Spatial Analysis and Prediction of Animal Vehicle Collisions (AVCs) in Iowa Using Environmental and Anthropogenic Factors

Spatial Analysis and Prediction of Animal Vehicle Collisions (AVCs) in Iowa Using Environmental and Anthropogenic Factors | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Spatial Analysis and Prediction of Animal Vehicle Collisions (AVCs) in Iowa Using Environmental and Anthropogenic Factors Xi Wei, Sudesh Bhagat, Aparna Joshi, Anuj Sharma This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8398693/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Animal–vehicle collisions (AVCs) present serious safety, economic, and ecological concerns, particularly in areas where wildlife habitats intersect with expanding road networks. While previous studies have examined land cover or traffic volume in relation to AVCs, many relied on coarse spatial scales, treated species as a single group, and omitted key roadway or operational variables. This study addresses these gaps by investigating the spatial determinants of terrestrial AVCs across Iowa using an integrated ecological and infrastructural modeling framework. Crash data from 2020 to 2024 were analyzed using a natural language processing (NLP) model to extract animal-related incidents and infer species types from narrative reports. Land cover, road characteristics (e.g., classification, AADT, speed limits), operational measures (e.g., incident clearance time), and population data were aggregated to Iowa’s 896 census tracts (2020 boundaries). A Negative Binomial (NB) regression model identified forest coverage, population density, speed limits, and road network complexity (e.g., length of Interstates and arterials) as significant predictors of higher AVC risk, while urban land cover was negatively associated. Stepwise AIC selection and 10-fold cross-validation enhanced model performance. Spatial clustering in residuals (local Moran’s I = 0.42, p < 0.01) led to a Spatial Error Model (SEM), which further improved fit and reduced spatial bias. This tract-level framework offers a scalable, data-driven approach for transportation and wildlife agencies to identify high-risk areas and implement localized mitigation strategies, such as fencing, signage, or speed control, tailored to both ecological and infrastructural conditions. Civil Engineering Artificial Intelligence and Machine Learning Animal-vehicle collisions wildlife mitigation spatial analysis Natural language processing Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 Figure 11 Figure 12 INTRODUCTION Each year, more than five million vehicle crashes occur across the United States, resulting in nearly three million injuries and approximately 40,000 fatalities ( 1 ). While public attention often centers on crashes caused by speeding, distracted driving, or weather, a less visible yet significant threat unfolds daily on rural roads: AVCs (animal-vehicle collisions). These crashes account for roughly 26,000 injuries, 200 deaths, and over $ 8 billion in damage annually in the United States ( 1 ). Yet despite their magnitude, AVCs remain overlooked in mainstream traffic safety discourse. In a state like Iowa, the risk of AVCs is especially pronounced. The landscape, comprising farmland, woodlands, streams, and ravines, supports a diverse range of wildlife while also being heavily intersected by road networks. Areas where dense vegetation or natural habitats lie close to roadways become common crossing points for animals such as deer. As a result, Iowa experiences 8,000 animal-related crashes each year, over 14 percent of the state’s total crash count ( 2 ). AVCs tend to follow spatial and seasonal patterns influenced by both environmental and human-related factors ( 3 )( 4 ). Because AVCs are spatially clustered ( 5 ), driven by both ecological corridors ( 6 ) and infrastructure design ( 7 ), statistical models must account for both crash frequency and spatial dependence. Prior research has examined these relationships using variables such as land cover ( 8 ), traffic volume ( 9 ), and speed limits ( 10 ), often at coarse spatial scales ( 11 ). However, these studies rarely consider operational and roadway characteristics that influence crash occurrence, crash severity, and post-crash recovery. These include incident clearance time, road clearance time, and surface width (lane width), which are key variables that directly affect emergency response effectiveness, secondary crash risk, and roadway hazard duration. Additionally, average annual daily traffic (AADT) and posted speed limits play critical roles in shaping driver response time and stopping distance but remain inconsistently integrated across ecological AVC models. This study addresses existing research gaps by integrating a comprehensive set of environmental, anthropogenic, and transportation-related variables into a spatial modeling framework at the census tract level across Iowa. We aggregate diverse datasets, including ecological land cover types such as needleleaf, broadleaf, mixed forest, wetlands, cropland, shrubland, grassland, and urban areas, as well as land-use designations associated with high animal activity such as Wildlife Refuges, Nature Preserves, City Parks, Greenbelts, and Trail systems to the tract scale. From the transportation domain, we include road network characteristics such as Annual Average Daily Traffic (AADT), speed limits, and surface widths, alongside crash-level operational variables such as incident and road clearance times. To evaluate the influence of these factors on AVC frequencies, we employ a negative binomial regression model, a widely used method for analyzing overdispersed crash count data. This tract-level modeling approach enables spatial resolution, which has typically relied on coarser spatial scales such as county-level data ( 10 )( 11 ), allowing for more localized insights into AVC risk factors. Furthermore, we introduce a novel layer of species-specific analysis by applying a natural language processing (NLP) model to infer the animal species involved in each crash. This allows us to move beyond treating AVCs as a homogeneous category and instead examine which species are most frequently affected and under what conditions. By integrating environmental context, roadway design, traffic operations, and species-level identification, this study provides a more holistic and actionable framework for detecting collision hotspots and improving mitigation strategies in wildlife-prone regions. LITERATURE REVIEW AVCs remain a persistent challenge at the human–wildlife interface. Research in this area has progressed through distinct phases, uncovering the multidimensional nature of these events. Early ecological studies established that land cover patterns are strongly associated with AVC risk. Hubbard et al. identified crash hotspots near habitat features such as forests, rivers, and wetlands, which facilitate wildlife movement across roadways and increase collision risk ( 12 ). A study by Galinskaitė et al. confirmed the role of wetlands and agricultural edges in attracting wildlife across varying biogeographic regions ( 13 ). These spatial ecological studies provided a foundation for predictive modeling but were limited in their analytical scope. Methodological advances soon followed to address the statistical complexity of AVC data. Lord and Mannering demonstrated that traditional Poisson models often fail to capture the overdispersed and rare nature of AVCs and introduced alternative approaches such as diagonal-inflated Poisson and Negative Binomial (NB) models, with the latter showing improved model fit ( 14 ). Technological developments have further enhanced AVC monitoring and analysis. Gunson et al. described how crowdsourcing platforms and automated tools, including lidar and thermal sensors, enable real-time wildlife detection and crash reporting ( 15 ). Geographic information systems (GIS) have also expanded spatial modeling capabilities, allowing for visualization of collision clusters across time and diverse landscape features. Despite ongoing advances, critical knowledge gaps remain. Many studies group all wildlife species together, overlooking biological differences in movement and behavior. For example, Huijser et al. found that treating deer, rodents, and reptiles as a single group ignores key differences, such as movement speed and avoidance ability, that influence the likelihood of a collision ( 16 ). Another limitation is that ecological models often leave out transportation infrastructure variables that influence crash risk. Bissonette and Kassar demonstrated that road type alone, from high-speed interstates to low-volume collectors, can significantly affect collision probability, regardless of habitat conditions ( 17 ). However, some studies continue to focus solely on habitat and land cover features while omitting traffic or road design information. For instance, Arévalo et al. examined patterns of amphibian road mortality near a national park in Costa Rica using ecological factors, but without accounting for road classification, traffic volume, or speed ( 18 ). This lack of integration between ecological and transportation data may lead to blind spots in high-risk areas, such as farmlands near protected lands, where animals frequently encounter fast-moving traffic. In addition to ecological gaps, operational characteristics, such as incident clearance time and road clearance time, remain largely absent from AVC studies. For example, Arévalo et al. and Galinskaitė et al. did not incorporate post-crash response or clearance duration variables into their spatial analyses, despite these factors being widely used in broader traffic safety research ( 18 ). Haule et al. demonstrated that such variables are critical for understanding crash severity, emergency response efficiency, and overall system resilience ( 19 ). Including these operational measures in AVC models may provide insights into both crash causation and exposure duration. Roadway geometry, such as surface width, shoulder presence, or lane configuration, —is also underrepresented in most AVC studies, even though these design features can influence both animal crossing behavior and driver avoidance maneuvers ( 15 ). While traffic volume (AADT) and speed limits are sometimes included, they are often treated as categorical or control variables, rather than continuous predictors capable of explaining collision frequency or species-specific vulnerability ( 20 ). Technological countermeasures have demonstrated potential in reducing AVC, but their implementation remains limited. Automated wildlife warning systems, for example, have been reported to reduce collisions by 30–50% under controlled conditions ( 21 ). However, these systems are typically species-insensitive and rarely integrate key roadway engineering variables such as road classification, traffic volume, or speed, factors known to influence collision likelihood. As a result, their effectiveness in real-world settings remains constrained. Furthermore, there is no comprehensive framework that explains why wildlife select specific road segments for crossing. The continued disconnect between ecological drivers (e.g., habitat suitability and migratory behavior) and anthropogenic risk factors (e.g., roadway infrastructure and human disturbance) hampers efforts to develop effective and targeted mitigation strategies. Land-use designations, such as wildlife refuges, city parks, nature preserves, and greenbelt trails, often act as attractants or movement corridors for wildlife. Despite their ecological importance, these areas are inconsistently represented in AVC prediction models. Their presence can increase the likelihood of animal crossings and elevate collision risk on adjacent road segments, particularly during migration periods or seasonal peaks in animal activity ( 22 ). As terrestrial species accounted for 99.7% of serious collisions in the dataset, birds and other flying species were excluded due to fundamental differences in their collision dynamics. This allowed for a focused analysis of ground-level interactions where vehicle trajectories intersect with terrestrial wildlife movement. To enable species-specific risk analysis, a machine learning model was used to classify the animal species involved in each crash, an approach that has not previously been applied at this scale. This innovation helps overcome the long-standing “species blind spot” noted by Litvaitis and Tash, who emphasized the risks of treating wildlife as a homogenous group in collision analyses ( 23 ). The core theoretical contribution of this study lies in its integrative analytical framework. We combined environmental predictors (e.g., forest cover, wetland density, cropland interface, and wildlife refuge buffers) with anthropogenic variables (e.g., tract-level population, AADT, speed limits, and road classification). Additionally, we incorporated geometric and operational characteristics, such as surface width, incident clearance time, road clearance time, and refined land-use categories. This multidimensional perspective revealed previously undocumented interactions: for example, wetlands near highway arterials reduce AVCs regardless of habitat quality, while agricultural edges adjacent to low-volume roads may act as ecological traps during harvest seasons. Wildlife corridors intersecting secondary arterials showed increased risk when speed limits exceeded average braking distance. Despite the growing use of flexible regression models, many AVCs studies continue to assume spatial independence of crash occurrences, an assumption often violated when nearby tracts share similar unobserved risks, such as localized patterns of human activity ( 24 ). Spatial econometric techniques, such as the Spatial Error Model (SEM), offer a robust approach to account for spatial dependence on model residuals. SEMs are particularly effective when omitted variables are spatially correlated, thereby biasing estimates if ignored. Prior traffic safety studies, including crash frequency models in Louisiana and Pennsylvania, have demonstrated that SEMs improve model fitness and reduce misspecification bias in segment-level crash data ( 25 ). Incorporating SEMs enhances model accuracy, enables detection of spatial clusters of error, and supports more reliable, spatially explicit predictions. Key Contributions of This Study: Developed a census tract–level modeling framework to capture localized AVC risk patterns with greater spatial precision. Integrated ecological, anthropogenic, geometric, and operational variables in a unified statistical framework. Applied a machine learning NLP pipeline to extract species-level crash data from narrative crash reports, addressing species-blind modeling limitations. Demonstrated the benefit of spatial error modeling in correcting residual autocorrelation and improving predictive accuracy. Provided a transferable methodology and practical insights to inform wildlife–vehicle conflict mitigation strategies across different regions and agencies. METHODS Data Sources and Processing This study integrates diverse spatial and tabular datasets to analyze terrestrial AVCs across Iowa at the U.S. Census tract level. The analysis covers 2020–2024, using 2020 census tract boundaries and population data as the spatial framework for geoprocessing and aggregation. Census tracts were selected for their finer granularity compared to counties and compatibility with available environmental, crash, and infrastructure datasets. Data formats