NatureScot Research Report 1400 - Deer Vehicle Collision Analysis 2022–2024

Published: 2026 

Authors: Lush, M.J. and Lush, C.E.

Cite as: Lush, M.J. and Lush, C.E. Deer Vehicle Collision Analysis 2022–2024. NatureScot Research Report 1400.

Keywords

Deer vehicle collision; wildlife vehicle collision; wildlife management; road management; road networks; risk mapping.

Non-technical summary

Overview

Deer numbers across Scotland have grown over recent decades, potentially bringing deer into closer contact with people and vehicles. This study updates Scotland’s long-running monitoring of Deer Vehicle Collisions (DVCs), using over 43,000 mapped incidents recorded between 2008 and 2024. The aim is to understand where and when collisions happen most often, so steps can be taken to reduce the risks to both people and wildlife.

What’s changing?

• Overall numbers steady: The total number of reported DVCs on trunk roads – major routes managed at a national level rather than by local authorities – has not increased since 2015, suggesting the overall risk at a national scale has remained stable.
• Shifting locations: The number of collisions is increasing in the Central Belt and around Aberdeen, but is decreasing in the north-west.
• Hotspots: The 10 locations with the most recorded DVCs in 2022–2024 were all in the Central Belt, mainly around junctions with small areas of woodland. Only one of these ‘hotspots’ matched a previous top-ten hotspot from 2019–2021, showing that high-risk locations can change.
• Human injuries: Incidents involving human injury and/or vehicle damage have declined since 2008, perhaps partly reflecting changes in reporting.

When and where collisions happen

• Time of day: Most collisions occur at dusk, extending into the late evening in winter.
• Time of year: May and June are the peak months for DVCs, especially on motorways. Autumn and early winter peaks also occur in some areas.
• Regions most affected: The highest numbers of DVCs are in Highland, Fife, Perth & Kinross and Aberdeenshire, but when the total length of the roads within a region is taken into account, Falkirk, North Lanarkshire and East Lothian show the highest collision rates.

Which deer species are involved?

• Roe deer are involved in the most collisions nationwide. Evidence suggests female roe deer are slightly more often involved than males.
• Red deer dominate the number of collisions in Highland and parts of western Scotland.

Other wildlife on roads

Deer are not the only animals affected. Reports also included:

• Badgers and foxes – frequently recorded victims.
• Otters, pine martens and red squirrels – occasionally killed on roads, showing impacts on rare and endangered species.

How the findings help

The information supports road managers, conservationists and local authorities in:

• Identifying high-risk road sections for possible signage, roadside habitat management or other mitigation measures.
• Improving data collection, including reports from the public through mobile apps.
• Developing future models to predict collision risk using traffic, habitat and road features.

Primary data suppliers

• Amey
• Autolink Concessionaires (M6) Plc.
• Balfour Beatty
• BEAR Scotland
• British Deer Society
• Forestry and Land Scotland
• Mammal Society
• Police Scotland
• Scottish Society for the Prevention of Cruelty to Animals
• Transport Scotland

Background

In recent decades, the number of deer in Scotland has increased and deer populations have expanded their range. At the same time, traffic volumes have also risen, potentially increasing the risk of collisions between deer and vehicles.

This report continues work begun in 2003 to analyse Deer Vehicle Collisions (DVCs) in Scotland. The purpose of this work is to track changing patterns in the frequency and distribution of DVCs on the trunk road network, and to inform actions aimed at reducing future collisions, particularly on stretches of road where incidents occur most often.

For this contract, data covering 2022–2024 have been added to the NatureScot DVC database, together with some earlier records that were previously unavailable. The database now contains 40,179 mapped incidents dated from 2008 to 2024. Around 65% of these come from four main sources: the SSPCA, Trunk Road Operating Companies, Forestry and Land Scotland rangers, and human-injury collisions recorded by Police Scotland. An additional 26% of records were drawn from Police Scotland’s STORM database, incorporated during this contract and covering the years 2020–2024.

The full dataset has been analysed, broadly replicating methods used in earlier studies while also introducing enhancements to improve the identification of trends and to refine the approach to DVC risk mapping.

Main findings

• Accounting for new data, there is limited evidence for an increase in the number of DVCs reported on trunk roads per annum between 2015 and 2024. This suggests that the risk of DVCs on the Scottish trunk road network as a whole is not increasing.
• However, the distribution of DVCs across the road network appears to have changed. Numbers of DVCs were increasing in the central belt across all roads and around Aberdeen for non-trunk roads. Fewer DVCs were reported elsewhere, especially in the wider road network in the north west.
• The largest number of DVC incidents occur in northern local authority areas in Scotland, notably Highland, Fife, Perth and Kinross and Aberdeenshire. However, when expressed as the number of DVCs per length of road within a local authority area, the largest number of incidents occur in the central belt, notably Falkirk, North Lanarkshire and East Lothian.
• May and June remain the peak months for DVCs. This was most notable on motorways and became less pronounced as roads became more minor. This may reflect levels of activity for roe deer only; the peak months for DVCs on Lewis and Harris, where only red deer are present, appear to be September to December. At Navidale, north Scotland, where red deer predominate, DVCs peak in September to November.
• Dusk remains the peak time for DVCs, though this peak is spread over a longer period from mid-afternoon until midnight in winter. This is likely due to high levels of both traffic and deer activity, plus factors such as driving conditions and driver fatigue.
• Roe deer remain the most frequently implicated species in DVC incidents and there is evidence to suggest that DVCs are more likely to involve female than male roe deer. However, red deer were most frequently implicated in Highland and North Ayrshire, whilst fallow deer were most frequently implicated in Dumfries and Galloway.
• The number of DVCs resulting in human injury, especially slight injuries, or vehicle damage appears to have declined from 2008–2024. However, this may be due to incomplete data or a change to the threshold for what constitutes a slight injury.
• Deer escape uninjured in the majority of reported DVCs resulting in human injury. The proportion of these DVCs that represent drivers avoiding the deer but having some sort of incident in the process and those using deer as an excuse for an incident with some other underlying cause is not known, though both are thought to occur.
• The number of DVC records made by the public has increased over time, reflecting the increased availability and use of recording tools such as the Mammal Society app, iRecord and iNaturalist.
• Large numbers of badger and red fox are also implicated in Wildlife Vehicle Collisions (WVCs), though it is not currently possible to estimate the total numbers. WVCs also account for mortality in rare and endangered species, including otters, pine martens and red squirrels.
• Most DVC hotspots between 2022 to 2024 were in the central belt.
• Ten DVC hotspots were shortlisted. All were within the central belt and around road junctions with adjacent small areas of woodland. A combination of deer using the woodland as daytime or birthing retreats, intermittent, fast-moving traffic on slip roads and reduced visibility on curved slip roads is believed to be responsible.
• Only one of the top ten DVC hotspots identified in 2019–2021 was included in the top ten hotspots in 2022–2024. This may reflect fluctuations in local deer numbers and suggests that the analysis needs to encompass a greater number of years. This would require DVC data with largely unbiased positional accuracy over multiple years.
• Mapping DVC hotspots using Kernel Density Estimation is recommended for a future study. This may be complementary to the hotspot mapping employed here, as it would provide additional evidence.
• Attempts to determine the relationship between DVC hotspots and physical and environmental factors failed, but suggested that a reliable model could be produced with more years of data. It is recommended that this work is repeated using reliable data from future years.
• Despite the failure of the model to converge, the results supported suggestions that DVCs are more likely to occur around road junctions, and close to woodland, moorland and moderately urban areas. Higher vehicle speeds and larger volumes of traffic also appeared to be related to a high incidence of DVCs.
• The DVC database should be improved by developing functionality that differentiates between records relating to collisions between deer and vehicles, and those representing uninjured deer on roads. This ensures that all records can be used to assess risk whilst allowing the identification of DVCs.
• Animal rescue centres other than the SSPCA, which already provides data, are recognised as a valuable potential source of DVC data. Recommendations are made regarding the collection and mobilisation of this data.

Acknowledgements

We thank the Project Steering Group members for their advice and support, and for funding this work: Dominic Sargent (NatureScot) and Angus Corby (Transport Scotland). Dr. Jochen Langbein is also thanked for useful conversations during this contract.

