse the 26 cases of collision between military aircrafts and birds that occurred between 1967 and 1991. The work is based exclusively on observations made by the Head of the Meteorological Department of the Romanian Military Aviation Commandment between 1990 and 1993, commander (retired) and meteorologist. This paper complements various scientific and statistical studies in the field, which lacked information about military aircraft bird strikes in Romania. In addition to these 26 collisions, 3 other incidents involving military aircrafts may have been bird-related. They occurred in 2002 (2, one of which was a catastrophe that lead to the pilot's death) and 2007 (1 incident). We analysed the collisions' locations (prevalence of certain aerodromes), the atmospheric and seasonal conditions in which the collisions occurred, and the types of aircrafts and bird species involved in the incidents, which resulted in a map of the analyses' main elements. It was found that in Romania, in the analysed period, most of the collisions occurred in areas where aerodromes were located in the vicinity of water sources, in autumn, followed by spring, which is linked to the birds' migration activity, as well as to agricultural practices, as birds relocate in order to look for food sources. Most collisions occurred during the day, under normal meteorological conditions, up to altitudes of 400-500 m. It was also noticed that in over 50% of cases collisions occurred in the vicinity of aerodromes, most often during landing. The most important element that favours bird presence near aerodromes is food and water source availability, which is followed by the availability of various nesting-suitable locations. This study also touches on flight ornithological insurance, factors that stimulate bird concentration near aerodromes, recommendations for ensuring flight safety from an ornithological perspective, as well as methods used to hinder bird presence near aerodromes.

Dessalegn Ejigu

Background Bahir Dar International Airport is known by its rich avifaunal diversity, and bird-aircraft collisions is becoming a serious problem in the area. Study on bird-aircraft strike problems in Bahir Dar International Airport was carried out from February - August 2020. The study area was classified into four habitat types based on its vegetation structures namely: bushland, grassland, wetland, and modified habitat. Point and transect count methods were used to collect data on diversity and abundance of birds. Interviews to people working at the Airport were used to gather information about the incident and prevention of bird-airport strike problems. Shannon-Wiener diversity, Simpson’s similarity indices, ANOVA, and chi-square tests were used for data analysis. Results A total of 80 bird species belonging to 15 orders and 40 families were identified in the present study area. The highest species diversity (H’=3.59) and species evenness (E=0.96) were recorded in the modified hab...

Human–Wildlife Interactions

A basic tenet of programs to mitigate the risks of bird strikes with aircraft has been to focus management efforts at airports because various historical analyses of bird-strike data for civil aviation have indicated the majority of strikes occur in this environment during take-off and landing at 500 feet AGL increased signifi cantly from about 25% in 1990 to 30% in 2009. The percentage of all damaging strikes that occurred at &gt;500 feet increased at a greater rate, from about 37% in the early 1990s to 45% during 2005 to 2009. I also examined trends in strike rates (strikes/1 million commercial aircraft movements) for strikes occurring at 500 feet. From 1990 to 2009, the damaging strike rate at &gt;500 feet increased from about 2.5 to 4.0, whereas the damaging strike rate for strikes at 1.8 kg, showed a pattern similar to that for all species. I conclude that mitigation efforts incrementally implemented at airports in the United States during the past 20 years have resulted in a r...

RELATED PAPERS

International Journal of Cancer Therapy and Oncology

Luis Santana-Blank

Journal of the American College of Cardiology

zafar hashim

Bernardo Aco Castañeda

Clinical Infectious Diseases

andrea petrucca

New Zealand Journal of Crop and Horticultural Science

Emilio Parrilla

Journal of Genetics

Nguyen Hai Lam

RSC Advances

Siphelo Ngqoloda

JAMA Oncology

Eric Y . H . Chen

Diagnostics

J. Cetacean Res. Manage.

Frank Cipriano

Hector Fernandez

Francisco Lenzi

Journal of Photochemistry and Photobiology A: Chemistry

SAMY EL-DALY

Gilmor Keshet

Proceedings of the 69th International Symposium on Molecular Spectroscopy

alexander alijah

Rachasak Boonhok

Stamford Journal of Pharmaceutical Sciences

Manik Shill

Gaceta de pedagogía

Ríchard Sosa

IEEE Transactions on Information Theory

RELATED TOPICS

  •   We're Hiring!
  •   Help Center
  • Find new research papers in:
  • Health Sciences
  • Earth Sciences
  • Cognitive Science
  • Mathematics
  • Computer Science
  • Academia ©2024