included shapefiles (e.g., road network, public lands), raster grids (e.g., land cover), and tabular CSVs (e.g., crash records, population). All datasets were processed using ArcGIS Pro and R through coordinating harmonization, spatial join, raster reclassification, zonal statistics, and aggregation to census tracts. These steps ensured consistent spatial alignment and standardized variable construction. Detailed processing steps are described in the following subsections. Population and Human Activity To represent human activity, we used 2020 population data from the U.S. Census Bureau ( 26 ) and tract boundaries from the 2020 TIGER/Line shapefiles ( 27 ). Population counts were joined to census tract geometries using the unique GEOID identifier. The resulting tract-level population values were used as a continuous predictor in our models and served as a proxy for anthropogenic pressure and development density, one of the factors known to influence wildlife movement and AVC likelihood. These patterns are visualized in Fig. 1 , which displays population distribution across Iowa’s census tracts. Road Network and Traffic Data The road network data were obtained as a shapefile from the Iowa Department of Transportation (Iowa DOT) Road Network Portal ( 28 ). This dataset includes roadway geometries along with embedded attributes such as Federal Functional Class, Average Annual Daily Traffic (AADT), posted speed limit, surface width, and a unique Route ID for each segment. Roads were classified into six categories: Interstate, Other Principal Arterial, Minor Arterial, Major Collector, Minor Collector, and Local, reflecting differences in capacity and network function (Fig. 2 ). The shapefile was then spatially joined to 2020 Census tract boundaries, assigning each road segment to the tract it intersects. Tract-level traffic variables, including mean AADT and mean speed limit, were calculated to serve as proxies for traffic intensity and driver exposure. Crash Data AVC records from 2020 to 2024 were obtained from the State of Iowa Crash Data Base ( 29 ) and spatially joined to 2020 U.S. Census tracts using geographic coordinates. Each crash record included information such as crash date, latitude and longitude, cause, and post-crash clearance times. To narrow the focus of our analysis, we excluded non-terrestrial species and retained only terrestrial AVCs, which represented 99.7% of all AVCs during this period. The total count of terrestrial AVCs per tract served as the response variable in our modeling framework. To support spatial interpretation and communication, we generated several visualizations. Figure 3 displays point locations of AVCs across Iowa from 2020 to 2024. Figure 3 presents tract-level AVC counts, classified using the natural breaks (Jenks) method to highlight meaningful variation in collision frequency. Table 1 provides a sample of the raw crash dataset, including key variables such as Route ID, geographic coordinates, crash time, cause, and both roadway and incident clearance times, which were used as operational predictors in the model. Table 1 Raw Crash data sample Crash Key Route ID Latitude Longitude Crash Date Major Cause Total Roadway Clearance Time Total Incident Clearance Time 20201154658 C000343420N 43.30339 -91.22849 1/4/2020 14:10 Animal 000:52 000:52 20201154717 S001920030E 42.00765 -93.38498 1/4/2020 17:25 Animal 000:11 000:35 20201154720 S001910080W 41.67961 -91.46715 1/5/2020 4:41 Animal 000:49 000:19 Land Cover Data The base land cover dataset consisted of multiple natural and human-modified land types, including needleleaf forest, broadleaf forest, mixed forest, shrubland, grassland, wetland, cropland, urban and built-up areas, barren land, and water. These classes were derived from the 2020 North American Land Cover 30m raster dataset provided by the Commission for Environmental Cooperation (CEC) ( 30 ). To simplify model input while preserving ecological relevance, the original land cover types were reclassified into six broader categories: Forest, Shrubland/Grassland, Wetland, Cropland, Urban/Built-up, and Water (Fig. 4 ). This generalization helped reduce classification noise from overly specific vegetation subtypes and ensured alignment with wildlife habitat associations reported in prior literature ( 12 , 13 , 16 ). Using spatial overlays and zonal statistics, the proportional area of each reclassified land cover type was calculated for every census tract. The resulting tract-level variables (e.g., forest proportion, wetland proportion) were used as key environmental predictors in the modeling framework. Public Land Use Reclassification To better represent wildlife-attractive environments, we reclassified the public land dataset provided by Iowa DOT ( 31 ) based on the designation and intended ecological use of each public land unit. Each unit was defined as a spatial polygon representing a distinct area of publicly managed land and was evaluated for its potential to support or attract wildlife. The units were grouped and weighed as follows: High relevance (weight = 1.0): Includes Wildlife Refuges, Nature Preserves, National Wildlife Refuges, and State Wildlife Refuges. Moderate relevance (weight = 0.5): Includes City Parks, Greenbelts, and Trail Systems. Low or no relevance (weight = 0): Includes public lands not directly associated with habitat or recreational use. Each polygon's area (in acres) was multiplied by its assigned weight. These weighted areas were then summed at the census tract level to compute a continuous variable: Wildlife-Related Area Acres. To account for tract size and enable cross-tract comparisons, this value was then divided by the total area of each tract to produce the final modeling variable: Wildlife-Related Area Proportion shown as Eq. 1. Wildlife-Related Area Proportion i = \(\:\frac{{\sum\:}_{j=1}^{n}\text{A}\text{ᵢ}\text{ⱼ}\:\times\:\:\text{w}\text{ⱼ}}{\text{T}\text{o}\text{t}\text{a}\text{l}\:\text{T}\text{r}\text{a}\text{c}\text{t}\:\text{A}\text{r}\text{e}\text{a}\text{ᵢ}}\) ( 1 ) Where: A i ⱼ = area (in acres) of public land unit j in tract i wⱼ = assigned weight based on land use relevance (1.0, 0.5, or 0) Total Tract Area i = total land area of tract i Modeling Approach NLP Modeling To identify and classify animal-related crashes, structured crash data was merged with their corresponding narrative descriptions. A natural language processing (NLP) pipeline was developed using a large language model (LLM) via the Ollama framework to extract relevant information from these unstructured texts ( 32 ). Narratives were preprocessed by converting date fields and removing records with missing or non-informative content. The primary objective was twofold: 1 is to determine whether an animal was involved in a crash, and 2 if so, to identify the specific animal type mentioned. The objective was not to identify additional animal crashes. The classification was performed using the LLaMA 3 (8B) model, a decoder-only transformer architecture released by Meta in April 2024 ( 33 ). With approximately 8 billion parameters, this model incorporates Grouped Query Attention, rotary position embeddings, and a 128k-token vocabulary, enabling efficient and accurate long-context reasoning. A classification accuracy of 90% was achieved. The model was hosted locally through Ollama and executed using a standardized prompt: "You are a crash narrative analysis assistant. Analyze the following narrative and answer these two questions: Is this crash related to an animal? Answer only 'yes' or 'no'. If yes, what kind of animal was involved? Answer with the animal type (e.g., deer, dog, cattle, bird, raccoon, etc.). If unknown or not applicable, say 'none'. Respond in this JSON format only: { 'animal_involved': 'yes' or 'no', 'animal_type': 'deer' or 'dog' or 'bird' or 'cattle' or 'raccoon' or 'other' or 'none' }." This constrained output format ensured consistency across thousands of crash narratives and minimized parsing errors. Exception handling was incorporated to manage incomplete or malformed responses. Once classification was complete, the results were merged back into the original crash dataset. From this enriched dataset, a filtered subset of animal-related crashes was generated. Figure 5 presents the distribution of the most frequently identified animal types. Table 2 Top 10 Animal Crash Count and Percentage Animal Type Count Percentage Deer 34658 95.56 Cattle 366 1.01 Dog 317 0.87 Raccoon 226 0.62 Cow 146 0.4 Coyote 137 0.38 Horse 92 0.25 Turkey 57 0.16 Calf 42 0.12 Fox 16 0.04 This structured classification allowed for subsequent research utilizing the negative binomial and geospatial visualization to discover where animal crashes are more likely to occur. Negative Binomial and Spatial Error Modeling To examine how environmental and anthropogenic factors influence the frequency of terrestrial AVCs, we developed a statistical modeling framework using census tract level data across Iowa. The dependent variable was the count of AVCs, which exhibited overdispersion where the variance exceeded the mean, a common feature of both crash and ecological data. We initially evaluated several generalized linear models including Poisson regression, Zero Inflated Poisson, and Negative Binomial regression. A dispersion test revealed significant overdispersion, making the Poisson model unsuitable. Although AVC counts included some zeros, Vuong’s test indicated that the Zero Inflated Poisson model did not provide a significantly better fit than the Negative Binomial model. Additionally, the Zero Inflated Poisson model assumes two separate data generating processes, one for excess zeros and one for count values, which was not theoretically justified in this context. Based on model diagnostics and fit statistics, including the Akaike Information Criterion and residual deviance, and consistent with prior literature, the Negative Binomial model was selected as the final specification. To improve model parsimony and predictive performance, we applied stepwise AIC selection in both forward and backward directions to identify the most informative subset of predictors. Variables were retained based on a combination of statistical significance and theoretical relevance. To assess residual spatial dependence after fitting the Negative Binomial model, we computed Moran’s I and conducted Local Indicators of Spatial Association analysis. These diagnostics revealed significant spatial clustering in the residuals, suggesting that unobserved spatial processes were influencing AVC outcomes. Based on these findings, we estimated a Spatial Error Model to account for latent spatial autocorrelation while retaining the structure of the selected covariates. The initial set of predictor variables included environmental factors such as the proportions of forest, wetland, shrubland and grassland, water, cropland, and wildlife related public lands, along with anthropogenic factors including log transformed total population, average speed limit, average annual daily traffic, and road lengths by functional classification. To improve interpretability and reduce skewness, the population variable was log transformed. Interaction terms between each land cover proportion and log transformed population were included to capture how human presence modifies the effect of land use on AVC risk. Rather than applying an offset for tract area, the model adjusted for variation in population and landscape composition through these transformed and interacted covariates. The final Negative Binomial specification included all land type variables interacted with population along with transportation related covariates. The Spatial Error Model was implemented in R using the spdep and spatialreg packages. A Queen contiguity based spatial weights matrix was constructed from tract geometries using poly2nb and nb2listw functions. The model was estimated using the errorsarlm function and compared with a spatial lag model using AIC values. Final predictions and residuals were joined back to the spatial data object for visualization and assessment. RESULTS The final Negative Binomial (NB) regression model identified several significant predictors of terrestrial AVCs frequency across Iowa census tracts (Table 3 ). The model exhibited a good overall fit (AIC = 6390.3; residual deviance = 847.21 on 790 df) and provided interpretable coefficients for both environmental and anthropogenic factors. A 10% increase in forest cover was associated with a 20–22% increase in AVC risk, consistent with the positive coefficient for forest proportion (β = 2.03, p < 0.001). This relationship reflects the ecological role of forests as high-use wildlife habitats and corridors for ungulate species such as deer, increasing frequency of crossing and collision exposure ( 13 , 15 ). Log-transformed population was also positively associated with AVC frequency (β = 0.196, p = 0.001), indicating that areas with greater human presence face elevated collision risk, likely due to more vehicle activity and habitat fragmentation. In contrast, urban built-up areas showed a strong negative association (β = − 1.947, p < 0.001), suggesting that highly developed regions with impervious surfaces and limited habitat are less frequented by terrestrial wildlife, thereby reducing collision risk. This is in line with findings by Bissonette and Kassar ( 17 ), who showed that urbanization acts as a movement barrier for wildlife. Transportation variables, including AADT, posted speed limit, and lengths of roads across all functional classes, were also positively associated with AVCs (p < 0.001), confirming that higher traffic exposure increases collision risk. These findings are consistent with past AVC studies that emphasize road speed and volume as critical collision predictors ( 15 , 16 ). Table 3 Final Negative Binomial regression model Variable Estimate Std. Error z value p-value (Intercept) 0.647 0.511 1.266 0.206 Forest_prop 2.03*** 0.2572 7.892 < 0.001 log_POP 0.196** 0.0598 3.281 0.001 UrbanBuiltup_prop -1.947*** 0.1049 -18.564 < 0.001 avg_AADT 1.71e-05*** 3.17e-06 5.402 < 0.001 avg_SPEED_LIMIT 0.01683*** 0.00323 5.216 < 0.001 Interstate 0.00833*** 0.00167 4.986 < 0.001 MinorArterial 0.01306*** 0.00172 7.579 < 0.001 MajorCollector 0.02004*** 0.00189 10.585 < 0.001 MinorCollector 0.01614*** 0.00478 3.377 < 0.001 Local 0.03384*** 0.00295 11.466 < 0.001 Despite the NB model’s reasonable predictive ability, as demonstrated by the 10-fold cross-validation plot of predicted versus actual AVC counts (Fig. 6 ), the residual plot (Fig. 7 ) revealed increasing underprediction at higher crash counts, suggesting unaccounted spatial structure. To assess spatial dependence in the NB model residuals, we first applied Moran’s I, which yielded a positive value of 0.136, indicating significant global spatial clustering (Fig. 8 ). Local Indicators of Spatial Association (LISA) analysis further identified distinct high–high and low–low residual clusters, particularly in eastern and southern Iowa (Fig. 9 ). The corresponding LISA significance map revealed that many of these clusters were statistically significant (p < 0.05), reinforcing the conclusion that unobserved spatial processes were influencing AVC outcomes (Fig. 10 ). Notably, several low–low clusters appeared in central Iowa, which includes more urbanized areas with high population density and AADT. These patterns suggest that despite greater human activity, such areas tend to have fewer AVCs, possibly due to lower wildlife presence or effective roadway barriers. In contrast, many