Thanks are also due to those that have regularly supplied data for use in this project from a range of organisations, primarily Kristoffer Thorbjornsen (Amey); Jock Laidlaw and Neil Brannock (Autolink Concessionaires (M6) Plc.); Carla Cummins (Galliford Try Investments); Stewart MacKenzie (Balfour Beatty); Boris Boum, David Patton, Sheila Thomson and Tommy Deans (BEAR Scotland); Laura Boyle (SSPCA); Sheila Baxter (Forestry and Land Scotland); Ross Clifton (Mammal Society). Additional data was also received from Hannah Crow (West Lothian Council), Ben Pacholek (Police Scotland), Mark Jamieson (NatureScot), Paul Reynolds (New Arc Wildlife Rescue) and Mark Scott (Transport Scotland).

Thanks also to all those that have uploaded Wildlife Vehicle Collision records to the NBN Gateway, often via iRecord, and iNaturalist. Organisations providing data via the NBN Gateway included Argyll Biological Records Centre, Highland Biological Recording Group, National Trust for Scotland and The Wildlife Information Centre. Particular thanks to those iNaturalist recorders that provided unobscured coordinates for several records, thus allowing these additional records to be included in the analysis – especially users sugarsnap_t and Mike G. Rutherford.

Abbreviations

Aberdeen Western Peripheral Route (AWPR)

Asset Management Performance System (AMPS)

Average Annual Daily Flow (AADF)

Deer Vehicle Collision (DVC)

Design-Build-Finance-Operate (DBFO)

Collision Reporting and Sharing system (CRaSH)

Forestry and Land Scotland (FLS)

Integrated Road Information System (IRIS)

Kernel Density Estimation (KDE)

Locally Estimated Scatterplot Smoothing (LOESS)

Road Traffic Accident (RTA)

Road Traffic Collision (RTC)

Scottish Society for the Prevention of Cruelty to Animals (SSPCA)

Trunk Road Operating Company (TROC)

Wildlife Vehicle Collision (WVC)

Introduction

Wild deer numbers in Scotland have risen substantially in recent decades. Although all species have expanded their ranges, roe deer have become especially well established across lowland Scotland, and deer are now increasingly present in urban areas and throughout the central belt (Pepper, Barbour and Glass, 2019).

These increases in deer populations and distribution have coincided with growth in road traffic, potentially leading to more Deer Vehicle Collisions (DVCs). Langbein (2019) defines DVCs as “any incidents where it may be concluded that a collision between a road vehicle and a deer has occurred; as evident either from live injured or dead deer casualties found at the roadside, or from reported road traffic collisions in which deer were implicated as an object or hazard in the carriageway (e.g. deer colliding with road vehicle, or deer presence causing drivers to swerve).”

This report presents findings from the sixth contract analysing the impact of DVCs on Scotland’s trunk roads, extending the dataset to cover the years 2022–2024. The database now includes records from 2008 onwards. Earlier contracts were reported by Langbein & Putman (2006), Langbein (2011; 2013; 2017; 2019), and Lush and Lush (2023). Langbein (2019) provides a consolidated summary of work undertaken between 2004 and 2018.

The current contract focused on updating and extending the dataset compiled in earlier studies, repeating analyses to enable comparison across time, and developing new analyses to provide additional insight. The results are intended to help guide measures to reduce the number and impact of DVCs in Scotland.

Data were sourced from the same organisations that contributed to previous contracts, primarily the Trunk Road Operating Companies (TROCs), supplemented by a range of other organisations that routinely deal with DVCs, as well as reports from the public submitted through various channels. Changes to participating organisations and available data are documented in this report.

This contract has also further refined methods of data collation, processing, and analysis, with the aim of improving consistency and efficiency and minimising the risk of human error. 

Data collection regime, processing and analysis

Records of DVC incidents between 2008 and 2021 had been collated as part of previous contracts (Langbein, 2011; 2013; 2017; 2019; Lush and Lush, 2023). Records for 2022 to 2024, as well as some for earlier years that had not been previously collated, were obtained from source organisations, as described below.

As before, data were collated into a PostgreSQL/PostGIS database. The structure of the data was rationalised during this contract to make it more relational and consistent, and therefore easier to analyse (Annex 1: Database main table structure).

Changes to the database included enhancing its ability to accommodate Wildlife Vehicle Collisions (WVCs) involving species other than deer. Efforts were made to collate such data for 2022 to 2024, and records from data received between 2019 and 2021 were reviewed to incorporate additional species into the database. This may have importance in the future if, for example, collisions with large animals such as feral pigs become more common. A fuller understanding of WVC numbers also helps to put DVC numbers in context, where recording is relatively consistent.

Records were retained in their original formats and as separate processed PostgreSQL tables, so that data provenance could be easily traced. PostgreSQL scripts used to process the data, both from original sources into usable spatial database tables and into the master table, were retained, allowing full traceability and enabling changes to be quickly implemented if errors were found. Any manual processing undertaken was carefully documented.

Core data sources

Trunk Road Operating Companies

Data on reported DVC incidents since 2022 were supplied by the Trunk Road Operating Companies (TROCs) and added to the existing data for 2008–2021. This included data from the four main trunk road units and the smaller Design-Build-Finance-Operate (DBFO) sections. Data from the M8 DBFO included DVC incidents from August 2020 to December 2021, which were missing in the previous analysis (Lush and Lush, 2023). All TROCs supplied data for all years covered by this study and these were incorporated into the DVC database.

Efforts were made to obtain records of reports that did not result in a deer carcass uplift (Lush and Lush, 2023) and those relating to other animals. Some TROCs supplied data for other species, but in many cases it was not possible to know whether supplied DVC data included ‘no trace’ records.

Figure 4 represents the trunk road network as of December 2021. However, the trunk road network is evolving and since December 2021 the following modifications have occurred:

• A9 Luncarty to Birnham – upgrade with minimal impact on road layout.
• A9 Cross Tay – realignment of the road by up to 170 m.
• A85 Ruthvenfield – addition of several new slip roads parallel to the A9.
• A9 Berridale – minor realignment of less than 15 m.
• M8 Inchinen – addition of two new slip roads onto the A8.
• M9 Winchburgh – addition of four new slip roads onto the B8020.
• A77 Maybole bypass – realignment of the road by up to one kilometre.

Of these, only the A77 Maybole bypass was considered likely to have the potential for significantly impacting DVC analysis, but it was situated in an area where numbers of DVCs were few. Analysis of all DVC records using the 2021 trunk road network was therefore considered unlikely to affect the results significantly.

Scottish Society for the Prevention of Cruelty to Animals

The SSPCA has supplied records of incidents reported by various bodies and sometimes attended by staff since 2008. Incidents involving DVCs up to and including 2024 were provided by them and integrated into the DVC database. Reports typically included a postcode and were geolocated using Ordnance Survey Code-Point data.

Forestry and Land Scotland Wildlife Rangers

Forestry and Land Scotland (FLS; formerly Forestry Commission Scotland) has provided extracts from its national cull database for records classified as Road Traffic Accidents (RTA) or Road Traffic Collisions (RTC) from 2008 onwards. Records up to and including 2024 have been incorporated into the DVC database.

Each entry in the FLS dataset represents an individual deer. Consequently, where multiple DVCs occurred at the same time and location, an equivalent number of records are present. In other datasets that distinguish sex and age class, the total number of deer is derived from the sum of males, females, and juveniles. Forestry and Land Scotland is unique among the core data providers in recording the sex of juvenile deer. Because each FLS record represents a single animal, values of one are entered into the relevant sex, juvenile, and total deer fields within the database. Accordingly, unlike other sources, the total count in FLS records is not derived from the aggregation of sex and age-class fields.

Police and Road Safety Teams’ RTC records

The primary source of information on incidents involving DVCs from the police and road safety teams since 2008 has been STATS19 accident reporting forms. STATS19 is Great Britain’s official road traffic casualty database. Complete or near complete STATS19 data was included in the DVC database for 2008–2017 inclusive. These data subsequently became unavailable, due to changes to the data structure supplied.