Bird-Strike

Bird Strike, what is It, How Common, How Dangerous?

write a term paper on bird strike

What is a ‘Bird Strike’?

How common are bird strikes, migratory patterns, time of day, surrounding features and fauna, are bird strikes dangerous, flight control surfaces, windscreens, how do pilots avoid birds, they fly slower, weather radar, engine spinner painting, climb, don’t dive.

Being honest, the birds were there long before we ever were, so we have to accept that we share the air with them. Birds nearly always come off worse when they make contact with an airplane too! However, even small birds can pose a significant danger to aviation. Thankfully bad outcomes are exceptionally rare. Today, we will look at bird strikes, what they are, how often they happen, and how dangerous they really are.

While bird strikes are common, it is rarely a cause for concern. Manufacturers build and test aircraft to withstand most bird strikes. The only danger comes if the bird makes contact with critical or fragile components. Thankfully the chances of this happening are slim.

Let’s take a more detailed look into what constitutes a ‘bird strike’.

According to the International Civil Aviation Organization (ICAO) , a “bird strike” is defined as: –

“A collision between a bird and an aircraft which is in flight, or on takeoff or landing roll”.

Essentially, this means that during any phase of flight, a bird strike can occur.

The size of the bird or the speed at which the aircraft is traveling doesn’t matter. While the biggest danger comes from a combination of larger birds at higher aircraft speeds, there is still a small risk presented even from tiny birds.

Bird strikes are certainly not uncommon. According to published figures on the FAA ‘wildlife strike’ database in New York airports alone, there were 493 incidents reported from 1 st January 2021 to 1 st January 2022. That is more than one per day!

Or, to spread that data over a broader range, according to the FAA , there have been over 17 thousand wildlife strikes recorded in 2019!

However, while this may sound like a lot, on the whole, bird strikes don’t pose a significant risk. According to the same website mentioned above, in a 31-year-period, wildlife strikes have only accounted for 292 fatalities worldwide .

Statistically, that is 9 per year. When you factor in that in 2019, 4.5 billion people flew worldwide , the odds of a fatality are infinitesimally small!

Several things may influence how often bird strikes occur. These are areas such as: –

There is a slight uptick in the number of bird strikes in the summer and fall months. There are more birds airborne due to young birds ‘flying the nest’, and seasonal migration begins as the weather cools.

Most bird strikes occur during daylight hours, with around a third occurring at night. Many birds are ‘day VFR’ only . It tends to be nocturnal species that airplanes collide with at night.

Most aircraft collide with birds during the takeoff and landing phase. While this isn’t good news, it minimizes the risk a little, as these flight phases are when the aircraft is at its slowest, meaning there is less force when a strike occurs.

Only 3% of bird strikes occur in the USA during the en-route phase.

Airports located in rural areas or near the coast can be particularly troublesome. Rivers and lakes also attract larger birds, presenting particular difficulties and dangers.

In fact, you might be familiar with one such case where a bird strike did cause a real problem.

On the whole, bird strikes are not dangerous. After all, most commercial aircraft weigh quite a lot and are made from strong alloys designed to withstand turbulence and thunderstorms. They are also tested to extremes, with simulations and technology that will far exceed what they could expect in the real world.

Want to see the ‘technology’?

Ever seen a chicken fired through a gun at an airplane? Here it is!

However, there are several vulnerable areas that could spell trouble if a bird hits. Such as: –

Jet engines are tuned to incredibly high-performance levels, and it doesn’t take much to induce a stall or failure. The only things that should go through a jet engine are air and fuel.