high–high clusters were observed in rural regions, where natural habitats are abundant and less fragmented, leading to consistently elevated AVC frequencies across neighboring tracts. Additionally, high–low and low–high clusters may reflect localized anomalies such as hazardous road segments or mitigation infrastructure. To correct these spatial dependencies, we estimated a Spatial Error Model (SEM) using the same covariates. The SEM’s spatial autoregressive coefficient (λ = 0.211, p < 0.001) was statistically significant, and the model demonstrated improved fit with a lower AIC (7090.7) compared to the NB model (7105). While coefficient interpretation was not the primary focus of the SEM, it substantially improved predictive accuracy and reduced residual spatial bias. As shown in Fig. 11 , the SEM-generated predicted AVC intensity map revealed smoother spatial gradients across Iowa, particularly reducing overprediction in low-density rural areas and underprediction in high-density zones. The corresponding residual map (Fig. 12 ) illustrates a notable decrease in spatial clustering and overall residual magnitude compared to the NB model. This suggests that the SEM more effectively captured latent, spatially structured risk factors and improved model calibration at both ends of the prediction spectrum. CONCLUSION This study presents a comprehensive spatial analysis of terrestrial AVCs across Iowa from 2020 to 2024, integrating ecological, anthropogenic, geometric, and operational variables into a robust census tract level framework. A novel contribution of this research lies in its use of species-specific crash classification derived from narrative data and the explicit incorporation of operational characteristics, such as incident and road clearance time, seldom addressed in prior AVC studies. At the census tract level, the Negative Binomial (NB) regression identified several key factors associated with increased AVC frequency. Forested areas were linked to higher collision risk, while urban and built-up zones saw fewer crashes. Human-related factors such as greater population density, higher traffic volumes, and faster speed limits were also positively associated with AVCs. Additionally, a more complex road network including arterials, collectors, and local roads was tied to elevated crash frequencies. These results emphasize how both ecological settings and roadway infrastructure influence wildlife–vehicle conflict patterns. However, Local Moran’s I and LISA cluster maps revealed spatial clustering in residuals (Moran’s I = 0.136), indicating unmeasured local risk factors and the presence of spatial autocorrelation not fully explained by the NB model. To address this, a Spatial Error Model (SEM) was implemented, yielding a lower AIC (7090.7) compared to the NB model (7105) and producing more spatially consistent AVC predictions. Residual clustering was substantially reduced, confirming that the SEM effectively captured latent spatial dependencies and improved overall predictive accuracy. This study demonstrates several key strengths. It integrates narrative-derived species information, fine scale spatial modeling, and a diverse set of predictors, including static factors like land use and public lands, and dynamic ones like traffic volume and incident clearance times. The framework also leverages multiple advanced methods: a Negative Binomial model to address overdispersed crash counts, a Spatial Error Model (SEM) to account for spatial autocorrelation, and a natural language processing (NLP) pipeline to extract species information from crash narratives. To our knowledge, this is the first AVC study to combine NLP-based species inference with spatial econometric modeling at the census tract level. This integrated approach supports geographically tailored mitigation strategies and is scalable to other regions with census-based crash and infrastructure data, making it valuable for transportation planners, conservationists, and public agencies. The findings of this study offer important guidance for transportation planners, wildlife managers, and policymakers. By identifying how factors such as forest coverage, road classification, population density, and operational variables influence AVC risk, this research enables the development of targeted, location specific mitigation strategies. These might include dynamic signage in high-risk corridors, speed reductions in semi urban zones near wildlife attractants, or prioritization of fencing and wildlife crossings where road types and habitat patterns intersect. The tract level approach provides finer spatial resolution than most county-based models, allowing local agencies to allocate resources more effectively. Moreover, the integration of species-specific information through NLP opens new possibilities for tailoring mitigation to wildlife populations. As statewide crash databases and real time wildlife detection systems continue to evolve, this framework can be adapted into operational tools for forecasting risk and evaluating the effectiveness of implemented countermeasures. LIMITATIONS AND FUTURE DIRECTIONS While this study provides a robust and multidimensional framework for analyzing terrestrial AVCs, several limitations highlight areas for future improvement and research. Spatial Resolution and Aggregation Bias: The use of U.S. Census tracts as the spatial unit of analysis offers a practical approach, but it may obscure localized risks that occur at smaller scales, such as road segments or specific crossing zones. Future research could explore finer spatial resolution to reduce potential aggregation bias. Species Identification Uncertainty: Species information was inferred using a NLP applied to crash report narratives. Although this represents an innovative advancement, classification accuracy may vary, and future research should aim to validate these results or incorporate manually verified data when feasible. Unmeasured Confounding Factors: Despite adjusting for spatial autocorrelation using the Spatial Error Model (SEM), certain variables—such as wildlife fencing, driver distraction, enforcement presence, or seasonal animal movement—were unavailable. Incorporating these in future datasets could improve model performance and policy relevance. Temporal Dynamics: The current model is based on crash data aggregated from 2020 to 2024. It does not capture short-term fluctuations like migration periods, seasonal trends, or time-of-day patterns. Future efforts should include time-stamped data to assess temporal variability in AVC risk. Model Interpretability: While the SEM significantly improved fit, it does not capture direct spatial spillovers—how risk in one tract may influence its neighbors. Future comparisons with models such as the Spatial Lag or Conditional Autoregressive (CAR) model could provide deeper insights into spatial interaction effects. Micro Scale Analysis Within Clusters: The tract level spatial analysis provided insight into regional AVC patterns, but further investigation at the road segment level, particularly within high-high, low-low, and high-low LISA clusters, could reveal site specific factors that contribute to localized crash risk. This would support more precise mitigation design. Declarations AUTHOR CONTRIBUTIONS The authors confirm contribution to the paper as follows: study conception, design, data preprocessing and primary writing: Xi Wei; NLP-based crash data preparation: Sudesh Bhagat; statistical modeling and result interpretation: Xi Wei, Sudesh Bhagat; manuscript structure and narrative refinement: Aparna Joshi; conceptual guidance and methodological recommendations: Anuj Sharma. All authors contributed to revisions and approved the final version of the manuscript. ACKNOWLEDGMENTS This manuscript was supported in part by language and structural editing using ChatGPT (OpenAI), which assisted with improving grammar, clarity, and academic tone. All content was reviewed and verified by the authors to ensure factual accuracy and alignment with TRB ethical guidelines. References U.S. Department of Transportation, Federal Highway Administration (2008) Wildlife-vehicle collision reduction study: Report to Congress (Report No. FHWA-HRT-08-034). Retrieved May 18, 2025, from https://www.fhwa.dot.gov/publications/research/safety/08034/ Iowa Department of Transportation. (n.d.). Iowa Crash Analysis Tool [Data tool]. Retrieved May 18 (2025) from https://icat.iowadot.gov/ Bartonička T, Andrášik R, Duľa M, Sedoník J, Bíl M (2018) Identification of local factors causing clustering of animal-vehicle collisions. J Wildl Manag 82(5):940–947 Taylor-Brown A, Booth R, Gillett A, Mealy E, Ogbourne SM, Polkinghorne A, Conroy GC (2019) The impact of human activities on Australian wildlife. PLoS ONE, 14(1), e0206958 Rahman MA, Das S, Codjoe J, Mitran E, Sun X, Abedi K, Hossain MM (2023) Applying Data Mining Methods to Explore Animal-Vehicle Crashes. Transp Res Record: J Transp Res Board, 2677(11) Gregory A, Spence E, Beier P, Garding E (2021) Toward best management practices for ecological corridors. Land 10(2):140 Andrews KM, Nanjappa P, Riley SP (eds) (2015) Roads and ecological infrastructure: concepts and applications for small animals Tsuda LS, Carneiro CC, Quintanilha JA (2022) Using Self-Organizing Maps to find spatial relationships between wildlife-vehicle crashes and land use classes. An Acad Bras Cienc, 94(suppl 4), e20210727 Lao Y, Zhang G, Wu YJ, Wang Y (2011) Modeling animal–vehicle collisions considering animal–vehicle interactions. Accid Anal Prev 43(6):1991–1998 Sullivan JM (2011) Trends and characteristics of animal-vehicle collisions in the United States. J Saf Res 42(1):9–16 Snow NP, Zhang Z, Finley AO, Rudolph BA, Porter WF, Williams DM, Winterstein SR (2018) Regional-based mitigation to reduce wildlife–vehicle collisions. J Wildl Manag 82(4):756–765 Hubbard MW, Danielson BJ, Schmitz RA (2000) Factors influencing the location of deer-vehicle accidents in Iowa. J Wildl Manage 64(3):707–713 Galinskaitė L, Ulevičius A, Valskys V, Samas A, Busher PE, Ignatavičius G (2022) The influence of landscape structure on wildlife–vehicle collisions: Geostatistical analysis on hot spot and habitat proximity relations. ISPRS Int J Geo-Information 11(1):63 Lord D, Mannering F (2010) The statistical analysis of crash-frequency data: A review and assessment of methodological alternatives. Transp Res Part A: Policy Pract 44(5):291–305 Gunson KE, Mountrakis G, Quackenbush LJ (2011) Spatial wildlife-vehicle collision models: A review of current work. J Environ Manage 92(4):1074–1082 Huijser, M. P., McGowen, P., Fuller, J., Hardy, A., Kociolek, A., Clevenger, A. P.,… Ament, R. (2009). Cost-benefit analyses of mitigation measures aimed at reducing collisions with large ungulates. Ecology and Society, 14(2), 15 Bissonette JA, Kassar CA (2008) Locations of deer-vehicle collisions are unrelated to traffic volume or posted speed limit. Human–Wildlife Conflicts 2(1):122–126 Arévalo JE, Honda W, Arce-Arias A, Häger A (2017) Spatiotemporal variation of roadkills show mass mortality events for amphibians in a highly trafficked road adjacent to a national park, Costa Rica. Revista de Biología Trop 65(4):1261–1276 Haule HJ, Sando T, Lentz R, Chuan CH, Alluri P (2019) Evaluating the impact and clearance duration of freeway incidents. Int J Transp Sci Technol 8(1):13–24 Clevenger AP, Chruszcz B, Gunson KE (2003) Spatial patterns and factors influencing small vertebrate fauna road-kill aggregations. Biol Conserv 109(1):15–26 Bissonette JA, Cramer PC (2008) Evaluation of the use and effectiveness of wildlife crossings, vol 615. Transportation Research Board Western Transportation Institute (2021) Cost-effective solutions: Literature review on wildlife warning signs and animal-detection systems. Pooled Fund Study TPF-5(358). Retrieved from https://westerntransportationinstitute.org/wp-content/uploads/2021/12/4w7576_Huijser_etal_WVC_ConnectivityLiteratureReview_PooledFundStudyFinalReport_2021.pdf Litvaitis JA, Tash JP (2008) An approach toward understanding wildlife-vehicle collisions. Environ Manage 42(4):688–697 Lewis JS, Spaulding S, Swanson H, Keeley W, Gramza AR, VandeWoude S, Crooks KR (2021) Human activity influences wildlife populations and activity patterns: implications for spatial and temporal refuges. Ecosphere, 12(5), e03487 Chen Y, Ma J, Wang S (2019) Spatial regression analysis of pedestrian crashes based on point-of-interest data. J Data Anal Inform Process 8(1):1–19 Census US (2020) Bureau. Census Demographic and Housing Characteristics (DHC) Data: Table P1, Total Population by Census Tract, Iowa. data.census.gov. Accessed at May 2025. https://data.census.gov/map/040XX00US19$1400000/DECENNIALDHC2020/P1?q=Iowa&y=2020&d=DEC+Demographic+and+Housing+Characteristics&layer=VT_2020_140_00_PY_D1&loc=41.9575,-93.8730,z6.0000 Census Bureau US, TIGER/Line Shapefile (2020) Iowa Census Tracts. Accessed at May 2025. https://www2.census.gov/geo/tiger/TIGER2020/TRACT/tl_2020_19_tract.zip Iowa Department of Transportation. Road Network Portal. Iowa Department of Transportation. Accessed at (2025) May https://data.iowadot.gov/datasets/road-network-portal/ State of Iowa. Crash Data Dashboard: Vehicle Crashes in Iowa. Iowa Open Data Portal. Accessed at (2025) May https://data.iowa.gov/stories/s/m73g-tzgc Commission for Environmental Cooperation. Land Cover 30m – 2020. North American Environmental Atlas (2021) Accessed at May 2025. https://www.cec.org/north-american-environmental-atlas/land-cover-30m-2020/ Iowa Department of Natural Resources. Public Lands Used for Conservation and Recreation in Iowa. Iowa Geodata Portal. Accessed at (2025) May https://geodata.iowa.gov/datasets/2e8be3c67288413f903276eea17cea43_0 Deepchecks. Ollama LLM, Tool Integration (2024) Guide Accessed at May 2025. https://www.deepchecks.com/llm-tools/ollama/ Meta (2025) Meta-Llama-3-8B-Instruct. Hugging Face. Accessed at May https://www.deepchecks.com/llm-tools/ollama/ Additional Declarations The authors declare no competing interests. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8398693","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":562586599,"identity":"dc138268-821f-474f-bac6-55011e1c2e82","order_by":0,"name":"Xi Wei","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAu0lEQVRIiWNgGAWjYDACZgbGBwlwHhtxWpgNgFokIKqJ0gJUJsFAkhaD4zxmFQ9q6ur45zcfYPhQdpiwFslmHrMbCccOS0gcY0tgnHGOCC38zLzbbiSwHZBgOMZjwMzbRoQWNqCWgoR/dRLyx/g/MP8lRgvIFobENmYJg2M8DMyMxGiRbOb/LJHYd1hy47E0g4M959IJazE4fyzx449vdfxyhw8/fPCjzJqwFhRwgET1o2AUjIJRMApwAQA7OTXIMYdsNAAAAABJRU5ErkJggg==","orcid":"https://orcid.org/0009-0007-0099-9140","institution":"Iowa State University","correspondingAuthor":true,"prefix":"","firstName":"Xi","middleName":"","lastName":"Wei","suffix":""},{"id":562586600,"identity":"d16ab3c6-68c6-4bf4-b198-ee02296c386f","order_by":1,"name":"Sudesh Bhagat","email":"","orcid":"","institution":"Iowa State University","correspondingAuthor":false,"prefix":"","firstName":"Sudesh","middleName":"","lastName":"Bhagat","suffix":""},{"id":562586601,"identity":"c528e8f5-741e-4279-937e-8dfb9b3841f9","order_by":2,"name":"Aparna Joshi","email":"","orcid":"","institution":"Iowa State University","correspondingAuthor":false,"prefix":"","firstName":"Aparna","middleName":"","lastName":"Joshi","suffix":""},{"id":562586602,"identity":"ce7f1ed2-08f5-4575-ab8d-a623baedf4e1","order_by":3,"name":"Anuj Sharma","email":"","orcid":"","institution":"Iowa State University","correspondingAuthor":false,"prefix":"","firstName":"Anuj","middleName":"","lastName":"Sharma","suffix":""}],"badges":[],"createdAt":"2025-12-18 20:22:49","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-8398693/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8398693/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":98734979,"identity":"08f58893-a88d-4e27-a93f-6c0a2b73e010","added_by":"auto","created_at":"2025-12-22 06:21:51","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1771601,"visible":true,"origin":"","legend":"","description":"","filename":"TRB2026XiWei.docx","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/3fd7a83ae42c04382bdcda65.docx"},{"id":98779172,"identity":"59bdd668-1647-45d6-91f6-3ceb564e8057","added_by":"auto","created_at":"2025-12-22 12:30:01","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":342,"visible":true,"origin":"","legend":"","description":"","filename":"rs8398693.json","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/c72f5ec61ad4f35708546905.json"},{"id":98734984,"identity":"37509b38-2e72-4edf-b621-f73468e1aaa1","added_by":"auto","created_at":"2025-12-22 06:21:51","extension":"xml","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":101961,"visible":true,"origin":"","legend":"","description":"","filename":"rs83986930enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/7165da998de350777ad7aeb6.xml"},{"id":98734986,"identity":"b8293bea-ecba-4942-aee7-4aa46e88c011","added_by":"auto","created_at":"2025-12-22 06:21:51","extension":"png","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":157540,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/eb84465cdee3c18e018a83b5.png"},{"id":98735003,"identity":"22a21e11-89e0-4765-bb4e-7dcc04670ac2","added_by":"auto","created_at":"2025-12-22 06:21:52","extension":"png","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":63368,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage10.png","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/a5f6921bb3fc1d16a16b5630.png"},{"id":98734980,"identity":"a6d195d6-d009-4e38-8803-6a2850709ed2","added_by":"auto","created_at":"2025-12-22 06:21:51","extension":"png","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":37028,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage11.png","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/1c0d2c3f515494a9f615e3e7.png"},{"id":98735008,"identity":"822d2f0d-6746-4f47-a072-2a95359b40f1","added_by":"auto","created_at":"2025-12-22 06:21:52","extension":"png","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":35361,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage12.png","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/7670e6e167e6180257c8ffb2.png"},{"id":98777005,"identity":"82073adf-dfed-4854-8595-0bb7177308ca","added_by":"auto","created_at":"2025-12-22 12:25:00","extension":"png","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":297657,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/b2812ca3c816d46bc5fa9b21.png"},{"id":98777130,"identity":"49fd1667-36ab-4b92-9173-c6115dc77843","added_by":"auto","created_at":"2025-12-22 12:25:27","extension":"png","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":134965,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/d4d2b6387681cb94bad5e628.png"},{"id":98734999,"identity":"1fe10d39-1e51-4d21-9703-533f9d1e7ccc","added_by":"auto","created_at":"2025-12-22 06:21:52","extension":"png","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":849256,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/6b7d2f1e151ab06723776ab1.png"},{"id":98735001,"identity":"4da57e2a-5e82-4c1d-a207-a1d3542761e5","added_by":"auto","created_at":"2025-12-22 06:21:52","extension":"png","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":38624,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/ecbee662a3c42e562032ffac.png"},{"id":98734987,"identity":"212645fe-6db9-4259-85ea-cfe7fbe87ee8","added_by":"auto","created_at":"2025-12-22 06:21:51","extension":"png","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":12738,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/d5a2e96e12a51a18a7010535.png"},{"id":98776856,"identity":"daa801a7-e2f8-4d79-9c25-28a93b159746","added_by":"auto","created_at":"2025-12-22 12:23:43","extension":"png","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":9834,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/d301100afbc98ca88c6a776f.png"},{"id":98734997,"identity":"41507042-39f2-4880-99a8-090cfa397e34","added_by":"auto","created_at":"2025-12-22 06:21:52","extension":"png","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":5723,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/19c572283a5b6ef4d7f192b3.png"},{"id":98777166,"identity":"937c4791-d57c-4ee4-a626-61ee86645660","added_by":"auto","created_at":"2025-12-22 12:25:32","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":64202,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage9.png","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/12ba98596126ad28271e91bd.png"},{"id":98735016,"identity":"70e15d46-3714-4feb-a535-95afa0f9ae25","added_by":"auto","created_at":"2025-12-22 06:21:53","extension":"png","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":62233,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/3d9af377830a32ff93d016b0.png"},{"id":98735006,"identity":"9eaf732c-a91f-4e0d-ad0c-320544dbf509","added_by":"auto","created_at":"2025-12-22 06:21:52","extension":"png","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":47268,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage10.png","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/146e12e55a8a80542ecf93cb.png"},{"id":98779175,"identity":"21d5e29a-98e1-444a-862c-1d1b24e6bde0","added_by":"auto","created_at":"2025-12-22 12:30:01","extension":"png","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":19479,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage11.png","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/85e3896e0a30ab5906949ce2.png"},{"id":98777039,"identity":"ce13376d-e60c-4165-95da-7d7d394c3e9c","added_by":"auto","created_at":"2025-12-22 12:25:12","extension":"png","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":19114,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage12.png","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/17b025c7d7f282890b379ec5.png"},{"id":98777565,"identity":"d873a8aa-8d71-4a09-b853-52b48f479632","added_by":"auto","created_at":"2025-12-22 12:28:06","extension":"png","order_by":19,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":98467,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/e6140d28f89bc0a24dc95470.png"},{"id":98734989,"identity":"36c50515-2b50-42a7-a153-7dde0dc6d71a","added_by":"auto","created_at":"2025-12-22 06:21:51","extension":"png","order_by":20,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":51703,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/66d5e387e8fd1dec3d1dcc8a.png"},{"id":98776590,"identity":"11edc563-a3c4-4c07-9a2b-3b335bb8068a","added_by":"auto","created_at":"2025-12-22 12:23:10","extension":"png","order_by":21,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":108861,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/37640b55638dc1313a188734.png"},{"id":98734996,"identity":"477f6f77-6535-4098-990d-1e5f8c014e79","added_by":"auto","created_at":"2025-12-22 06:21:52","extension":"png","order_by":22,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":8948,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/7a86b021e7f16e15905b00d8.png"},{"id":98778774,"identity":"7d9ae574-d405-4dbb-a96e-17c69fb2df31","added_by":"auto","created_at":"2025-12-22 12:29:38","extension":"png","order_by":23,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":12484,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/bb52e211204b5aad4840eb35.png"},{"id":98735015,"identity":"fab48c83-2883-4ff2-a397-e076bf440342","added_by":"auto","created_at":"2025-12-22 06:21:53","extension":"png","order_by":24,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":10059,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/73fd056d40ffd7ecb7bfe626.png"},{"id":98735018,"identity":"c58921ea-c92f-409b-86f3-29760ff15c2a","added_by":"auto","created_at":"2025-12-22 06:21:53","extension":"png","order_by":25,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":3495,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/288872ad32a8372560b7405f.png"},{"id":98734988,"identity":"6b46c7cc-9b43-4fce-862e-9c2e2e44257b","added_by":"auto","created_at":"2025-12-22 06:21:51","extension":"png","order_by":26,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":47782,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage9.png","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/bf46e6c582430c647f471833.png"},{"id":98735002,"identity":"56c6acd8-1be3-48c0-8003-f316d4ceaa7a","added_by":"auto","created_at":"2025-12-22 06:21:52","extension":"xml","order_by":27,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":100919,"visible":true,"origin":"","legend":"","description":"","filename":"rs83986930structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/e967eb5fe5823d027c5e3e93.xml"},{"id":98734994,"identity":"7201316e-6a1b-480d-b6d3-9009a2d3944c","added_by":"auto","created_at":"2025-12-22 06:21:52","extension":"html","order_by":28,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":110969,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/ccff20d3567605fdc61f0a43.html"},{"id":98734981,"identity":"5fa8a444-1a12-487a-8e8f-73d39ee0e554","added_by":"auto","created_at":"2025-12-22 06:21:51","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":444659,"visible":true,"origin":"","legend":"\u003cp\u003eIowa 2020 Census Tract Map\u003c/p\u003e","description":"","filename":"Picture1.png","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/bc9b8652af00cec8694dd755.png"},{"id":98776732,"identity":"eff18746-1e89-493f-bf7b-fdcbd48e0f35","added_by":"auto","created_at":"2025-12-22 12:23:25","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":1028287,"visible":true,"origin":"","legend":"\u003cp\u003e2024 Iowa Road Network\u003c/p\u003e","description":"","filename":"Picture2.png","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/847fac93b7d99804b442821c.png"},{"id":98734978,"identity":"fde58e7f-cbfc-4b09-85bf-cebea23167d5","added_by":"auto","created_at":"2025-12-22 06:21:51","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":367095,"visible":true,"origin":"","legend":"\u003cp\u003eIowa tract level AVCs 2020-2024\u003c/p\u003e","description":"","filename":"Picture3.png","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/a711c448d88e10624b8fce55.png"},{"id":98734992,"identity":"2e34f503-1b75-4d4e-afd8-2539fa85301c","added_by":"auto","created_at":"2025-12-22 06:21:52","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":265616,"visible":true,"origin":"","legend":"\u003cp\u003e2020 Iowa Forest Land Cover Map after Reclassification\u003c/p\u003e","description":"","filename":"Picture4.png","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/8479e826e6df158ce1d82685.png"},{"id":98735004,"identity":"3125fba7-3138-4dc0-bbde-7200b1d63f96","added_by":"auto","created_at":"2025-12-22 06:21:52","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":51729,"visible":true,"origin":"","legend":"\u003cp\u003eTop 10 Animal Species Identified in Crash Narratives Using NLP Classification (2020–2024)\u003c/p\u003e","description":"","filename":"Picture5.png","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/afa1a3f497af9e9afeede6ff.png"},{"id":98778692,"identity":"ffadadf0-6303-4a7f-b717-62f5656f6c5f","added_by":"auto","created_at":"2025-12-22 12:29:32","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":26311,"visible":true,"origin":"","legend":"\u003cp\u003ePredicted Terrestrial AVC vs. Actual Terrestrial AVC\u003c/p\u003e","description":"","filename":"Picture6.png","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/2097fbab089f0fbd6a290c47.png"},{"id":98776936,"identity":"4c975dff-a2f3-4880-93ae-7bc531ba02cd","added_by":"auto","created_at":"2025-12-22 12:24:33","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":20218,"visible":true,"origin":"","legend":"\u003cp\u003eResidual Plot\u003c/p\u003e","description":"","filename":"Picture7.png","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/a99fe90fecb7d8903e7e8d9b.png"},{"id":98735010,"identity":"b0d5c59a-94ed-4596-9b03-e8d693071f16","added_by":"auto","created_at":"2025-12-22 06:21:53","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":14718,"visible":true,"origin":"","legend":"\u003cp\u003eMoran’s I for predicted Terrestrial AVCs\u003c/p\u003e","description":"","filename":"Picture8.png","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/9ca2c0ea7b1d4ab9baed0c78.png"},{"id":98734991,"identity":"6365bdb0-33be-468c-bd79-5e6bec3e36fd","added_by":"auto","created_at":"2025-12-22 06:21:51","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":255222,"visible":true,"origin":"","legend":"\u003cp\u003eLISA Cluster Map\u003c/p\u003e","description":"","filename":"Picture9.png","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/e5804d7a07c04fc733e3fcd0.png"},{"id":98735013,"identity":"2be6f885-e716-472f-9763-2775630dcd2c","added_by":"auto","created_at":"2025-12-22 06:21:53","extension":"png","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":267085,"visible":true,"origin":"","legend":"\u003cp\u003eLISA Significant Map\u003c/p\u003e","description":"","filename":"Picture10.png","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/7a81b0fa6d4b1b1f0f6a8e5d.png"},{"id":98735009,"identity":"8a4a00ad-dc32-47ac-b3a6-10d6b40a4100","added_by":"auto","created_at":"2025-12-22 06:21:52","extension":"png","order_by":11,"title":"Figure 11","display":"","copyAsset":false,"role":"figure","size":130303,"visible":true,"origin":"","legend":"\u003cp\u003ePredicted AVCs - SEM based\u003c/p\u003e","description":"","filename":"Picture11.png","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/11015849bd11bc14a06d1995.png"},{"id":98734995,"identity":"794a39a9-d1d4-432e-9882-100ff231d691","added_by":"auto","created_at":"2025-12-22 06:21:52","extension":"png","order_by":12,"title":"Figure 12","display":"","copyAsset":false,"role":"figure","size":151687,"visible":true,"origin":"","legend":"\u003cp\u003eResiduals Map\u003c/p\u003e","description":"","filename":"Picture12.png","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/7c368252055f5ca33ac7e184.png"},{"id":98797651,"identity":"ce0c8da1-4a23-498f-8c3a-c7afc6ea1e34","added_by":"auto","created_at":"2025-12-22 13:40:09","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3355495,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8398693/v1/4bb27860-3b51-4bb6-90d9-40e3486e0129.pdf"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003eSpatial Analysis and Prediction of Animal Vehicle Collisions (AVCs) in Iowa Using Environmental and Anthropogenic Factors\u003c/p\u003e","fulltext":[{"header":"INTRODUCTION","content":"\u003cp\u003eEach year, more than five million vehicle crashes occur across the United States, resulting in nearly three million injuries and approximately 40,000 fatalities (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e). While public attention often centers on crashes caused by speeding, distracted driving, or weather, a less visible yet significant threat unfolds daily on rural roads: AVCs (animal-vehicle collisions). These crashes account for roughly 26,000 injuries, 200 deaths, and over \u003cspan\u003e$\u003c/span\u003e8\u0026nbsp;billion in damage annually in the United States (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e). Yet despite their magnitude, AVCs remain overlooked in mainstream traffic safety discourse.