In 2024, Police Scotland supplied an extract from their CRaSH database covering January 2020 to September 2024 (see Police Scotland). Whilst these data contained only an accident reference, date and description, and were therefore not incorporated into the database, the accident reference could be cross-referenced with related STATS19 records downloaded from the Department for Transport (2025). Because it was possible to identify which CRaSH records involved deer, these were linked to the relevant STATS19 record. It is not clear whether all relevant STATS19 records were identified this way, but it was more comprehensive than could be achieved in the previous contract (Lush and Lush, 2023).

Combined data were added to the DVC database, incorporating the description from CRaSH and other data from STATS19 where the accident references matched. This included adding records to the existing incomplete data for 2020 and 2021.

It was therefore considered unnecessary to repeat the Freedom of Information requests reported in Lush and Lush (2023), as these were time consuming and resulted in little data.

Note that there is often a lag between STATS19 data being collected and it being made available to other users. STATS19 data for 2024 was unavailable at the time of writing, so only data from 2020 to 2023 could be incorporated. 2024 data will need to be incorporated in a future project.

Supplementary data sources

Mammal Society

Data from the Mammal Society’s Mammal Mapper and Mammal Tracker schemes have been available since 2016. The Mammal Tracker scheme was phased out and replaced by the Mammal Mapper scheme in 2019. Data from both schemes up to and including 2024 were incorporated into the DVC database. These usually contribute between 10 and 50 records relating to DVC incidents in Scotland per year (Table 1).

Data from the National Mammal Atlas Project were also downloaded from the NBN Atlas (see NBN Atlas). Those records relating to DVCs were extracted and added to the database. These dated from 2013 to 2024 and contributed between zero and five DVC records in Scotland per year (Table 1). Any duplicates between this and the Mammal Mapper and Mammal Tracker datasets were flagged and removed from the analysis.

British Deer Society

From 2009 members of the public were able to report DVCs online via the DeerAware website. This generally involved fewer than 50 records for Scotland per year. The data were very poorly structured and required significant manual processing to make them usable. The British Deer Society took control of the DeerAware website in 2020 and started collating the online reports in a structured format.

Regular recorders had been encouraged to use the British Deer Society app instead, after its launch in 2019. This collected DVC data in a more structured format than DeerAware, which allowed for easier collation and analysis. However, the app was unavailable and undergoing an extensive rebuild at the time of writing.

The British Deer Society provided data for both DeerAware and the app up to February 2021, which had been incorporated into the DVC database (Lush and Lush, 2023). They have been unable to process or provide more recent data since 2021, which means no data were collated during this contract.

Animal rescue records

The database included two records from an animal rescue centre other than the SSPCA from 2010. During this contract, 157 records provided by New Arc Wildlife Rescue were incorporated into the DVC database, covering 2023 and 2024. These comprised the locations and details of call-outs related to DVCs, mainly covering the area around Aberdeen (Figure 5). Unlike many other data sources, these records were considered to have been accurately identified to species level.

Road Cleansing Departments

Langbein and Putman (2006) found that only a minority of local authority road cleansing departments would be able to provide useable records of DVCs. Because of this, and as in previous projects (Langbein, 2017; 2019; Lush and Lush, 2023), data from local authority road cleansing departments was not actively sought. All such data collated between 2008 and 2018 was included in the database, to which one record was added in the previous analysis that was included in a local authority response to the request for STATS19 data (Lush and Lush, 2023).

West Lothian provided additional data from their road cleansing department covering 16 February 2004 to 27 August 2024. These data included a date and a description of the location, and coordinates for some records since 2017 (Table 2). The location descriptions were inadequate for determining the precise location of those records without coordinates.

The West Lothian data included records of a range of species, of which deer were the most frequently reported (Table 3). Whilst all were incorporated into the DVC database, records of non-deer species and those without coordinates were excluded from the analysis, where appropriate.

Police Scotland

Police Scotland provided records containing the word “deer” from their STORM and CRaSH databases.

STORM Unity is the national command and control incident recording system used by Police Scotland. It therefore contains logs of incidents reported by various bodies, including those involving DVCs. The supplied data included coordinates, a date and time, and a short description of the incident. The extract covered the period from 1 January 2019 to 25 November 2024. The time related to the time the call was received, not the time of the incident.

Several records had coordinates that could not be used and a few records referred to deer but were unrelated to DVCs, such as “John Deere” or “Deer Park”, and were excluded. Nevertheless, STORM records contributed a huge number of records spanning the years 2019 and 2024 (Table 4).

The Collision Recording and SHaring system (CRaSH) is used to record data on road traffic collisions attended by Police Scotland (Figure 6). The supplied CRaSH data contained no location information and was not incorporated into the database. However, it was cross-referenced with STATS19 data to identify those records related to DVCs (Police and Road Safety Teams’ RTC records).

Transport Scotland AMPS database

Transport Scotland provided records of ‘animal defects’ from their Asset Management Performance System (AMPS) database. AMPS is a centralised asset management system that was gradually rolled out to the TROCs whilst this study was underway. Some data had also been imported into AMPS from the previous IRIS database. The AMPS ‘animal defects’ data supplemented the data received from the TROCs, as it contained additional records.

The supplied data included point geometries, a date and a description of the defect, which often included the name of the animal involved. The extract covered the period from 3 April 2022 to 30 April 2024, with an outlying record from 21 September 2020.

The records included species other than deer (Table 5). Whilst all were incorporated into the DVC database, records of non-deer species were excluded from the analysis, where appropriate.

NatureScot

NatureScot provided 40 records of DVCs that had been casually recorded by staff in 2023 and 2024. Species identification was considered to be accurate and records were flagged where the incident had been witnessed, meaning that the time was reliable.

NatureScot also provided a summary of DVCs made by the Lewis and Harris Deer Management Group in 2023 and 2024. Since the data was aggregated per month and lacked accurate locations it was not incorporated into the DVC database.

NBN Atlas

All records relating to deer that were licensed for commercial use were downloaded from the NBN Atlas. Those records that related to DVCs were extracted and incorporated into the DVC database. Some of these came from the National Mammal Atlas Project (see Mammal Society). The remaining relevant records came from four datasets and spanned the years 1992 to 2023 (Table 6). All pre-2008 records were retained in the database, but were not included in this analysis.

iNaturalist records

The GLOBAL Roadkill Observations project gathers records of WVCs crowdsourced using the iNaturalist platform. Wider research on iNaturalist records found this project to be comprehensive. All data from Scotland was downloaded and filtered for those licensed for commercial use, which reduced the number of usable records from 153 to 98. The coordinates on some records were obscured and could not be used, further reducing the number of usable records. A few prolific recorders were asked to change the settings on the records so that they could be used, resulting in 83 usable records.

These records were split into research and non-research grades. All records start as non-research grade. Records become research grade if they meet validity criteria and two thirds of users agree on the identification. Research grade records do not necessarily meet the verification standards of records on the NBN Atlas, but it is nevertheless a useful indicator of confidence in the species identification.

The records dated from 2018 to 2025 (Table 7) and included species other than deer (Table 8). Whilst all were incorporated into the DVC database, those from 2025 were not included in this analysis. Records of non-deer species were excluded from the analysis, where appropriate. The five records of deer shown in Table 8 dated from 2020 to 2023.

Supporting data

Several national-level datasets were collated to underpin the analysis and provide additional context:

• Integrated Road Information System (IRIS): An extract of the IRIS database, including Average Annual Daily Flow (AADF) figures, was supplied by Transport Scotland on 16 December 2021 and was considered current at the time of writing. Trunk road sections (Figure 4) were identified by selecting records where the OWNER_UID field was not equal to LOCAL AUTHORITY and the SECTION_CO field did not begin with 14935. The excluded codes (14935) corresponded to parts of the A737 that were removed from the live network on 7 April 2020 (Angus Corby, pers. comm.).
• Asset Management Performance System (AMPS): Transport Scotland also supplied an extract of various trunk road asset data from AMPS. This included the locations of basic, matrix and variable message signs, barriers and habitats.
• Ordnance Survey Open Roads: Data downloaded in October 2021 provided full coverage of the road network, enabling analysis of DVC frequencies across both trunk and non-trunk roads, as well as changes over time.
• Ordnance Survey Code Point: These data were obtained to support georeferencing of records supplied only as postcodes. Updates were made annually to ensure all new records could be incorporated into the database.
• Marker post data: Information released through a Freedom of Information request (https://www.gov.scot/publications/foi-19-01392/) had been obtained during the previous contract (Lush and Lush, 2023). The dataset was incomplete and contained some errors, but no comprehensive alternative existed. The missing marker post locations were supplemented and errors in the data were corrected manually.
• Habitat and Land Cover Map 2022: Data downloaded in July 2023 provided broad habitat data for the whole of Scotland.
• Deer Management Priority Areas: NatureScot provided boundaries for ten Deer Management Priority Areas in Scotland.