When birds strike the engine, this can disrupt or damage the precise angle of the blades and stators, causing an airflow reversal. The core of a jet engine is particularly sensitive. You can see how bad it can get in this video.

Propellors already spin at phenomenal speeds, and they aren’t designed to take structural shocks while rotating.

A bird strike could shock load the engine, bend piston rods, or even break the propellor entirely.

Probes are vital to gather air data to relay to the cockpit. If these are blocked or damaged, then the data isn’t reliable. Probes tend to be one of the worst areas to sustain a bird strike. Fortunately, most aircraft have more than one system acting as a backup.

Flight control surfaces, particularly on smaller aircraft, are vulnerable to bird strikes. Because they are lighter and smaller, they are more readily damaged.

Because most bird strikes happen during takeoff and landing, often high lift devices are deployed. If a bird becomes stuck in the flaps, it can’t be retracted!

This is the biggest threat from bird strikes. Airplanes fly at high speeds . A bird flying one direction at 30mph and an airplane flying in the other at 130 mph results in a relative collision speed of around 160mph.

Glass really isn’t designed to hold up too well to collisions at this speed. If a bird penetrates the cockpit at this speed, it could severely injure the pilot!

This pilot was extremely lucky when he made contact with a goose!

Ultimate Guide to Becoming a Pilot

Pilots utilize a few techniques to minimize the danger presented by a bird strike. Here are a few to think about.

If you are familiar with physics, you’ll already know that force equals mass multiplied by acceleration. The less force there is, the smaller the likelihood of damage occurring.

We can’t change the mass of the bird or the aircraft, but what we can change is our acceleration. Or, to put it even more simply, slow down!

Not only will this minimize the risk of severe damage, but it also gives the pilots more time to ‘see and avoid’ (and quite possibly gives the birds the same luxury too!)

There have been studies conducted by major aircraft manufacturers that suggest that weather radar can effectively ‘scare’ birds away. However, the jury is still out, and bird strikes still occur on aircraft with active weather radar.

Ever flown on a jet with white painted spiral spinners inside the engine? There are a couple of reasons for this. First, it allows people to see that the engine is turning…

But, there is another reason…

The theory is that when viewed from afar , the engines look like eyes, which the birds interpret as predators, and steer clear as a result!

Bird strikes at night can often come from birds being entirely unaware of the presence of an aircraft.

There is an obvious answer.

Make yourself visible.

The best way to do this at night is by using the aircraft lights. Some aircraft manufacturers actually recommend this as part of their standard operating procedures .

Here is a top tip for avoiding bird strikes if flying in a light aircraft.

A pilot’s first temptation is often to ‘push the nose down to get away from the impending threat.

However, it is a bird’s natural behavior to also ‘dive’ away from predators. If you encounter either a single bird or flock, it is far better to try and climb over them to avoid a collision.

While bird strikes are not uncommon, the statistics prove that they aren’t normally dangerous. With an increase in traffic, the likelihood of bird strikes may go up in the future. Thankfully airplanes are built to withstand bird strikes. And, when combined with other risk mitigation strategies, they shouldn’t be something to be unduly concerned about.

Similar Articles

Roadblocks-to-Becoming-a-Private-Pilot

Top 10 Roadblocks to Becoming a Private Pilot

How-Much-Do-Airplanes-Weigh

How Much Do Airplanes Weigh?

Turns-Around-a-Point

How to Fly Turns Around a Point Like a Pro

write a term paper on bird strike

Scored % on their FAA Exam

Bird Strikes in Aviation: Causes and Results Essay

Introduction, how many species of birds exist, what frequencies do birds hear, the species of birds altitude, current technology that deters birds in airports and planes, list of references.