\u003c/p\u003e \u003cp\u003eIn a state like Iowa, the risk of AVCs is especially pronounced. The landscape, comprising farmland, woodlands, streams, and ravines, supports a diverse range of wildlife while also being heavily intersected by road networks. Areas where dense vegetation or natural habitats lie close to roadways become common crossing points for animals such as deer. As a result, Iowa experiences 8,000 animal-related crashes each year, over 14 percent of the state\u0026rsquo;s total crash count (\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eAVCs tend to follow spatial and seasonal patterns influenced by both environmental and human-related factors (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e)(\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e). Because AVCs are spatially clustered (\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e), driven by both ecological corridors (\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e) and infrastructure design (\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e), statistical models must account for both crash frequency and spatial dependence. Prior research has examined these relationships using variables such as land cover (\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e), traffic volume (\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e), and speed limits (\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e), often at coarse spatial scales (\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e). However, these studies rarely consider operational and roadway characteristics that influence crash occurrence, crash severity, and post-crash recovery. These include incident clearance time, road clearance time, and surface width (lane width), which are key variables that directly affect emergency response effectiveness, secondary crash risk, and roadway hazard duration. Additionally, average annual daily traffic (AADT) and posted speed limits play critical roles in shaping driver response time and stopping distance but remain inconsistently integrated across ecological AVC models.\u003c/p\u003e \u003cp\u003eThis study addresses existing research gaps by integrating a comprehensive set of environmental, anthropogenic, and transportation-related variables into a spatial modeling framework at the census tract level across Iowa. We aggregate diverse datasets, including ecological land cover types such as needleleaf, broadleaf, mixed forest, wetlands, cropland, shrubland, grassland, and urban areas, as well as land-use designations associated with high animal activity such as Wildlife Refuges, Nature Preserves, City Parks, Greenbelts, and Trail systems to the tract scale. From the transportation domain, we include road network characteristics such as Annual Average Daily Traffic (AADT), speed limits, and surface widths, alongside crash-level operational variables such as incident and road clearance times. To evaluate the influence of these factors on AVC frequencies, we employ a negative binomial regression model, a widely used method for analyzing overdispersed crash count data. This tract-level modeling approach enables spatial resolution, which has typically relied on coarser spatial scales such as county-level data (\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e)(\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e), allowing for more localized insights into AVC risk factors.\u003c/p\u003e \u003cp\u003eFurthermore, we introduce a novel layer of species-specific analysis by applying a natural language processing (NLP) model to infer the animal species involved in each crash. This allows us to move beyond treating AVCs as a homogeneous category and instead examine which species are most frequently affected and under what conditions. By integrating environmental context, roadway design, traffic operations, and species-level identification, this study provides a more holistic and actionable framework for detecting collision hotspots and improving mitigation strategies in wildlife-prone regions.\u003c/p\u003e"},{"header":"LITERATURE REVIEW","content":"\u003cp\u003eAVCs remain a persistent challenge at the human–wildlife interface. Research in this area has progressed through distinct phases, uncovering the multidimensional nature of these events. Early ecological studies established that land cover patterns are strongly associated with AVC risk. Hubbard et al. identified crash hotspots near habitat features such as forests, rivers, and wetlands, which facilitate wildlife movement across roadways and increase collision risk (\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e). A study by Galinskaitė et al. confirmed the role of wetlands and agricultural edges in attracting wildlife across varying biogeographic regions (\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e). These spatial ecological studies provided a foundation for predictive modeling but were limited in their analytical scope.\u003c/p\u003e \u003cp\u003eMethodological advances soon followed to address the statistical complexity of AVC data. Lord and Mannering demonstrated that traditional Poisson models often fail to capture the overdispersed and rare nature of AVCs and introduced alternative approaches such as diagonal-inflated Poisson and Negative Binomial (NB) models, with the latter showing improved model fit (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e). Technological developments have further enhanced AVC monitoring and analysis. Gunson et al. described how crowdsourcing platforms and automated tools, including lidar and thermal sensors, enable real-time wildlife detection and crash reporting (\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e). Geographic information systems (GIS) have also expanded spatial modeling capabilities, allowing for visualization of collision clusters across time and diverse landscape features.\u003c/p\u003e \u003cp\u003eDespite ongoing advances, critical knowledge gaps remain. Many studies group all wildlife species together, overlooking biological differences in movement and behavior. For example, Huijser et al. found that treating deer, rodents, and reptiles as a single group ignores key differences, such as movement speed and avoidance ability, that influence the likelihood of a collision (\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e). Another limitation is that ecological models often leave out transportation infrastructure variables that influence crash risk. Bissonette and Kassar demonstrated that road type alone, from high-speed interstates to low-volume collectors, can significantly affect collision probability, regardless of habitat conditions (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e). However, some studies continue to focus solely on habitat and land cover features while omitting traffic or road design information. For instance, Arévalo et al. examined patterns of amphibian road mortality near a national park in Costa Rica using ecological factors, but without accounting for road classification, traffic volume, or speed (\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e). This lack of integration between ecological and transportation data may lead to blind spots in high-risk areas, such as farmlands near protected lands, where animals frequently encounter fast-moving traffic.\u003c/p\u003e \u003cp\u003eIn addition to ecological gaps, operational characteristics, such as incident clearance time and road clearance time, remain largely absent from AVC studies. For example, Arévalo et al. and Galinskaitė et al. did not incorporate post-crash response or clearance duration variables into their spatial analyses, despite these factors being widely used in broader traffic safety research (\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e). Haule et al. demonstrated that such variables are critical for understanding crash severity, emergency response efficiency, and overall system resilience (\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e). Including these operational measures in AVC models may provide insights into both crash causation and exposure duration. Roadway geometry, such as surface width, shoulder presence, or lane configuration, —is also underrepresented in most AVC studies, even though these design features can influence both animal crossing behavior and driver avoidance maneuvers (\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e). While traffic volume (AADT) and speed limits are sometimes included, they are often treated as categorical or control variables, rather than continuous predictors capable of explaining collision frequency or species-specific vulnerability (\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eTechnological countermeasures have demonstrated potential in reducing AVC, but their implementation remains limited. Automated wildlife warning systems, for example, have been reported to reduce collisions by 30–50% under controlled conditions (\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e). However, these systems are typically species-insensitive and rarely integrate key roadway engineering variables such as road classification, traffic volume, or speed, factors known to influence collision likelihood. As a result, their effectiveness in real-world settings remains constrained. Furthermore, there is no comprehensive framework that explains why wildlife select specific road segments for crossing. The continued disconnect between ecological drivers (e.g., habitat suitability and migratory behavior) and anthropogenic risk factors (e.g., roadway infrastructure and human disturbance) hampers efforts to develop effective and targeted mitigation strategies.\u003c/p\u003e \u003cp\u003eLand-use designations, such as wildlife refuges, city parks, nature preserves, and greenbelt trails, often act as attractants or movement corridors for wildlife. Despite their ecological importance, these areas are inconsistently represented in AVC prediction models. Their presence can increase the likelihood of animal crossings and elevate collision risk on adjacent road segments, particularly during migration periods or seasonal peaks in animal activity (\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eAs terrestrial species accounted for 99.7% of serious collisions in the dataset, birds and other flying species were excluded due to fundamental differences in their collision dynamics. This allowed for a focused analysis of ground-level interactions where vehicle trajectories intersect with terrestrial wildlife movement. To enable species-specific risk analysis, a machine learning model was used to classify the animal species involved in each crash, an approach that has not previously been applied at this scale. This innovation helps overcome the long-standing “species blind spot” noted by Litvaitis and Tash, who emphasized the risks of treating wildlife as a homogenous group in collision analyses (\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe core theoretical contribution of this study lies in its integrative analytical framework. We combined environmental predictors (e.g., forest cover, wetland density, cropland interface, and wildlife refuge buffers) with anthropogenic variables (e.g., tract-level population, AADT, speed limits, and road classification). Additionally, we incorporated geometric and operational characteristics, such as surface width, incident clearance time, road clearance time, and refined land-use categories. This multidimensional perspective revealed previously undocumented interactions: for example, wetlands near highway arterials reduce AVCs regardless of habitat quality, while agricultural edges adjacent to low-volume roads may act as ecological traps during harvest seasons. Wildlife corridors intersecting secondary arterials showed increased risk when speed limits exceeded average braking distance.\u003c/p\u003e \u003cp\u003eDespite the growing use of flexible regression models, many AVCs studies continue to assume spatial independence of crash occurrences, an assumption often violated when nearby tracts share similar unobserved risks, such as localized patterns of human activity (\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e). Spatial econometric techniques, such as the Spatial Error Model (SEM), offer a robust approach to account for spatial dependence on model residuals. SEMs are particularly effective when omitted variables are spatially correlated, thereby biasing estimates if ignored. Prior traffic safety studies, including crash frequency models in Louisiana and Pennsylvania, have demonstrated that SEMs improve model fitness and reduce misspecification bias in segment-level crash data (\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e). Incorporating SEMs enhances model accuracy, enables detection of spatial clusters of error, and supports more reliable, spatially explicit predictions.\u003c/p\u003e \u003cp\u003eKey Contributions of This Study:\u003c/p\u003e \u003cp\u003e \u003c/p\u003e\u003cul\u003e \u003cli\u003e \u003cp\u003eDeveloped a census tract–level modeling framework to capture localized AVC risk patterns with greater spatial precision.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eIntegrated ecological, anthropogenic, geometric, and operational variables in a unified statistical framework.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eApplied a machine learning NLP pipeline to extract species-level crash data from narrative crash reports, addressing species-blind modeling limitations.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eDemonstrated the benefit of spatial error modeling in correcting residual autocorrelation and improving predictive accuracy.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eProvided a transferable methodology and practical insights to inform wildlife–vehicle conflict mitigation strategies across different regions and agencies.\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003cp\u003e\u003c/p\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003cdiv id=\"Sec4\" class=\"Section3\"\u003e \u003c/div\u003e \u003c/div\u003e\n\n \u003cp\u003e \u003c/p\u003e\n\n \u003cp\u003e \u003c/p\u003e\n\n \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e\n\n \n\n\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e\u003c/p\u003e \u003c/div\u003e "},{"header":"METHODS","content":"\u003ch2\u003eData Sources and Processing\u003c/h2\u003e\u003cp\u003eThis study integrates diverse spatial and tabular datasets to analyze terrestrial AVCs across Iowa at the U.S. Census tract level. The analysis covers 2020–2024, using 2020 census tract boundaries and population data as the spatial framework for geoprocessing and aggregation. Census tracts were selected for their finer granularity compared to counties and compatibility with available environmental, crash, and infrastructure datasets.\u003c/p\u003e\u003cp\u003eData formats included shapefiles (e.g., road network, public lands), raster grids (e.g., land cover), and tabular CSVs (e.g., crash records, population). All datasets were processed using ArcGIS Pro and R through coordinating harmonization, spatial join, raster reclassification, zonal statistics, and aggregation to census tracts. These steps ensured consistent spatial alignment and standardized variable construction. Detailed processing steps are described in the following subsections.