NatureScot also provided vehicle speed data that had been collected by BEAR Scotland for the A9 at Navidale, which was previously identified as a DVC hotspot (Lush and Lush, 2023). Two mobile Variable Message Signs (VMS) warning about the risk of deer were deployed at this location from 1–25 November 2024. The speed data covered the period 25 October to 18 December 2024, and therefore included a few weeks before and after deployment of the VMS. The data were affected by the presence of snow on the road from 16–23 November 2024.

Identification of duplicate records

A semi-automated process was used to identify potential duplicate DVC incident records. The procedure compared records from the same date for grid references within 100 m of each other, excluding any records that had already been flagged as duplicates in earlier checks.

Where species information was available, this was also taken into account: records listing different deer species – or cases where species was not identified – were generally treated as separate incidents. It is acknowledged that in some situations a single incident may have been entered multiple times with different species recorded.

The process generated a set of candidate duplicate pairs which were then reviewed manually. For each pair, the record judged to contain the least detail was typically marked as the duplicate. In cases where duplication could not be confidently established, both records were retained in the database without amendment.

Analysis

Analysis was undertaken using a combination of standard and bespoke PostgreSQL/PostGIS functions, and the R statistical environment. Maps were produced using QGIS, drawing upon the data in PostgreSQL, either directly or using PostgreSQL views.

All charts, tables and statistical tests were produced or undertaken in R v.4.4.2, drawing directly upon PostgreSQL using the packages DBI v.1.2.3 and RPostgres v.1.4.7 (R Special Interest Group on Databases (R-SIG-DB), Wickham and Müller, 2024; Wickham, Ooms and Müller, 2025). Data reshaping in R was undertaken using core R functions and the packages tidyverse v.2.0.0 and janitor v.2.2.1 (Wickham, 2023; Firke, 2024). Most charts were produced using the packages ggplot2 v.3.5.2, patchwork v.1.3.1, scales v1.4.0 and ggpubr v.0.6.0 (Kassambara, 2023; Pedersen, 2025; Wickham et al., 2025; Wickham, Pedersen and Seidel, 2025). Reshaped, publication ready data tables were copied from R using clipr v.0.8.0 (Lincoln, 2022).

A mixed-effects regression was used to analyse the Navidale speed data and deer sex ratios, using the R package lmerTest 3.1-3 to estimate p-values using Satterthwaite's method or the Laplace Approximation (Kuznetsova, Bruun Brockhoff and Haubo Bojesen Christensen, 2020). For the Navidale analysis, VMS deployment, time of day and vehicle class were treated as fixed effects, whilst road direction and day were treated as random effects. Snow during the survey period was also treated as a fixed effect. For the sex ratio analysis, sex was treated as a fixed effect and year was treated as a random effect.

Generalized linear regression was used to model the relationship between DVC hotspots and physical and environmental factors, using the R package MASS 7.3-61 for the regression, and car 3.1-3 and vcd 1.4-13 for model diagnostics (Fox, Weisberg and Price, 2024; Meyer et al., 2024; Ripley and Venables, 2024)

Results

Change in volume of data by source, type and years

A total of 11,789 records of DVCs were collated for the period 2022 to 2024. Of these, 71 had no or insufficient location information to map, and in many cases could not be analysed. A further 1,088 were thought to be potential duplicate records and, where relevant, were excluded from the analysis. 

In addition, 5,488 non-duplicate records with sufficient location information were added for previous years, mostly between 2019 and 2021. These largely came from extracts from Police Scotland’s STORM database, which contributed 5,414 records for this period. 2021 data for the M8 DBFO, which was not available for the previous study (Lush and Lush, 2023), added 40 records. A further 22 records from STATS19 accident reporting were also added for 2020–2021. Other pre-2022 records came from various sources, such as data extracted from iNaturalist and the NBN Atlas.

The change in volume of DVC records received from core and non-core sources since the initiation of these studies in 2003 is shown in Table 9, and since 2008 in Figure 7. The dramatic increase in the available records from 2019 onwards was largely due to extracts from Police Scotland’s STORM database and Transport Scotland’s AMPS database. These were both regarded as non-core sources, to allow for the analysis of a consistent set of data covering 2008 to 2024.

Volume and trends of DVC records from core sources across years

The number of records collated from core and other sources has increased since 2008 (Table 10). STATS19 road traffic collision data are likely incomplete from 2018 onwards and at the time of writing were not available for 2024, as described in Police and Road Safety Teams’ RTC records. Note that records extracted from Transport Scotland’s AMPS database are considered to be non-core, as they are not consistently available across all years and regions.

The number of DVCs reported from core sources has remained fairly constant since 2015. The constancy was due to relatively stable combined totals from Trunk Road Operating Companies (TROCs), with 2023 showing an above average number of reports from TROCs (Figure 8). Nearly all of the variation in the number DVCs reported by core sources comes from SSPCA data. This has masked the changes to incidents reported by FLS described in Changes in the volume of SSPCA and FLS records and certain trunk road regions, plus a significant strong decline in DVCs from road traffic collision reporting described in Changes to the number of human injury collisions and damage-only DVCs attended by .

Volume and trends of DVC records from supplementary data sources across years

Non-core data and accurately identified deer records have come from a range of sources that have varied since 2008. The categorisation of Table 11 is different from the equivalent table in Lush and Lush (2023), in that the “other deer experts” and “general public” has been replaced with terms that reflect whether the record has undergone some form of expert verification. Additionally, “council uplifts” has been replaced with a broader term encompassing all road carcass uplifts, to accommodate data for the non-trunk roads received from TROCs.

Up until March 2011, data were available for a special project commissioned by the Deer Commission for Scotland (DCS; Langbein, 2011), but such data have not been available for subsequent studies. Recent road uplift data comes mainly from West Lothian Council and Amey supplying data for a non-trunk road contract. Data from Police Control since 2019 has come largely from STORM data.

Data from deer experts other than the DCS project and FLS (Figure 15) has been supplemented with reliable data from New Arc Wildlife Rescue and NatureScot, and verified data from the Mammal Society. Verified data from the NBN Atlas and Research Grade records from iNaturalist have increased the number of pre-2022 DVC records for some years, as well as adding records for 2023–2024.

Unverified records from the public were supplemented with data from the NBN Atlas and non-research-grade observations from iNaturalist. Despite the loss of DeerAware records after February 2021 – which only contributed a few records annually – verified and unverified data from the public shows a strong increasing trend (Spearman’s rank correlation, r = 0.66, p < 0.01), supported by contributions of unverified records from other sources.

Overview of Scotland-wide distribution of DVC records from core sources

The distribution of DVC records within Scotland from core sources from 2022 to 2024 is shown in Figure 9.

Data from the TROCs reflects the trunk road network, though the density of records is clearly higher along some parts of the network than others. Fewer records from TROCs were located outside the network than reported in Lush and Lush (2023), suggesting an improvement in data collection methods.

SSPCA records cover the entire road network, but their spatial distribution closely resembles that of TROC data. This reflects underlying factors such as traffic density, human and deer population distribution, and the locations of wildlife rescue centres. SSPCA data also provides relatively high record densities in urbanised areas of the central belt and Aberdeenshire. Forestry and Land Scotland records, while also present on the road network, include forest tracks and more remote locations, with clusters around FLS-managed sites. In contrast, STATS19 data cover the full road network but are comparatively sparse and widely scattered, with no clear spatial pattern.

The overall distribution of DVCs has changed little between 2013 and 2024 (Figure 10). Deer Vehicle Collisions are most frequently recorded on the trunk road network, through the central belt and around Aberdeen. There appears to be a greater density of DVCs on the trunk road network in 2022–2024, with more dark squares positioned on trunk roads. There are also indications that DVCs are less dense but more widespread in Aberdeenshire and Moray in 2022–2024.