A bird strike happens when an aircraft in a flight and a bird collides or when landing or takeoff. Bird Strike is prevalent, and it can cause substantial hazards to aircraft safety. For instance, substantial damage may be caused in a small aircraft, particularly in jet-engine aircraft vulnerable to thrust loss. It causes bird ingestion in the engine air intake. This has been the cause of various fatal accidents. Since large bird numbers in a particular flight are at reduced ranks, bird strikes are more likely during the stages of takeoff, approach, landing, and initial climb. Since the majority of birds fly in the daytime, the majority of birds associated strikes happen in the daytime. Consequently, the following research aims to assess the bird species globally, the birds hearing frequencies, the species of birds’ altitude, and current technology that deters birds in airports and planes. The figures below indicate the images of birds striking.

Figure 1

Birds are widely regarded as a well-studied cohort, with more than 95 percent of their global species composition approximated to have been explained. According to most scorecards used by bird watchers and scientists, there are approximately 9,000 to10, 000 birds species (Van Gasteren et al., 2019, p. 907). However, those figures are based on the “biological species concept,” which characterizes species based on what creatures can breed together. Another study conducted by Princeton University professionals found approximately 9,700 bird species globally, though this figure is debatable (Miłek and Blicharz-Domańska, 2018, p. 251). According to BirdLife.org, the world is home to nearly 11,000 different bird species (Miłek and Blicharz-Domańska, 2018, p. 251). Even though the precise number of bird species in the world varies, nearly all projections are in the 10,000-species range. So, it is safe to say that there are approximately 10,000 bird species on the planet.

Birds are a hazard to aviation and prey on a variety of crops. Even though a lethal monitor is required in many circumstances, it is normally preferable to use nonlethal methodologies to dissipate birds from specific environments for several reasons. Sound is one type of deterrent dispersal methodology (Heffner et al., 2020, p. 901). To be effective, the noises must be loud enough for birds to hear, within the frequency spectrum that the birds’ earlobes can identify, and include a biologically relevant signal that induces the birds to escape.

This makes it necessary to consider the frequency at which birds hear sounds. Hearing is number two only to vision in controlling their surroundings for birds. Avian hearing is most susceptible to sounds between 1 and 4 kHz, even if they can hear increasing and decreasing frequencies (Heffner et al., 2020, p. 902). Moreover, no bird species has demonstrated responsiveness to ultrasonic frequencies. Although responsiveness to frequencies below 20 Hz (sound) has attracted great attention, bird species have demonstrated physiological (Zeyl et al., 2020) and behavioral reactions to these low frequencies. Within the 1 – 4 kHz range, the frequency of institutional racism in birds is only around half or one-third as effective as in human beings (Zeyl et al., 2020, p. 1037). Destruction of the auditory receptors caused by loud noises is an issue that birds face as humans do.

The quantity of destruction and the intensity of the sound created varies based on the species. Birds living in airport populated places may be regularly exposed to sound thresholds that affect their hearing. Consequently, auditory alerts must be at frequency bands detectable by the damaged auditory receptors to disperse birds using sound accurately (Zeyl et al., 2020). Even though some, if not all, bird species can revolutionize hair cells, sustained exposure to loud noises would result in hearing loss.

Many bird species are located in natural environments. These environments are greater than 13,123 ft above sea level, while others regularly fly to heights of 10,000-13,000 feet which is 3,000-4,000 meters, particularly when relocating (Conkling et al., 2018). Scott wrote that a multitude of bird species navigate at even high elevations (Conkling et al., 2018). Scott is interested in how vertebrates operate in physically demanding surroundings (Conkling et al., 2018, p. 1171). Scott emphasizes that birds such as hummingbirds and sparrows exist in heights of 16,404 ft (5,000 m) in the Alpine region (Conkling et al., 2018, p. 1171). Other birds, such as massive Andean condors, are found at 18,044 feet which is 5,500 m. Mallard ducks have been observed at heights of 21,000 ft., which is 6,401 m. Moreover, the bar-headed geese in Central Asia are explicitly monitored at 23,917 ft (Conkling et al., 2018).

Such high-flyers are capable of exerting themselves at incredible heights. While the size of these birds varies, they all have one commonality: they have a larger wingspan comparative to their bodies when contrasted to birds that fly lower. However, as Scott points out, high-altitude flight necessitates more than just longer wings, which present immense physical challenges. “The first great hurdle is that the air becomes less intense” (Conkling et al., 2018, p. 1172).