\u003c/p\u003e\u003ch3\u003ePopulation and Human Activity\u003c/h3\u003e\u003cp\u003eTo represent human activity, we used 2020 population data from the U.S. Census Bureau (\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e) and tract boundaries from the 2020 TIGER/Line shapefiles (\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e). Population counts were joined to census tract geometries using the unique GEOID identifier. The resulting tract-level population values were used as a continuous predictor in our models and served as a proxy for anthropogenic pressure and development density, one of the factors known to influence wildlife movement and AVC likelihood. These patterns are visualized in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, which displays population distribution across Iowa’s census tracts.\u003c/p\u003e\u003ch3\u003eRoad Network and Traffic Data\u003c/h3\u003e\u003cp\u003eThe road network data were obtained as a shapefile from the Iowa Department of Transportation (Iowa DOT) Road Network Portal (\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e). This dataset includes roadway geometries along with embedded attributes such as Federal Functional Class, Average Annual Daily Traffic (AADT), posted speed limit, surface width, and a unique Route ID for each segment. Roads were classified into six categories: Interstate, Other Principal Arterial, Minor Arterial, Major Collector, Minor Collector, and Local, reflecting differences in capacity and network function (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eThe shapefile was then spatially joined to 2020 Census tract boundaries, assigning each road segment to the tract it intersects. Tract-level traffic variables, including mean AADT and mean speed limit, were calculated to serve as proxies for traffic intensity and driver exposure.\u003c/p\u003e\u003ch3\u003eCrash Data\u003c/h3\u003e\u003cp\u003eAVC records from 2020 to 2024 were obtained from the State of Iowa Crash Data Base (\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e) and spatially joined to 2020 U.S. Census tracts using geographic coordinates. Each crash record included information such as crash date, latitude and longitude, cause, and post-crash clearance times.\u003c/p\u003e\u003cp\u003eTo narrow the focus of our analysis, we excluded non-terrestrial species and retained only terrestrial AVCs, which represented 99.7% of all AVCs during this period. The total count of terrestrial AVCs per tract served as the response variable in our modeling framework.\u003c/p\u003e\u003cp\u003eTo support spatial interpretation and communication, we generated several visualizations. Figure\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e displays point locations of AVCs across Iowa from 2020 to 2024. Figure\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e presents tract-level AVC counts, classified using the natural breaks (Jenks) method to highlight meaningful variation in collision frequency. Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e provides a sample of the raw crash dataset, including key variables such as Route ID, geographic coordinates, crash time, cause, and both roadway and incident clearance times, which were used as operational predictors in the model.\u003c/p\u003e\u003cdiv class=\"gridtable\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eRaw Crash data sample\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e\u003ccolgroup cols=\"8\"\u003e\u003c/colgroup\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCrash Key\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRoute ID\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eLatitude\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLongitude\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eCrash Date\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eMajor Cause\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eTotal Roadway Clearance Time\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eTotal Incident Clearance Time\u003c/p\u003e \u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e20201154658\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eC000343420N\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e43.30339\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-91.22849\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1/4/2020 14:10\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAnimal\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e000:52\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e000:52\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e20201154717\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eS001920030E\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e42.00765\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-93.38498\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1/4/2020 17:25\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAnimal\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e000:11\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e000:35\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e20201154720\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eS001910080W\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e41.67961\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-91.46715\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1/5/2020 4:41\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAnimal\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e000:49\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e000:19\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/table\u003e\u003c/div\u003e\u003ch2\u003eLand Cover Data\u003c/h2\u003e\u003cp\u003eThe base land cover dataset consisted of multiple natural and human-modified land types, including needleleaf forest, broadleaf forest, mixed forest, shrubland, grassland, wetland, cropland, urban and built-up areas, barren land, and water. These classes were derived from the 2020 North American Land Cover 30m raster dataset provided by the Commission for Environmental Cooperation (CEC) (\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eTo simplify model input while preserving ecological relevance, the original land cover types were reclassified into six broader categories: Forest, Shrubland/Grassland, Wetland, Cropland, Urban/Built-up, and Water (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e). This generalization helped reduce classification noise from overly specific vegetation subtypes and ensured alignment with wildlife habitat associations reported in prior literature (\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eUsing spatial overlays and zonal statistics, the proportional area of each reclassified land cover type was calculated for every census tract. The resulting tract-level variables (e.g., forest proportion, wetland proportion) were used as key environmental predictors in the modeling framework.\u003c/p\u003e\u003ch3\u003ePublic Land Use Reclassification\u003c/h3\u003e\u003cp\u003eTo better represent wildlife-attractive environments, we reclassified the public land dataset provided by Iowa DOT (\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e) based on the designation and intended ecological use of each public land unit. Each unit was defined as a spatial polygon representing a distinct area of publicly managed land and was evaluated for its potential to support or attract wildlife. The units were grouped and weighed as follows:\u003c/p\u003e\u003cp\u003eHigh relevance (weight = 1.0): Includes Wildlife Refuges, Nature Preserves, National Wildlife Refuges, and State Wildlife Refuges.\u003c/p\u003e\u003cp\u003eModerate relevance (weight = 0.5): Includes City Parks, Greenbelts, and Trail Systems.\u003c/p\u003e\u003cp\u003eLow or no relevance (weight = 0): Includes public lands not directly associated with habitat or recreational use.\u003c/p\u003e\u003cp\u003eEach polygon's area (in acres) was multiplied by its assigned weight. These weighted areas were then summed at the census tract level to compute a continuous variable: Wildlife-Related Area Acres. To account for tract size and enable cross-tract comparisons, this value was then divided by the total area of each tract to produce the final modeling variable: Wildlife-Related Area Proportion shown as Eq.\u0026nbsp;1.\u003c/p\u003e\u003cp\u003eWildlife-Related Area Proportion\u003csub\u003ei\u003c/sub\u003e = \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\frac{{\\sum\\:}_{j=1}^{n}\\text{A}\\text{ᵢ}\\text{ⱼ}\\:\\times\\:\\:\\text{w}\\text{ⱼ}}{\\text{T}\\text{o}\\text{t}\\text{a}\\text{l}\\:\\text{T}\\text{r}\\text{a}\\text{c}\\text{t}\\:\\text{A}\\text{r}\\text{e}\\text{a}\\text{ᵢ}}\\)\u003c/span\u003e\u003c/span\u003e (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e)\u003c/p\u003e\u003cp\u003eWhere:\u003c/p\u003e\u003cp\u003eA\u003csub\u003ei\u003c/sub\u003eⱼ = area (in acres) of public land unit j in tract i\u003c/p\u003e\u003cp\u003ewⱼ = assigned weight based on land use relevance (1.0, 0.5, or 0)\u003c/p\u003e\u003cp\u003eTotal Tract Area\u003csub\u003ei\u003c/sub\u003e = total land area of tract i\u003c/p\u003e\u003ch3\u003eModeling Approach\u003c/h3\u003e\u003ch2\u003eNLP Modeling\u003c/h2\u003e\u003cp\u003eTo identify and classify animal-related crashes, structured crash data was merged with their corresponding narrative descriptions. A natural language processing (NLP) pipeline was developed using a large language model (LLM) via the Ollama framework to extract relevant information from these unstructured texts (\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e). Narratives were preprocessed by converting date fields and removing records with missing or non-informative content. The primary objective was twofold: 1 is to determine whether an animal was involved in a crash, and 2 if so, to identify the specific animal type mentioned. The objective was not to identify additional animal crashes. The classification was performed using the LLaMA 3 (8B) model, a decoder-only transformer architecture released by Meta in April 2024 (\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e). With approximately 8\u0026nbsp;billion parameters, this model incorporates Grouped Query Attention, rotary position embeddings, and a 128k-token vocabulary, enabling efficient and accurate long-context reasoning. A classification accuracy of 90% was achieved. The model was hosted locally through Ollama and executed using a standardized prompt:\u003c/p\u003e\u003cp\u003e\"You are a crash narrative analysis assistant. Analyze the following narrative and answer these two questions:\u003c/p\u003e\u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eIs this crash related to an animal? Answer only 'yes' or 'no'.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eIf yes, what kind of animal was involved? Answer with the animal type (e.g., deer, dog, cattle, bird, raccoon, etc.). If unknown or not applicable, say 'none'. Respond in this JSON format only: { 'animal_involved': 'yes' or 'no', 'animal_type': 'deer' or 'dog' or 'bird' or 'cattle' or 'raccoon' or 'other' or 'none' }.\"\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e\u003cp\u003eThis constrained output format ensured consistency across thousands of crash narratives and minimized parsing errors. Exception handling was incorporated to manage incomplete or malformed responses. Once classification was complete, the results were merged back into the original crash dataset. From this enriched dataset, a filtered subset of animal-related crashes was generated. Figure\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e presents the distribution of the most frequently identified animal types.\u003c/p\u003e\u003cdiv class=\"gridtable\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eTop 10 Animal Crash Count and Percentage\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e\u003ccolgroup cols=\"3\"\u003e\u003c/colgroup\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAnimal Type\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCount\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePercentage\u003c/p\u003e \u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDeer\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e34658\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e95.56\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCattle\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e366\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.01\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDog\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e317\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.87\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRaccoon\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e226\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.62\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCow\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e146\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.4\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCoyote\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e137\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.38\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHorse\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e92\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.25\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTurkey\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e57\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.16\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCalf\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e42\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.12\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFox\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e16\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.04\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/table\u003e\u003c/div\u003e\u003cp\u003eThis structured classification allowed for subsequent research utilizing the negative binomial and geospatial visualization to discover where animal crashes are more likely to occur.\u003c/p\u003e\u003ch2\u003eNegative Binomial and Spatial Error Modeling\u003c/h2\u003e\u003cp\u003eTo examine how environmental and anthropogenic factors influence the frequency of terrestrial AVCs, we developed a statistical modeling framework using census tract level data across Iowa. The dependent variable was the count of AVCs, which exhibited overdispersion where the variance exceeded the mean, a common feature of both crash and ecological data. We initially evaluated several generalized linear models including Poisson regression, Zero Inflated Poisson, and Negative Binomial regression. A dispersion test revealed significant overdispersion, making the Poisson model unsuitable. Although AVC counts included some zeros, Vuong’s test indicated that the Zero Inflated Poisson model did not provide a significantly better fit than the Negative Binomial model. Additionally, the Zero Inflated Poisson model assumes two separate data generating processes, one for excess zeros and one for count values, which was not theoretically justified in this context. Based on model diagnostics and fit statistics, including the Akaike Information Criterion and residual deviance, and consistent with prior literature, the Negative Binomial model was selected as the final specification.