Changes in volume of TROC reports overall and by region

Overall, the number of DVCs reported by TROCs remained relatively stable between 2009 and 2024, but show a clear spike in numbers in 2023 (Figure 11).

Figure 11 excludes additional data exported from Transport Scotland’s AMPS, as AMPS data was only available from April 2022 to April 2024. The number of additional records provided by AMPS varied (Figure 12), with the greatest number of additional records provided in 2023, which was also the only year with full data.

The number of DVC records reported by TROCs in the SE region increased gradually from 2008 and 2019, but showed a dramatic increase in reported DVCs between 2020 and 2023, after which numbers returned to normal in 2024. There was a significant strong increase between 2008 and 2024 (Spearman's rank-order correlation, r = 0.85, p < 0.01).

The NE region also showed an increase in the number DVC records reported by TROCs in 2023 and 2024, though numbers had remained remarkably stable between 2008 and 2022.

In contrast, the number of DVC records reported by TROCs in the NW region showed a significant strong decline from 2008 to 2024 (Spearman's rank-order correlation, r = -0.76, p < 0.01) (Figure 13).

The SW region shows great variation in the number of DVCs reported by TROCs per year, but no statistically significant change overall.

Lush and Lush (2023) described an apparent increase in the number of DVC records in the SE unit since 2019, but the updated data instead shows that the number of records remained fairly consistent in 2021–2023 (Figure 14). This jump in the number of DVC records likely coincided with the contract change from Amey to BEAR Scotland in August 2020 and the subsequent provision of direct data extracts from the IRIS and AMPS databases, including some records from before August 2020.

The data also show generally fewer records from the SE unit in 2024 than in previous years across most months, especially in the normal May–June peak. Figure 14 suggests a dramatic drop between higher-than-normal records in December 2023 and lower than normal records in January 2024.

As before, most reports of DVCs in all regions come from the Trunk Road Units, with relatively small numbers being reported from the DBFOs (Table 12).

Changes in the volume of SSPCA and FLS records

Unlike the data received from the TROCs, which is focussed on the trunk road network, data from the SSPCA and FLS offer an insight into DVC frequencies across the full Scottish road network.

The average number of records obtained from the SSPCA shows a significant strong increase between 2008 and 2017 (Spearman’s rank-order correlation, r = 0.95, p < 0.001), with a slight dip in 2014 (Figure 15). However, the annual number of SSPCA records seems to have plateaued since its peak of c.1,250 records in 2017. The apparent dip in 2020 ascribed to the lockdown imposed due to the Covid pandemic by Lush and Lush (2023) is clearly within the annual variation between 2018 and 2024.

The number of records obtained from FLS has seen a general decline since 2011 (Figure 15). There was a significant strong decline in the annual number of records between 2008 and 2024, despite relatively stable numbers until 2012 (Spearman’s rank-order correlation, r = -0.83, p < 0.001).

The distribution of these records in each local authority area in Scotland is shown in Table 13. Aberdeenshire has the largest total number of records spanning all years, followed by Fife and then North Lanarkshire. This is similar to the pattern reported in Lush and Lush (2023), except that Highland has been relegated to fourth place. The Western Isles continue to contain the lowest frequency of reported DVCs.

However, when expressed per kilometre of road within each local authority, East Dunbartonshire has the highest number of DVCs reported by SSPCA and FLS between 2008 and 2024, followed by Aberdeen City between 2008 and 2017, and North Lanarkshire between 2018 and 2024. This likely reflects relatively high numbers of DVCs, dense road networks, high traffic volumes and the location of SSPCA animal rescue centres.

Changes to the number of human injury collisions and damage-only DVCs attended by the police

Overall, there has been a significant strong decline in DVCs from road traffic collision reporting (Figure 16; Spearman’s rank-order correlation, r = -0.80, p < 0.01). Most of this decline has been since 2018, with no significant decline in the number of records between 2008 and 2017. The average number of DVC records from road traffic collision reporting between 2018 and 2023 was little over half the average from previous years, despite the better availability of STATS19 data then reported in Lush and Lush (2023). This decline may be real or reflect incomplete data from 2018 to 2023 if not all relevant STATS19 records were included in Police Scotland’s CRaSH database (see Police and Road Safety Teams’ RTC records).

DVC distribution and change on the wider road network

The changing distribution and frequencies of DVCs across the entire Scottish road network, including non-trunk roads, was assessed by using a hexagonal grid using the method described in Lush and Lush (2023). In summary:

• Each cell was approximately 100 km².
• Cells were attributed with the total length of road they contained, taken from Scotland only Ordnance Survey Open Roads data (see Supporting data).
• Cells were attributed with DVC counts in 2008–2017 and 2018–2024 in the categories: all records, core records only, core records excluding TROC data.
• The average number of DVCs per kilometre of road per year was calculated for the six combinations.
• Change was calculated by subtracting 2008–2017 values from the equivalent values for 2018–2024.

The change in the average number of DVCs per kilometre of road per year using all DVC records is shown in Figure 17, core DVC records only in Figure 18and core records excluding those from TROCs in Figure 19. In each case there was an overall increasing trend, so change per cell relative to the overall change is highlighted by styling the average change as white, instead of no change. All show a relative increase in the average number of DVCs per kilometre of road per year in the central belt and an overall relative decrease in the north. This is less pronounced but still clear when data from the TROCs is excluded (Figure 19), which suggests that the trend is not solely due to an increase in incidents recorded from trunk roads in central belt.

There is a clear outlier on the southern border in Figures 17 and 18. This relates to two records on the A68 at Carter Bar in 2023 and 2024, although there were no previous records for this area. The short length of road within Scotland in this cell (c.1.1 km) exaggerates the change. It is absent from Figure 19, which demonstrates that it has not affected the overall pattern of change.

DVC distribution and change on the trunk road network

The hexagonal grid approach was also applied to DVCs on the trunk road network. Only records that were on the trunk road or within 250 m were included.

The change in the average number of DVCs per kilometre of trunk road per year using all DVC records is shown in Figure 20 and core DVC records only in Figure 21. In both cases, there was an overall increasing trend in DVC records and the strongest increase occurred in the central belt, though for core records this predominantly affected only the eastern part. The north of Scotland showed the strongest overall decline when all and only core records were considered.

Combined records from all core sources by local authority areas

Highland continues to contain the largest number of DVC incident reports from core sources (Table 14), largely due to large numbers reported by the North West unit trunk road operating company (Figure 22, top). Fife, Perth and Kinross, and Aberdeenshire, also contribute large numbers of recorded DVC incidents.

However, Falkirk has the largest number of DVCs per length of road from core sources (Figure 22, bottom), followed by North Lanarkshire and East Lothian. Fife has the fourth largest number of DVCs per length of road, but Perth and Kinross, and especially Highland and Aberdeenshire rank much lower. This reflects road network, deer density and likely recording bias, with local authorities in the central belt ranking higher.

Human injury collisions and damage-only DVCs attended by the police

Whilst most human personal injuries sustained through deer-related road incidents are slight, the higher proportion of serious incidents reported in Lush and Lush (2023) continued between 2021 and 2023, but no fatal collisions were reported (Figure 23). Unfortunately, it is difficult to draw any conclusions about this due to possible gaps in the data: if all missing data between 2018 and 2023 related to slight injuries, the overall severity would be similar to previous years.

An assessment of the 65 CRaSH records of Road Traffic Collisions caused by deer between 1 January 2020 and 15 September 2024, shows that the deer escaped injury in 65% of cases. The fate of the deer was not clear in a further 5% of cases. These findings suggest that the majority of incidents resulting in more serious damage to both the vehicle and its occupants involve drivers either successfully taking evasive action to avoid the deer but having an incident as a consequence, or using the avoidance of deer as a justification for an incident with another cause. For example, in one incident the driver alleged that they swerved to avoid a deer, but subsequently failed a road-side breath test for alcohol. Without an injured/deceased deer or independent corroboration, verification of a deer’s presence may not be possible.

DVC frequency in relation to Time of day and Season

Diurnal patterns of DVC occurrence

Although many DVC records include a time stamp, these often refer to the time of reporting or carcass recovery rather than the incident itself. The data most consistently recording the actual time of collisions are the police incident reports, where times are assumed to accurately reflect the incident. A few other records were also known to have accurate times where the incident had been witnessed.