They must flutter as they fly greater. Hence, they struggle to stay skyward, which increases their metabolism requirements. The availability of oxygen (O 2 ) becomes confined and gets colder at greater heights, and birds have to warm them. Moreover, as the air becomes dry, they are more prone to losing water through respiration and water loss, leaving them thirsty. Charles Bishop asserts that physiological modifications undoubtedly permit birds to achieve extraordinary altitudes (Biteman et al., 2018). Bishop, a strong bar-headed bird’s research associate, told Reporters via email indicating that the geese birds do not struggle from heights sickness. These birds hyperventilate while piloting to continue increasing their oxygen consumption. Because of their rapid breathing, their blood becomes more basic, influencing blood flow to the brain in humans (Biteman et al., 2018, p.28). Consequently, this is why hyperventilating causes most individuals to faint or be dizzy.

However, geese are very impervious to alkaline environments or high PH states. as a result, blood flow to the creatures’ biological makeup remains normal, as the Bishop suggests. “Finally,” Bishop told Live Science, “the hemoglobin in their blood has a relatively high affinity for oxygen binding” (El-Sayed, 2019, p.109).

All through their long waves of migration, ranging from 1,243 to 3,107 miles which are 2,000 to 5,000Km, and last 5 to 200 hours, bar-headed geese use a “roller-coaster strategy” (El‐Sayed, 2019, P 108). Observations made directly of the geese’s height happened at a rate of 98% below 18,044 ft. “Whenever the geese had to fly over a high obstacle, they would immediately come down” (El-Sayed, 2019, p. 109). Furthermore, as per Scott, flying higher may give birds fair prospects for long-haul flights. Migration patterns flights at higher elevations introduce the living creatures to fewer predatory animals, while wind conditions can help the birds fly more efficiently, and cold weather can keep the living creatures from excessive heat.

When birds are prevalent in an area of aircraft action, among the most popular techniques of scaring them off-airport land is to fire air cannons. Nevertheless airports regularly alter the neighboring surroundings to make it less bird-friendly or substituting grass with crushed rock (Biteman et al., 2018, p. 28). Some airports are compelled to get innovative to keep birds at bay. This suggests that Salt Lake City’s airport uses pigs to feed gull eggs (Arrondo et al., 2021 p. 31). Similarly, border collies chase away egrets and herons at Southwest Florida International.

A bird strike occurs when an aircraft in flight collides with a bird or when landing or taking off. Bird strikes are common, posing significant risks to aircraft safety. Because of most birds in flight at subordinates, bird strikes seem to be more probable during the departure airport, preliminary climb, attitude, and arrival stages. Birds are widely regarded as a well-studied group, with more than 95 percent of their global species composition roughly explained. Birds threaten aviation and prey on a wide range of crops. Although lethal control is required in many situations, it is frequently preferable to use nonlethal methods to disperse or deter birds from specific locations for various reasons.

Most of these species live in natural environments that are more than 13,123 ft (4,000 meters) above sea level. Others normally tend to fly heights 10,000-13,000 ft when relocating (Miłek and Blicharz-Domańska, 2018, p. 252). When birds are abundant in an area of aircraft activity, one of the most common methods of scaring them off-airport land is to fire air cannons, but airports also frequently alter the surrounding environment to make it less bird-friendly, such as filling in ponds or replacing grass with crushed rock. In the appearance of bird exercise, postpone takeoff or touching down. The plane should keep the speed below 10,000 feet to less than 250 tangles if feasible (Biteman et al., 2018, 29). Afterward, they should ascend at the fastest possible rate below 2,000 feet to reduce flight time vulnerability to a potential attack danger.