\u003c/p\u003e\u003cp\u003eTo improve model parsimony and predictive performance, we applied stepwise AIC selection in both forward and backward directions to identify the most informative subset of predictors. Variables were retained based on a combination of statistical significance and theoretical relevance. To assess residual spatial dependence after fitting the Negative Binomial model, we computed Moran’s I and conducted Local Indicators of Spatial Association analysis. These diagnostics revealed significant spatial clustering in the residuals, suggesting that unobserved spatial processes were influencing AVC outcomes. Based on these findings, we estimated a Spatial Error Model to account for latent spatial autocorrelation while retaining the structure of the selected covariates.\u003c/p\u003e\u003cp\u003eThe initial set of predictor variables included environmental factors such as the proportions of forest, wetland, shrubland and grassland, water, cropland, and wildlife related public lands, along with anthropogenic factors including log transformed total population, average speed limit, average annual daily traffic, and road lengths by functional classification. To improve interpretability and reduce skewness, the population variable was log transformed. Interaction terms between each land cover proportion and log transformed population were included to capture how human presence modifies the effect of land use on AVC risk. Rather than applying an offset for tract area, the model adjusted for variation in population and landscape composition through these transformed and interacted covariates. The final Negative Binomial specification included all land type variables interacted with population along with transportation related covariates.\u003c/p\u003e\u003cp\u003eThe Spatial Error Model was implemented in R using the spdep and spatialreg packages. A Queen contiguity based spatial weights matrix was constructed from tract geometries using poly2nb and nb2listw functions. The model was estimated using the errorsarlm function and compared with a spatial lag model using AIC values. Final predictions and residuals were joined back to the spatial data object for visualization and assessment.\u003c/p\u003e"},{"header":"RESULTS","content":"\u003cp\u003eThe final Negative Binomial (NB) regression model identified several significant predictors of terrestrial AVCs frequency across Iowa census tracts (Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). The model exhibited a good overall fit (AIC\u0026thinsp;=\u0026thinsp;6390.3; residual deviance\u0026thinsp;=\u0026thinsp;847.21 on 790 df) and provided interpretable coefficients for both environmental and anthropogenic factors.\u003c/p\u003e \u003cp\u003eA 10% increase in forest cover was associated with a 20\u0026ndash;22% increase in AVC risk, consistent with the positive coefficient for forest proportion (β\u0026thinsp;=\u0026thinsp;2.03, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). This relationship reflects the ecological role of forests as high-use wildlife habitats and corridors for ungulate species such as deer, increasing frequency of crossing and collision exposure (\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eLog-transformed population was also positively associated with AVC frequency (β\u0026thinsp;=\u0026thinsp;0.196, p\u0026thinsp;=\u0026thinsp;0.001), indicating that areas with greater human presence face elevated collision risk, likely due to more vehicle activity and habitat fragmentation. In contrast, urban built-up areas showed a strong negative association (β = \u0026minus;\u0026thinsp;1.947, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), suggesting that highly developed regions with impervious surfaces and limited habitat are less frequented by terrestrial wildlife, thereby reducing collision risk. This is in line with findings by Bissonette and Kassar (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e), who showed that urbanization acts as a movement barrier for wildlife.\u003c/p\u003e \u003cp\u003eTransportation variables, including AADT, posted speed limit, and lengths of roads across all functional classes, were also positively associated with AVCs (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), confirming that higher traffic exposure increases collision risk. These findings are consistent with past AVC studies that emphasize road speed and volume as critical collision predictors (\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eFinal Negative Binomial regression model\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVariable\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eEstimate\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eStd. Error\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003ez value\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003ep-value\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e(Intercept)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.647\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.511\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.266\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.206\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eForest_prop\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2.03***\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.2572\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e7.892\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003elog_POP\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.196**\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.0598\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3.281\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUrbanBuiltup_prop\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-1.947***\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.1049\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e-18.564\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eavg_AADT\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1.71e-05***\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3.17e-06\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e5.402\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eavg_SPEED_LIMIT\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.01683***\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.00323\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e5.216\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInterstate\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.00833***\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.00167\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e4.986\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMinorArterial\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.01306***\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.00172\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e7.579\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMajorCollector\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.02004***\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.00189\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e10.585\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMinorCollector\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.01614***\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.00478\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3.377\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLocal\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.03384***\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.00295\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e11.466\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eDespite the NB model\u0026rsquo;s reasonable predictive ability, as demonstrated by the 10-fold cross-validation plot of predicted versus actual AVC counts (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e), the residual plot (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e) revealed increasing underprediction at higher crash counts, suggesting unaccounted spatial structure.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eTo assess spatial dependence in the NB model residuals, we first applied Moran\u0026rsquo;s I, which yielded a positive value of 0.136, indicating significant global spatial clustering (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e). Local Indicators of Spatial Association (LISA) analysis further identified distinct high\u0026ndash;high and low\u0026ndash;low residual clusters, particularly in eastern and southern Iowa (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e). The corresponding LISA significance map revealed that many of these clusters were statistically significant (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05), reinforcing the conclusion that unobserved spatial processes were influencing AVC outcomes (Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e10\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eNotably, several low\u0026ndash;low clusters appeared in central Iowa, which includes more urbanized areas with high population density and AADT. These patterns suggest that despite greater human activity, such areas tend to have fewer AVCs, possibly due to lower wildlife presence or effective roadway barriers. In contrast, many high\u0026ndash;high clusters were observed in rural regions, where natural habitats are abundant and less fragmented, leading to consistently elevated AVC frequencies across neighboring tracts. Additionally, high\u0026ndash;low and low\u0026ndash;high clusters may reflect localized anomalies such as hazardous road segments or mitigation infrastructure.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eTo correct these spatial dependencies, we estimated a Spatial Error Model (SEM) using the same covariates. The SEM\u0026rsquo;s spatial autoregressive coefficient (λ\u0026thinsp;=\u0026thinsp;0.211, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) was statistically significant, and the model demonstrated improved fit with a lower AIC (7090.7) compared to the NB model (7105). While coefficient interpretation was not the primary focus of the SEM, it substantially improved predictive accuracy and reduced residual spatial bias.\u003c/p\u003e \u003cp\u003eAs shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e11\u003c/span\u003e, the SEM-generated predicted AVC intensity map revealed smoother spatial gradients across Iowa, particularly reducing overprediction in low-density rural areas and underprediction in high-density zones. The corresponding residual map (Fig.\u0026nbsp;\u003cspan refid=\"Fig12\" class=\"InternalRef\"\u003e12\u003c/span\u003e) illustrates a notable decrease in spatial clustering and overall residual magnitude compared to the NB model. This suggests that the SEM more effectively captured latent, spatially structured risk factors and improved model calibration at both ends of the prediction spectrum.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e"},{"header":"CONCLUSION","content":"\u003cp\u003eThis study presents a comprehensive spatial analysis of terrestrial AVCs across Iowa from 2020 to 2024, integrating ecological, anthropogenic, geometric, and operational variables into a robust census tract level framework. A novel contribution of this research lies in its use of species-specific crash classification derived from narrative data and the explicit incorporation of operational characteristics, such as incident and road clearance time, seldom addressed in prior AVC studies.\u003c/p\u003e \u003cp\u003eAt the census tract level, the Negative Binomial (NB) regression identified several key factors associated with increased AVC frequency. Forested areas were linked to higher collision risk, while urban and built-up zones saw fewer crashes. Human-related factors such as greater population density, higher traffic volumes, and faster speed limits were also positively associated with AVCs. Additionally, a more complex road network including arterials, collectors, and local roads was tied to elevated crash frequencies. These results emphasize how both ecological settings and roadway infrastructure influence wildlife\u0026ndash;vehicle conflict patterns.\u003c/p\u003e \u003cp\u003eHowever, Local Moran\u0026rsquo;s I and LISA cluster maps revealed spatial clustering in residuals (Moran\u0026rsquo;s I\u0026thinsp;=\u0026thinsp;0.136), indicating unmeasured local risk factors and the presence of spatial autocorrelation not fully explained by the NB model. To address this, a Spatial Error Model (SEM) was implemented, yielding a lower AIC (7090.7) compared to the NB model (7105) and producing more spatially consistent AVC predictions. Residual clustering was substantially reduced, confirming that the SEM effectively captured latent spatial dependencies and improved overall predictive accuracy.\u003c/p\u003e \u003cp\u003eThis study demonstrates several key strengths. It integrates narrative-derived species information, fine scale spatial modeling, and a diverse set of predictors, including static factors like land use and public lands, and dynamic ones like traffic volume and incident clearance times. The framework also leverages multiple advanced methods: a Negative Binomial model to address overdispersed crash counts, a Spatial Error Model (SEM) to account for spatial autocorrelation, and a natural language processing (NLP) pipeline to extract species information from crash narratives. To our knowledge, this is the first AVC study to combine NLP-based species inference with spatial econometric modeling at the census tract level. This integrated approach supports geographically tailored mitigation strategies and is scalable to other regions with census-based crash and infrastructure data, making it valuable for transportation planners, conservationists, and public agencies.\u003c/p\u003e \u003cp\u003eThe findings of this study offer important guidance for transportation planners, wildlife managers, and policymakers. By identifying how factors such as forest coverage, road classification, population density, and operational variables influence AVC risk, this research enables the development of targeted, location specific mitigation strategies. These might include dynamic signage in high-risk corridors, speed reductions in semi urban zones near wildlife attractants, or prioritization of fencing and wildlife crossings where road types and habitat patterns intersect. The tract level approach provides finer spatial resolution than most county-based models, allowing local agencies to allocate resources more effectively. Moreover, the integration of species-specific information through NLP opens new possibilities for tailoring mitigation to wildlife populations. As statewide crash databases and real time wildlife detection systems continue to evolve, this framework can be adapted into operational tools for forecasting risk and evaluating the effectiveness of implemented countermeasures.\u003c/p\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eLIMITATIONS AND FUTURE DIRECTIONS\u003c/h2\u003e \u003cp\u003eWhile this study provides a robust and multidimensional framework for analyzing terrestrial AVCs, several limitations highlight areas for future improvement and research.\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eSpatial Resolution and Aggregation Bias: The use of U.S. Census tracts as the spatial unit of analysis offers a practical approach, but it may obscure localized risks that occur at smaller scales, such as road segments or specific crossing zones. Future research could explore finer spatial resolution to reduce potential aggregation bias.