Analysis of records with reliable time information indicates that the highest frequency of DVCs occurs between 21:00 and midnight (Figure 24). However, during the winter months (December to February), this late-evening peak is less evident, with equally high numbers of incidents spread across the period from 15:00 to 00:00 (Figure 25).

High frequencies of DVCs are also apparent between 18:00 and 21:00 and 00:01 and 03:00 (Figure 24). The early evening peak is most marked in autumn (September to November), whilst the nighttime peak occurs mostly in the spring and summer (March to August).

Seasonal patterns of DVC occurrence by road type and region

As found in the previous studies (Langbein, 2019; Lush and Lush, 2023), seasonal patterns in reported DVC occurrences showed a peak in May to June, regardless of road type (Figure 26).

Police Scotland’s STORM database provides a consistently collected six-year dataset against which seasonal changes may be assessed. This shows a distinct peak in May and roughly equal troughs in September and March (Figure 27). It also shows a significant decline in the May peak over time (Spearman’s, r = -0.81, p < 0.05), whilst figures for the other months remain broadly unchanged and there is no evidence of an annual decline.

The size of the May to June peak varied between the different road types (Figure 28) across all years. It was greatest on motorways, where it accounted for around 50% of all reported DVCs. It was less pronounced on A-roads within the trunk road network (c.36%) and other A-roads (c.28%), and accounted for only about 24% on other non-trunk roads. This suggests a reduction in the May to June peak from major to minor roads.

The results are broadly similar to those of the previous studies, but it is worth revisiting the possible trends identified in Lush and Lush (2023):

• The average annual numbers of DVCs reported from motorways was higher in 2019–2024 than 2008–2018. The monthly distribution was flatter in 2019–2021, but in 2022–2024 it was more similar to 2008–2018 (Table 15).
• The average annual numbers of DVCs reported from A-class trunk roads was highest in 2022–2024, closely followed by 2019–2021, and substantially lower in 2008–2018. This was probably due to increased data availability in recent years. The proportions in each period across the entire seventeen-year dataset was fairly consistent in 2008–2021, but the distribution was flatter in 2022–2024 (Table 16).
• The average annual numbers of DVCs reported from non-trunk A-roads was highest in 2019–2021, closely followed by 2021–2024, and substantially lower in 2008–2018. In addition, the proportion reported in May and June was much higher in 2008–2018 and has decreased in 2019–2021 and 2022–2024. The result of this decrease is that the distribution in 2021–2024 was remarkably similar to the distribution seen for other non-trunk roads (Table 17).
• The annual numbers of DVCs reported from other non-trunk roads was also highest in 2019–2021, followed by 2021–2024, and substantially lower in 2008–2018. The proportions in each period across the entire seventeen-year dataset appear to be remarkably consistent, with a much flatter distribution across all periods than other road types (Table 13).

DVCs on Lewis and Harris

Records made by the Lewis and Harris Deer Management Group in 2023 and 2024 show a peak of DVCs between September and December (Figure 29). The peak in 2024 appears to be about two months earlier, though no data were available for December 2024 and any trend is unclear. Almost twice the number of DVCs had been recorded in 2024 compared to 2023, though this was thought to be due to better surveillance in 2024 (Mark Jamieson pers. comm.).

DVC reports with reliable detail of deer species

As noted by Langbein (2019), the accuracy of deer species identification in the submitted datasets is variable, with only a proportion considered sufficiently reliable for analysis. For records prior to 2019, only those deemed reliable by Jochen Langbein were included in this study. Since 2019, the dataset has been supplemented with verified contributions from the Mammal Society’s Mammal Mapper app, the British Deer Society app, Forestry and Land Scotland, NatureScot, selected animal rescue organisations, validated records from the NBN Atlas, and research-grade records from iNaturalist. In total, these sources accounted for approximately 4.3% of all available records.

Analysis of incidents where species identification was regarded as reliable indicates that roe deer were the species most frequently involved, followed by red deer, with smaller numbers of sika and fallow deer (Table 19). The highest numbers of incidents involving roe, red and sika deer were all recorded in Highland. Red deer outnumbered roe deer only in Highland and North Ayrshire, while incidents involving fallow deer were most frequently reported from Dumfries and Galloway.

Of the records where identification was considered reliable where the sex or age had also been recorded, there were too few records of fallow and sika deer to suggest any trends or bias (Figure 30). Both red and roe deer had more records of adults than juveniles. Whilst red deer showed declining numbers but no clear sex bias overall, female roe deer were significantly more likely to be recorded than males (GLMM, β = -0.33, p < 0.001).

Wildlife Vehicle Collisions including non-deer species

There is a clear increase in the number of Wildlife Vehicle Collisions (WVCs) in the DVC database that did not involve deer between 2008 and 2024 (Table 20), but this is because data for earlier years is less complete. Data for 2022–2024, when most efforts were made to collate data on WVCs not including deer, are also known to be patchy and incomplete.

Badgers were the most frequently implicated animal after deer, amounting to 1,540 records over the full period. There was also evidence of the impact of cars on rare and endangered species, especially otters, but also pine marten and red squirrel. Most if not all records of cats are likely to be domesticated or feral animals rather than true Scottish wildcats, and are the most frequently recorded of several domesticated animals on the list.

Whilst larger animals are more likely to be recorded, rats are surprisingly frequently recorded. It is likely that these mostly occurred on minor roads, as they are unlikely to be spotted on trunk roads and few TROCs provide such records.

DVC risk based on recent incidents

Lush and Lush (2023) developed an approach to identifying DVC hotspots along the trunk road network. DVCs were repositioned on the trunk road where they were within 500 m of that trunk road, in stages:

1. Where the section code associated with the DVC matched a section code in the most recent available trunk road dataset the DVC was repositioned on the closest point on that trunk road section.
2. Where the road number (e.g. A9) associated with the DVC matched the road number in the most recent available trunk road dataset the DVC was repositioned on the closest point on that road.
3. All other DVCs were repositioned on the closest point on the trunk road network.

Repositioned DVCs were also duplicated on nearby trunk road sections where they were within 50 m, meaning that individual DVCs could be represented on multiple trunk roads. The DVCs in the last three years associated with a moving window of 500 m length along each trunk road section were counted, with the window moving 10 m along the section after each count. Short trunk road sections were treated as single lengths, down to a minimum of 50 m.

Non-overlapping trunk road lengths with the highest number of DVCs were selected. The number of DVCs per km per year and normalised DVC risk index (Figure 31) were calculated for each length.

A PostgreSQL function was developed to apply the methodology, allowing it to be rapidly repeated using different parameters. Whilst this function had been enhanced during this contract, the basic approach remained unchanged.

This approach was repeated for DVCs from 2022–2024. Including more years of data was rejected due to suspected positional bias in the data before 2022, whilst the data included appeared to contain limited bias (Annex 2: Accuracy of incident locations). Note that the results are not directly comparable with previous results, due to differences in the data sources used and the way that the normalised risk index is calculated.

DVC hotspots were calculated using a minimum of two DVCs per length of trunk road and zero DVCs per length of trunk road. Due to the way that it is calculated, these produce different normalised DVC risk index results. A minimum of two DVCs per length of trunk road is most useful for identifying hotspots, whilst using zero DVCs produces a dataset with near complete coverage – small gaps between lengths with high normalised DVC risk index values are not filled. The former are considered here, whilst the latter data were used in Predicting DVC risk using environmental data.

The results of the risk analysis for 2022–2024 are shown in Figures 32 to 38. These show that hotpots remain densest in the central belt, as in 2018–2021 (Lush and Lush, 2023).

The top 20 DVC hotspots for all road lengths between 50 m and 500 m includes several on each of the A1, A9 and A90 (Table 21). However, none of the identified top 20 hotspots relate to 500 m road lengths, which suggests bias towards shorter road sections having highest calculated DVC risk. It is therefore also appropriate to consider only those hotspots that relate to 500 m road lengths, which additionally identifies several each on the A87, M876 and M90 (Table 22). Combining the roads in the two tables suggests that the M90 has the largest number of DVC hotspots (seven), followed by the A1 and M876 (four each), and the A87, A9 and A90 (three each). Section 14402/95 on the M876 includes two separate hotspots that appear in the top twenty road hotspots that relate to 500 m road lengths.