Arrondo, E., García-Alfonso, M., Blas, J., Cortes-Avizanda, A., De la Riva, M., Devault, T., Fiedler, W., Flack, A., Jimenez, J., Lambertucci, S., Margalida, A., Oliva-Vidal, P., Phipps, L., Sanchez-Zapata, J., Wikelski, M. and Donazar, J. 2021). Use of avian GPS tracking to mitigate human fatalities from bird strikes caused by large soaring birds. Journal of Applied Ecology . Volume 2, pp.31-34

Biteman, D., Collins, D. and Washburn, B., 2018. Sunshine, Beaches, and Birds: Managing Raptor-Aircraft collisions at airports in Southern California. Proceedings of the Vertebrate Pest Conference , 28. pp. 28-35.

Conkling, T., Belant, J., DeVault, T. and Martin, J., 2018. Impacts of biomass production at civil airports on grassland bird conservation and aviation strike risk. Ecological Applications , 28(5), pp. 1168-1181.

El-Sayed, A., 2022. Bird Strike in Aviation . Zagazig University Press. 25(3), 98-111.

Heffner, R., Cumming, J., Koay, G. and Heffner, H., 2020. Hearing in Indian peafowl (Pavo cristatus): sensitivity to infrasound. Journal of Comparative Physiology A , 206(6), pp. 899-906.

Miłek, J. and Blicharz-Domańska, K., 2018. Coronaviruses in avian species – review with focus on epidemiology and diagnosis in wild birds. Journal of Veterinary Research , 62(3), pp. 249-255.

Van Gasteren, H., Krijgsveld, K., Klauke, N., Leshem, Y., Metz, I., Skakuj, M., Sorbi, S., Schekler, I. and Shamoun‐Baranes, J., 2019. Aeroecology meets aviation safety: early warning systems in Europe and the Middle East prevent collisions between birds and aircraft. Ecography , 42(5), pp. 899-911.

Zeyl, J., Ouden, O., Köppl, C., Assink, J., Christensen-Dalsgaard, J., Patrick, S. and Clusella‐Trullas, S., 2020. Infrasonic hearing in birds: a review of audiometry and hypothesized structure–function relationships. Biological Reviews , 95(4), pp. 1036-1054.

  • Chicago (A-D)
  • Chicago (N-B)

IvyPanda. (2023, January 1). Bird Strikes in Aviation: Causes and Results. https://ivypanda.com/essays/bird-strikes-in-aviation-causes-and-results/

"Bird Strikes in Aviation: Causes and Results." IvyPanda , 1 Jan. 2023, ivypanda.com/essays/bird-strikes-in-aviation-causes-and-results/.

IvyPanda . (2023) 'Bird Strikes in Aviation: Causes and Results'. 1 January.

IvyPanda . 2023. "Bird Strikes in Aviation: Causes and Results." January 1, 2023. https://ivypanda.com/essays/bird-strikes-in-aviation-causes-and-results/.

1. IvyPanda . "Bird Strikes in Aviation: Causes and Results." January 1, 2023. https://ivypanda.com/essays/bird-strikes-in-aviation-causes-and-results/.

Bibliography

IvyPanda . "Bird Strikes in Aviation: Causes and Results." January 1, 2023. https://ivypanda.com/essays/bird-strikes-in-aviation-causes-and-results/.

  • About Canada Goose Inc
  • International Success and Failures of Canada Goose Inc
  • Canada Goose Inc.'s South Korean Opportunity
  • Microbial Interactions in the Gut of Canadian Geese
  • An Essential Chemical Process: Fischer Tropsch (FT) Catalysis
  • Free Trade: An Ambiguous Phenomenon
  • Neighbours Noises Problem and Solutions
  • Marty’s Martini Bar: Grey Goose Inventory. Sale Challenge
  • Aviation Weather: Atmospheric Pressure
  • Canada Goose Inc. in the South Korean Market
  • Examining The Boeing 737 MAX Case Study
  • Air Cargo Planning Approach and Process
  • Closing the Gaps in Air Cargo Security
  • Knowledge Gained on the Air Cargo Industry
  • Reliever Airports in the Air Transport System