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eSpecies Identification Uncertainty: Species information was inferred using a NLP applied to crash report narratives. Although this represents an innovative advancement, classification accuracy may vary, and future research should aim to validate these results or incorporate manually verified data when feasible.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eUnmeasured Confounding Factors: Despite adjusting for spatial autocorrelation using the Spatial Error Model (SEM), certain variables\u0026mdash;such as wildlife fencing, driver distraction, enforcement presence, or seasonal animal movement\u0026mdash;were unavailable. Incorporating these in future datasets could improve model performance and policy relevance.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eTemporal Dynamics: The current model is based on crash data aggregated from 2020 to 2024. It does not capture short-term fluctuations like migration periods, seasonal trends, or time-of-day patterns. Future efforts should include time-stamped data to assess temporal variability in AVC risk.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eModel Interpretability: While the SEM significantly improved fit, it does not capture direct spatial spillovers\u0026mdash;how risk in one tract may influence its neighbors. Future comparisons with models such as the Spatial Lag or Conditional Autoregressive (CAR) model could provide deeper insights into spatial interaction effects.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eMicro Scale Analysis Within Clusters: The tract level spatial analysis provided insight into regional AVC patterns, but further investigation at the road segment level, particularly within high-high, low-low, and high-low LISA clusters, could reveal site specific factors that contribute to localized crash risk. This would support more precise mitigation design.\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"Declarations","content":"\u003ch2\u003eAUTHOR CONTRIBUTIONS\u003c/h2\u003e \u003cp\u003eThe authors confirm contribution to the paper as follows: study conception, design, data preprocessing and primary writing: Xi Wei; NLP-based crash data preparation: Sudesh Bhagat; statistical modeling and result interpretation: Xi Wei, Sudesh Bhagat; manuscript structure and narrative refinement: Aparna Joshi; conceptual guidance and methodological recommendations: Anuj Sharma. All authors contributed to revisions and approved the final version of the manuscript.\u003c/p\u003e\u003ch2\u003eACKNOWLEDGMENTS\u003c/h2\u003e \u003cp\u003eThis manuscript was supported in part by language and structural editing using ChatGPT (OpenAI), which assisted with improving grammar, clarity, and academic tone. All content was reviewed and verified by the authors to ensure factual accuracy and alignment with TRB ethical guidelines.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eU.S. Department of Transportation, Federal Highway Administration (2008) Wildlife-vehicle collision reduction study: Report to Congress (Report No. FHWA-HRT-08-034). Retrieved May 18, 2025, from \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.fhwa.dot.gov/publications/research/safety/08034/\u003c/span\u003e\u003cspan address=\"https://www.fhwa.dot.gov/publications/research/safety/08034/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIowa Department of Transportation. (n.d.). Iowa Crash Analysis Tool [Data tool]. Retrieved May 18 (2025) from \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://icat.iowadot.gov/\u003c/span\u003e\u003cspan address=\"https://icat.iowadot.gov/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBartonička T, Andr\u0026aacute;šik R, Duľa M, Sedon\u0026iacute;k J, B\u0026iacute;l M (2018) Identification of local factors causing clustering of animal-vehicle collisions. J Wildl Manag 82(5):940\u0026ndash;947\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTaylor-Brown A, Booth R, Gillett A, Mealy E, Ogbourne SM, Polkinghorne A, Conroy GC (2019) The impact of human activities on Australian wildlife. PLoS ONE, 14(1), e0206958\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRahman MA, Das S, Codjoe J, Mitran E, Sun X, Abedi K, Hossain MM (2023) Applying Data Mining Methods to Explore Animal-Vehicle Crashes. Transp Res Record: J Transp Res Board, 2677(11)\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGregory A, Spence E, Beier P, Garding E (2021) Toward best management practices for ecological corridors. Land 10(2):140\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAndrews KM, Nanjappa P, Riley SP (eds) (2015) Roads and ecological infrastructure: concepts and applications for small animals\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTsuda LS, Carneiro CC, Quintanilha JA (2022) Using Self-Organizing Maps to find spatial relationships between wildlife-vehicle crashes and land use classes. An Acad Bras Cienc, 94(suppl 4), e20210727\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLao Y, Zhang G, Wu YJ, Wang Y (2011) Modeling animal\u0026ndash;vehicle collisions considering animal\u0026ndash;vehicle interactions. Accid Anal Prev 43(6):1991\u0026ndash;1998\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSullivan JM (2011) Trends and characteristics of animal-vehicle collisions in the United States. J Saf Res 42(1):9\u0026ndash;16\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSnow NP, Zhang Z, Finley AO, Rudolph BA, Porter WF, Williams DM, Winterstein SR (2018) Regional-based mitigation to reduce wildlife\u0026ndash;vehicle collisions. J Wildl Manag 82(4):756\u0026ndash;765\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHubbard MW, Danielson BJ, Schmitz RA (2000) Factors influencing the location of deer-vehicle accidents in Iowa. J Wildl Manage 64(3):707\u0026ndash;713\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGalinskaitė L, Ulevičius A, Valskys V, Samas A, Busher PE, Ignatavičius G (2022) The influence of landscape structure on wildlife\u0026ndash;vehicle collisions: Geostatistical analysis on hot spot and habitat proximity relations. ISPRS Int J Geo-Information 11(1):63\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLord D, Mannering F (2010) The statistical analysis of crash-frequency data: A review and assessment of methodological alternatives. Transp Res Part A: Policy Pract 44(5):291\u0026ndash;305\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGunson KE, Mountrakis G, Quackenbush LJ (2011) Spatial wildlife-vehicle collision models: A review of current work. J Environ Manage 92(4):1074\u0026ndash;1082\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuijser, M. P., McGowen, P., Fuller, J., Hardy, A., Kociolek, A., Clevenger, A. P.,\u0026hellip; Ament, R. (2009). Cost-benefit analyses of mitigation measures aimed at reducing collisions with large ungulates. Ecology and Society, 14(2), 15\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBissonette JA, Kassar CA (2008) Locations of deer-vehicle collisions are unrelated to traffic volume or posted speed limit. Human\u0026ndash;Wildlife Conflicts 2(1):122\u0026ndash;126\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAr\u0026eacute;valo JE, Honda W, Arce-Arias A, H\u0026auml;ger A (2017) Spatiotemporal variation of roadkills show mass mortality events for amphibians in a highly trafficked road adjacent to a national park, Costa Rica. Revista de Biolog\u0026iacute;a Trop 65(4):1261\u0026ndash;1276\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHaule HJ, Sando T, Lentz R, Chuan CH, Alluri P (2019) Evaluating the impact and clearance duration of freeway incidents. Int J Transp Sci Technol 8(1):13\u0026ndash;24\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eClevenger AP, Chruszcz B, Gunson KE (2003) Spatial patterns and factors influencing small vertebrate fauna road-kill aggregations. Biol Conserv 109(1):15\u0026ndash;26\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBissonette JA, Cramer PC (2008) Evaluation of the use and effectiveness of wildlife crossings, vol 615. Transportation Research Board\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWestern Transportation Institute (2021) Cost-effective solutions: Literature review on wildlife warning signs and animal-detection systems. Pooled Fund Study TPF-5(358). Retrieved from \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://westerntransportationinstitute.org/wp-content/uploads/2021/12/4w7576_Huijser_etal_WVC_ConnectivityLiteratureReview_PooledFundStudyFinalReport_2021.pdf\u003c/span\u003e\u003cspan address=\"https://westerntransportationinstitute.org/wp-content/uploads/2021/12/4w7576_Huijser_etal_WVC_ConnectivityLiteratureReview_PooledFundStudyFinalReport_2021.pdf\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLitvaitis JA, Tash JP (2008) An approach toward understanding wildlife-vehicle collisions. Environ Manage 42(4):688\u0026ndash;697\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLewis JS, Spaulding S, Swanson H, Keeley W, Gramza AR, VandeWoude S, Crooks KR (2021) Human activity influences wildlife populations and activity patterns: implications for spatial and temporal refuges. Ecosphere, 12(5), e03487\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen Y, Ma J, Wang S (2019) Spatial regression analysis of pedestrian crashes based on point-of-interest data. J Data Anal Inform Process 8(1):1\u0026ndash;19\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCensus US (2020) Bureau. Census Demographic and Housing Characteristics (DHC) Data: Table P1, Total Population by Census Tract, Iowa. data.census.gov. Accessed at May 2025. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://data.census.gov/map/040XX00US19$1400000/DECENNIALDHC2020/P1?q=Iowa\u0026amp;y=2020\u0026amp;d=DEC+Demographic+and+Housing+Characteristics\u0026amp;layer=VT_2020_140_00_PY_D1\u0026amp;loc=41.9575,-93.8730,z6.0000\u003c/span\u003e\u003cspan address=\"https://data.census.gov/map/040XX00US19$1400000/DECENNIALDHC2020/P1?q=Iowa\u0026amp;y=2020\u0026amp;d=DEC+Demographic+and+Housing+Characteristics\u0026amp;layer=VT_2020_140_00_PY_D1\u0026amp;loc=41.9575,-93.8730,z6.0000\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCensus Bureau US, TIGER/Line Shapefile (2020) Iowa Census Tracts. Accessed at May 2025. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www2.census.gov/geo/tiger/TIGER2020/TRACT/tl_2020_19_tract.zip\u003c/span\u003e\u003cspan address=\"https://www2.census.gov/geo/tiger/TIGER2020/TRACT/tl_2020_19_tract.zip\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIowa Department of Transportation. Road Network Portal. Iowa Department of Transportation. Accessed at (2025) May \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://data.iowadot.gov/datasets/road-network-portal/\u003c/span\u003e\u003cspan address=\"https://data.iowadot.gov/datasets/road-network-portal/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eState of Iowa. Crash Data Dashboard: Vehicle Crashes in Iowa. Iowa Open Data Portal. Accessed at (2025) May \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://data.iowa.gov/stories/s/m73g-tzgc\u003c/span\u003e\u003cspan address=\"https://data.iowa.gov/stories/s/m73g-tzgc\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCommission for Environmental Cooperation. Land Cover 30m \u0026ndash; 2020. North American Environmental Atlas (2021) Accessed at May 2025. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.cec.org/north-american-environmental-atlas/land-cover-30m-2020/\u003c/span\u003e\u003cspan address=\"https://www.cec.org/north-american-environmental-atlas/land-cover-30m-2020/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIowa Department of Natural Resources. Public Lands Used for Conservation and Recreation in Iowa. Iowa Geodata Portal. Accessed at (2025) May \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://geodata.iowa.gov/datasets/2e8be3c67288413f903276eea17cea43_0\u003c/span\u003e\u003cspan address=\"https://geodata.iowa.gov/datasets/2e8be3c67288413f903276eea17cea43_0\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDeepchecks. Ollama LLM, Tool Integration (2024) Guide Accessed at May 2025. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.deepchecks.com/llm-tools/ollama/\u003c/span\u003e\u003cspan address=\"https://www.deepchecks.com/llm-tools/ollama/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMeta (2025) Meta-Llama-3-8B-Instruct. Hugging Face. Accessed at May \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.deepchecks.com/llm-tools/ollama/\u003c/span\u003e\u003cspan address=\"https://www.deepchecks.com/llm-tools/ollama/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"Iowa State University","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Animal-vehicle collisions, wildlife mitigation, spatial analysis, Natural language processing","lastPublishedDoi":"10.21203/rs.3.rs-8398693/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8398693/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eAnimal\u0026ndash;vehicle collisions (AVCs) present serious safety, economic, and ecological concerns, particularly in areas where wildlife habitats intersect with expanding road networks. While previous studies have examined land cover or traffic volume in relation to AVCs, many relied on coarse spatial scales, treated species as a single group, and omitted key roadway or operational variables. This study addresses these gaps by investigating the spatial determinants of terrestrial AVCs across Iowa using an integrated ecological and infrastructural modeling framework. Crash data from 2020 to 2024 were analyzed using a natural language processing (NLP) model to extract animal-related incidents and infer species types from narrative reports. Land cover, road characteristics (e.g., classification, AADT, speed limits), operational measures (e.g., incident clearance time), and population data were aggregated to Iowa\u0026rsquo;s 896 census tracts (2020 boundaries). A Negative Binomial (NB) regression model identified forest coverage, population density, speed limits, and road network complexity (e.g., length of Interstates and arterials) as significant predictors of higher AVC risk, while urban land cover was negatively associated. Stepwise AIC selection and 10-fold cross-validation enhanced model performance. Spatial clustering in residuals (local Moran\u0026rsquo;s I\u0026thinsp;=\u0026thinsp;0.42, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01) led to a Spatial Error Model (SEM), which further improved fit and reduced spatial bias. This tract-level framework offers a scalable, data-driven approach for transportation and wildlife agencies to identify high-risk areas and implement localized mitigation strategies, such as fencing, signage, or speed control, tailored to both ecological and infrastructural conditions.\u003c/p\u003e","manuscriptTitle":"Spatial Analysis and Prediction of Animal Vehicle Collisions (AVCs) in Iowa Using Environmental and Anthropogenic Factors","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-12-22 06:21:42","doi":"10.21203/rs.3.rs-8398693/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"b8047f08-1085-4f9d-87b5-974d73a2d731","owner":[],"postedDate":"December 22nd, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":60041575,"name":"Civil Engineering"},{"id":60041576,"name":"Artificial Intelligence and Machine Learning"}],"tags":[],"updatedAt":"2025-12-22T06:21:42+00:00","versionOfRecord":[],"versionCreatedAt":"2025-12-22 06:21:42","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8398693","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8398693","identity":"rs-8398693","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}