Several roads were flagged as DVC hotspots in both 2018–2021 and 2022–2024, in several cases with the same link section (Table 23). Most of these had been considered questionable due to positional bias of DVCs along the trunk roads. It appears that previous concerns were therefore unfounded and these road links are of concern.

As in Lush and Lush (2023), the 40 hotspots listed in Tables 21 and 22 were manually assessed. This assessment prioritised hotspots with the highest normalised risk index with high numbers of implicated and reliably located DVCs, longer lengths of road, and clusters of hotspots. This resulted in a shortlist of the ten locations with the highest risk of DVCs, based on reliably positioned DVC records (Table 24). 38% of shortlisted hotspots included 500 m parts of longer road sections.

The locations of the combined top 10 DVC hotspots are shown in Annex 3: Maps of DVC hotspots for 2022–2024.

Predicting DVC risk using environmental data

Heckels (2018) used several models to investigate the relationship between DVCs and a range of road characteristics and environmental factors. They used the original point-based Scottish DVC records for 2009–2016 and unmodified trunk road sections, which are highly variable in length. They found that a generalised additive model performed best and explained 36–37% of the variability in the data, but had poor predictive ability.

The availability of DVC count data related to standardised lengths of trunk roads (DVC risk based on recent incidents) presented an opportunity to develop new statistical models. The focus was to investigate the factors related to the spatial distribution of DVCs and did not consider temporal variation.

The review by Putman and Langbein (2024) identified several factors that could be used to explain the incidence of DVCs. The variables used in this analysis were:

• The number of DVCs along non-overlapping lengths of road, including zero DVCs.
• The length of road considered. Because of the algorithm used, these ranged from 50–500 m and were strongly biased towards 500 m.
• Estimated Annual Average Daily Flow (AADF) data for each road length. These were based on the sum of individual counts for different vehicle types on each trunk road section.
• A measure of road sinuosity. The ratio of the length of the road and the straight-line length between start and end nodes was used.
• The maximum speed limit for the road. Most were 60 or 70 miles per hour, though small numbers of road lengths were 30, 40 or 50 miles per hour.
• The distance to the nearest woodland, moorland and urban habitat within 2 km. Habitats were mapped on a 20 m grid in the Scotland Habitat and Land Cover Map 2022 EUNIS level 1 data (Space Intelligence Ltd., 2023). Additional mapped woodland areas wider than five metres were extracted from AMPS habitat data and included, as these were thought likely to provide additional information on woodland immediately adjacent to roads.
• The area of woodland, moorland and urban habitat within 25 m perpendicular to the road.
• The area of woodland, moorland and urban habitat within 100 m perpendicular to the road.
• The area of contiguous woodland, moorland and urban habitat intersecting a 25 m buffer around the road.
• The road type. Most lengths were classified as carriageway, as they related to the major stretches of trunk roads. Fewer lengths were classified as slipways or roundabouts, which represented typically shorter lengths of road at junctions.
• The number of lanes, extracted from AMPS data.
• An estimation of the road width, based on:
  • 3.65 m for each normal carriageway.
  • 4.00 m for each wide carriageway.
  • 3.30 m for each hard shoulder.
  • 5.00 m for each central reservation where the road was dual carriageway or motorway.
• The total length of road barrier within 10 m perpendicular to the road.
• The number of “Wild animals likely to be in road ahead” signs within 100 m. Sadly, it was not possible to limit these to those within 100 m along the road network.

Many of these variables were expected to be collinear, but were included so that the most informative could be selected.

Other possible factors not considered included:

• Actual traffic speeds. Estimated average traffic speed data for roads is available from Ordnance Survey and could be included in a future analysis.
• Landscape heterogeneity and patch size. It was felt that the habitat data available were inadequate to calculate these metrics.
• Deer species. Very few DVCs could be adequately assigned to species for the years considered. Given different deer have different ecological requirements and will be found in different locations, this lack of information is likely to have perturbed the model.
• Time of year and day. Due to the nature of the data analysed, this information could not be considered by the analysis.
• The presence of wildlife corridors perpendicular to and crossing the road. This would be challenging to achieve at a national level.

Only the data for 2022–2024 were included in the analysis, due to concerns about positional bias in earlier data (Annex 2: Accuracy of incident locations). Data diagnostics suggested that the data had a negative binomial distribution. Zero-inflation in the data was rejected by comparing predicted with observed zeros and through Vuong's non-nested test. One road length, containing 34 DVCs over the three-year period, was removed as an outlier following model diagnostics.

Models were assessed for collinear predictors and those with the poorest performance were removed. This resulted in the removal of several predictor variables:

• The area of woodland within 100 m – collinear with the area of woodland within 25 m.
• The area of moorland within 100 m – collinear with the area of moorland within 25 m.
• The area of urban habitat within 100 m – collinear with the area of urban habitat within 25 m.
• The number of lanes – collinear with the estimated road width. 

Predictor variables were also removed using a combination of stepwise regression and manual removal of poorly performing predictors. This resulted in a more parsimonious model containing eleven predictors, all of which showed a highly significant relationship with the number of DVCs within each length of road over the three-year period. However, all models failed to converge, most likely due to the relatively few DVCs over the whole trunk road network in the three-year period, and were regarded as unreliable.

DVCs in Deer Management Priority Areas

The Operational Delivery Workstream of the Deer Management Strategic Board identified several Deer Management Priority Areas. These aimed to deliver a cohesive and dynamic ‘on the ground’ approach to deer management, with all partners focussing on opportunities for landscape scale delivery (NatureScot, 2025).

A series of maps summarising the occurrence of DVCs within the Priority Areas were developed in collaboration with NatureScot staff. These were based on the average number of DVCs per kilometre of road per year within 5 km² hexagonal cells. The length of road was calculated using Ordnance Survey OpenRoads data. A PostgreSQL function was developed that could undertake the analysis on any set of areas using any cell size.

Two maps were produced covering two five-year periods for the previous decade of data, allowing comparison of the two maps. Simplified versions of the maps are provided in Annex 4: Deer Management Priority Area DVC maps. These show that the Central Scotland Priority Area had the most DVCs per kilometre of road per year, though DVCs appeared fewer and more patchy in 2020–2024 than in 2015–2019 (Figure 67). Large parts of the Clyde Climate Forest Area Priority Area also had large numbers of DVCs, though again fewer in 2020–2024 (Figure 68). In most other areas DVCs were more scattered.

Navidale mobile VMS deployment

Mixed-effects regression analysis of vehicle speed data from the VMS deployment at Navidale showed that time of day, vehicle type and snow affected vehicle speeds, but the deployment of the VMS had no significant effect. This was true regardless of the direction of traffic. Ignoring vehicle class, the snow had the biggest effect on vehicle speed, accounting for a reduction of about 4.5 miles per hour controlling for other variables (p < 0.01).

Discussion and recommendations arising

Data collection and collation

Data collation was successful, with data from all core sources and several supplementary data sources added to the DVC database. The DVC database now includes 43,482 mapped incident records for the period 2008–2024. During this contract, 20,729 records were added, which includes previously missing 2020–2021 data for the M8 and pre‑2022 data from several significant new sources (Lush and Lush, 2023). Non-deer species account for 3,303 records.

The supply of CRaSH data by Police Scotland has allowed STATS19 records to be included for all local authority areas for the first time since 2017. The completeness of the data is not known, but it is likely to include all or most DVCs that resulted in a human injury, since most STATS19 incidents are expected to be recorded in CRaSH.

In terms of numbers, data from Police Scotland’s STORM database were the most significant addition to the DVC database. However, some records relate to deer on roads, rather than DVCs. All STORM records have been included in the database, since all represent an increased risk of DVCs. Assuming that it is possible to separate DVCs from other records in the STORM data, it would be appropriate to add a field to the DVC database to flag collisions.

The positional accuracy of the STORM data is also unknown, but a review of records around the DVC hotspots suggested that it was not accurate. For example, it included a small cluster of records from a fast-food outlet. Records were also often unrealistically clustered on single points. Future work would benefit from investigating how incidents are reported and assigned locations, and any bias apparent in the data.