IMAGES

  1. Bird Strike Report Form

    write a term paper on bird strike

  2. Best Tips on How to Write a Term Paper: Outline, Format, Example

    write a term paper on bird strike

  3. Sample Of A Term Paper / When asked to write an essay, a term paper, or

    write a term paper on bird strike

  4. Handy Guide on How to Write a Term Paper

    write a term paper on bird strike

  5. Essay on Bird Within 10 Lines for Class 1,2,3,4,5 Kids

    write a term paper on bird strike

  6. (PDF) Bird strike: prevention and proofing

    write a term paper on bird strike

VIDEO

  1. Special paper bird 🕊️ #paperbird #short

  2. Write an Essay on the Birds in English || Essay Writing || Short essays//zima learning corner//Birds

  3. bird strike

  4. Facts about strike Bird

  5. What happens if a bird strikes an airplane engine #fact #interestingfacts #shorts

  6. 'Paper Bird' #aianimation #aiexperiments #paperquilling

COMMENTS

  1. (PDF) Bird strike: prevention and proofing

    Abstract. The bird-strike statistics of the USA and the world, as presented in Chapter 2, demonstrated the large number of injuries and fatalities imposed upon passengers and aircraft crew, as ...

  2. Three novel bird strike likelihood modelling techniques: The case of

    In Australia alone, 17,336 bird-strikes occurrences and 401 animal-strike incidents were reported to the Australian Transport Safety Bureau between 2010 and 2019 and 17,360 strikes reported to the US Federal Aviation Authority (FAA) in a single year (2019) . Wildlife strikes can have catastrophic consequences.

  3. (PDF) The Bird Strike Challenge

    This paper focuses on collisions involving birds and the term bird strike is used. First, the vast majority of wildlife strikes occur with birds, for example: 98% in Australia, 95% in Canada and 95%

  4. Three novel bird strike likelihood modelling techniques: The case of

    The risk posed by wildlife to air transportation is of great concern worldwide. In Australia alone, 17,336 bird-strike incidents and 401 animal-strike incidents were reported to the Air Transport Safety Board (ATSB) in the period 2010-2019. Moreover, when collisions do occur, the impact can be catastrophic (loss of life, loss of aircraft) and involve significant cost to the affected airline ...

  5. A birdstrike risk assessment model and its application at ...

    Here, we propose and validate a new model for assessing birdstrike risk in order to fill that need. The model consists of two elements. First, empirical data are collected on the occurrence of ...

  6. PDF The Bird Strike Challenge

    the bird strike statistics [2,6,7,10]. This paper focuses on collisions involving birds and the term bird strike is used. First, the vast majority of wildlife strikes occur with birds, for example: 98% in Australia, 95% in Canada and 95% in the USA [2,7,10]. Second, terrestrial animals can be prevented from entering airport perimeters,

  7. Bird Strike: An Experimental, Theoretical and Numerical Investigation

    Bird strikes occur in every 2000 flights. 90% of foreign body damage in aviation is caused by bird strikes. In the event of a bird strike, the most critical parts of the aircraft are the nose ...

  8. PDF A birdstrike risk assessment model and its application at ...

    Here, we propose and validate a new model for assessing birdstrike risk in order to fill that need. The model consists of two elements. First, empirical data are collected on the occurrence of ...

  9. PDF Bird Strike Analyses on The Parts of Aircraft Structure

    The submitted paper describes using air gun with smooth borehole for proof of bird strike resistance of aircraft parts. The finite element code ABAQUS Explicit was mainly employed in the study of sensitivity on the simulations to material properties, mesh characteristics, and impactor models. 1 Introduction

  10. Bird Strike

    Abstract. In this chapter, the term "bird strike" is defined and its effects - as the most important foreign object damage (FOD) are presented. The most important bird strike events in history are also reviewed. The importance of bird strike as a threat to both the international aviation industry and its passengers, which is the ...

  11. Analysis of Risk-Based Operational Bird Strike Prevention

    Bird strike prevention in civil aviation has traditionally focused on the airport perimeter. Since the risk of especially damaging bird strikes outside the airport boundaries is rising, this paper investigates the safety potential of operational bird strike prevention involving pilots and controllers. In such a concept, controllers would be equipped with a bird strike advisory system, allowing ...

  12. An investigation of bird strike cases in the aviation sector with a

    Bird strike Aviation Principal-agent problem Aviation insurance 1. Introduction 1.1. Motivating factor behind the research Markets advance with news and information. Standard pricing theory assumes that all participants in the market have the same information. However, in reality, market actors have different information.

  13. Bird Strike

    A bird strike is strictly defined as a collision between a bird and an aircraft which is in flight or on a take off or landing roll. The term is often expanded to cover other wildlife strikes - with bats or ground animals. Bird Strike is common and can be a significant threat to aircraft safety.

  14. (PDF) Bird Strike to Aircrafts An Assessment of Changing Bird

    A basic tenet of programs to mitigate the risks of bird strikes with aircraft has been to focus management efforts at airports because various historical analyses of bird-strike data for civil aviation have indicated the majority of strikes occur in this environment during take-off and landing at 500 feet AGL increased signifi cantly from about ...

  15. Bird strike

    A bird strike (sometimes called birdstrike, bird ingestion (for an engine), bird hit, or bird aircraft strike hazard ( BASH )) is a collision between an airborne animal (usually a bird or bat) [1] and a moving vehicle (usually an aircraft ). The term is also used for bird deaths resulting from collisions with structures, such as power lines ...

  16. Bird Strike, what is It, How Common, How Dangerous?

    Windscreens. This is the biggest threat from bird strikes. Airplanes fly at high speeds. A bird flying one direction at 30mph and an airplane flying in the other at 130 mph results in a relative collision speed of around 160mph. Glass really isn't designed to hold up too well to collisions at this speed. If a bird penetrates the cockpit at ...

  17. Review of current research trends in bird strike and hail impact

    These methods and models are useful to analyse airworthiness requirements for damage tolerance regarding bird-strike and hail impact and haves been subjected to critical review in this paper.

  18. Bird Strike on Final Approach: Guidance for Flight Crews

    On 31 July 2012, a Boeing 737-900 struck a single large bird whilst descending to land at Denver in day VMC and passing approximately 6000 feet aal, sustaining damage to the radome, one pitot head and the vertical stabiliser. The flight crew declared an emergency and continued the approach with ATC assistance to an uneventful landing.

  19. PDF The Bird Strike Challenge

    the bird strike statistics [2,6,7,10]. This paper focuses on collisions involving birds and the term bird strike is used. First, the vast majority of wildlife strikes occur with birds, for example: 98% in Australia, 95% in Canada and 95% in the USA [2,7,10]. Second, terrestrial animals can be prevented from entering airport perimeters,

  20. PDF International Birdstrike Committee

    Confirmed strikes: Any reported collision between a bird or other wildlife and an aircraft for which evidence in the form of a carcass, remains or damage to the aircraft is found. Any bird/wildlife found dead on an airfield where there is no other obvious cause of death (e.g. struck by a car, flew into a window etc.).

  21. Bird Strikes in Aviation: Causes and Results Essay

    Conclusion. A bird strike occurs when an aircraft in flight collides with a bird or when landing or taking off. Bird strikes are common, posing significant risks to aircraft safety. Because of most birds in flight at subordinates, bird strikes seem to be more probable during the departure airport, preliminary climb, attitude, and arrival stages.

  22. Description Of Bird Strike Engineering Essay

    "A bird strike (sometimes bird strike, bird hit, or BASH - Bird Aircraft Strike Hazard) is a collision between an airborne animal (usually a bird) and a man-made vehicle, especially aircraft. It is a common threat to flight safety, and has caused a number of accidents with human casualties.

  23. Dangers and Controls of Bird-Strikes

    1.Who has the right to fly? Bird or man? God has created the bird to fly in the sky, whereas man is not designed to fly. But man has conquered the sky and now sharing the sky with the bird. Therefore, we can say that it is our responsibility to avoid the collision with the bird and ensure the safe living of the bird species.