Data extracted by Transport Scotland from its AMPS database also made a valuable new contribution. Whilst many records had also been obtained from TROCs, it also contained numerous non-duplicate records that have enhanced the analysis. The AMPS database was rolled out to TROCs and DBFOs gradually, which means that the chances of duplicate records may increase alongside use of the AMPS. However, it is also known that some TROCs and DBFOs maintain separate systems for recording DVCs that may not be readily recorded in AMPS.

Data from the STORM and AMPS databases were not treated as core sources in this contract and were therefore excluded from many analyses. However, they could be valuably included in many of these analyses if the data is added to the DVC database in future years. This would essentially involve different analyses to look at trends from 2008 and a later year from which STORM and AMPS data is consistently available.

The addition of records from the NBN Atlas and iNaturalist has increased the number of records where species are thought to have been reliably identified. These will need to be regularly updated, which must include the deletion of existing data and addition of up-to-date records for all years, as this allows for redeterminations and other changes to historic records.

Data from NatureScot and New Arc Wildlife Rescue also increased the number of reliably identified records.

Basic analysis, trends and distribution of DVC occurrence

Evidence for DVC trends

Trends in DVCs appear to be in line with previous years once new data sources are taken into account, with an overall plateauing of records between 2017–2024 (Lush and Lush, 2023). The addition of supplementary data from new sources, notably the large Police Scotland STORM dataset, means that there are now many more records from 2019–2024. Nevertheless, the overall number of DVCs per year between 2019–2024 also appears to be relatively stable.

The numbers of DVCs reported by TROCs remained broadly stable between 2009–2024, though individual regions show large variability. Whilst the North East region had reported relatively stable numbers of DVCs per year until 2022, the numbers in 2023 and 2024 were clearly higher. Whether the apparent increase in the North East region continues and proves to be significant will become clear as more years are analysed. The significant declining trend in DVCs in the North West region continued, though shows signs of stability between 2019–2024. Meanwhile, annual numbers of DVCs in the South West region continued to show great variability, with no apparent trend in recent years.

The significant increasing trend in the South East TROC region has continued, with numbers in 2021–2023 approximately twice those of previous years and 2024. The reasons for the increase and subsequent decrease are unclear, but may be due to the move from IRIS to AMPS and subsequent adoption of AMPS modules – an issue that may also have affected other TROCs and DBFOs. Staff changes may also mean that different techniques were used to extract the data, which may have affected consistency. The extent to which records where a deer carcass was not uplifted were including in 2024 data is unclear, but may explain the peak in 2021–2023 if they were subsequently removed from the data supply.

Despite the variation in records from TROCs, the trunk road network shows a clear increase in the number of DVCs per length of road per year between 2008–2017 and 2018–2024. This is due to records from other sources and is true regardless of whether all or only core records are considered. Perhaps more interestingly, no part of the trunk road saw a decline in DVCs per length of road per year.

Most variation in the numbers of DVCs per year from core sources is due to data from the SSPCA. Whilst this has significantly increased since 2008, the previously speculated dip in numbers in 2020 now appears to be a general plateauing of numbers since 2017 (Lush and Lush, 2023). The reasons for this are not clear, but will likely relate to a range of perturbing factors including availability of suitable staff, SSPCA locations in relation to DVC incidents, changes to whether the SSPCA are called by key individuals and organisations, and the location and availability of other wildlife rescue centres. Whether this represents a plateauing of DVCs within the wider road network is therefore unclear.

DVC reports from Forestry and Land Scotland (FLS) continue to decline. It is likely that this is due to culling activities around FLS sites that have reduced the number of incidents that they are called to attend (Lush and Lush, 2023). As such, FLS data represents an interesting case study in how consistent, long-term deer culling may reduce local DVC numbers, though it is acknowledged that culling is not possible or as effective in all areas (Putman and Langbein, 2024).

On a finer scale, DVCs remained concentrated in the central belt, with evidence of increases from 2008–2024 across all relevant data sources. Whilst numbers remained comparatively low, there was also evidence of above average increases in DVC frequencies in the south and north east, along with below average increases in the north and west. Two local authority areas, Argyll and Bute and Aberdeen City, showed a decline in the number of DVCs reported by the SSPCA and Forestry and Land Scotland, which may be representative of overall trends in the wider road network. Indications of an increase in DVCs in the Aberdeen area vanish when only core data sources are considered and are likely due to the inclusion of data supplied by New Arc Wildlife Rescue.

Any apparent localised increases or decreases in the data are likely to be within the expected range of random variation and could be due to fluctuating deer numbers, changes in traffic levels or reporting likelihood, or the implementation of localised DVC mitigation, such as signage or culling.

Overall, this suggests a relative decline in DVC frequencies in north Scotland and a relative increase in the central belt. No clear change in DVC frequencies could be discerned for the area south of the central belt.

Diurnal and seasonal distribution, and species involved

The available evidence suggests that the DVC incidents continue to peak between 21:00 and 00:00, or between 15:00 and 00:00 between December and February. These patterns reflect those previously reported (Langbein, 2019; Lush and Lush, 2023), though improvements to the data more clearly show the nighttime peak in spring and summer.

Although deer are active throughout the 24-hour cycle, they show strong crepuscular peaks (Mitchell, Staines and Welch, 1977; Georgii, 1981; Pagon et al., 2013; Sándor et al., 2013; Ensing et al., 2014; Ikeda et al., 2019). In upland habitats, deer typically descend from higher ground at dusk to forage, returning at dawn, while in lowland landscapes they emerge from woodland at dusk to feed and likewise return at dawn (Mitchell, Staines and Welch, 1977). These movements place deer close to roads in the evening and early morning, with evening activity occurring earlier in winter months. Activity also increases, and becomes less predictable, during the rut (Sándor et al., 2013), which can shift risk to earlier in the day – a pattern evident in this study between September and November.

Traffic levels also fluctuate daily, generally peaking during morning and evening commutes. When deer activity coincides with these peaks, DVCs are more frequent, as seen between 15:00 and 18:00 during December to February.

Driver behaviour and road conditions add further complexity. A consistent peak in DVCs occurs between 21:00 and 00:00 throughout the year, which may reflect driver fatigue, reducing their ability both to see and then react to deer. Alcohol or drug use could also play a role, though collisions linked to these factors may not display a consistent daily pattern.

Adverse driving conditions – such as reduced visibility or slippery roads – can heighten risk. However, the evenings of December to February, when conditions are poorest, do not show a sharp peak in DVCs. This may result from reduced deer activity or lower traffic volumes. Red deer in particular are less active during January and February (Georgii, 1981), though the available evidence also suggests that the daily peak in DVCs may be spread across a longer period, from 15:00 to 00:00, during the winter months. Evidence from Lewis and Harris, where red deer are the only deer species present, suggests a peak of DVCs involving red deer between September and December. A similar peak occurs between September and November at Navidale, north Scotland, where red deer also predominate. These seasonal peaks coincide with the annual rut, which peaks in October in Scotland (Mitchell, Staines and Welch, 1977).

The primary reason for the increased average number of DVCs per year on all roads, especially trunk roads, is likely to be better reporting, due to the increased number of reports received from the SSPCA and the inclusion of Police Scotland’s STORM data. However, this is unlikely to have affected the monthly proportions of DVCs.

DVC frequencies continue to show a peak in May and June across all road types, but especially for motorways and ‘A’ class trunk roads. The proportion of DVCs in May and June has remained broadly consistent on all roads except non‑trunk A‑roads, which show evidence of decline. At the same time, the proportion of DVCs in November and December has increased on non-trunk A-roads. Police Scotland’s STORM dataset, which is expected to be consistent between years and reflect the entire road network, also shows a significant decline in the May peak. The reasons for these apparent trends are not clear.

Analysis of records where the species of deer was thought to have been reliably identified suggests no change from previous analysis (Lush and Lush, 2023). Roe deer appear to be the most frequently involved species throughout Scotland and in most local authorities, and are reported from all local authorities except West Dunbartonshire. Red deer are the next most frequent, having been reported from nine local authority areas. Red deer are also the most frequently reported species in Highland and North Ayrshire. Sika and fallow deer are less frequently recorded, though fallow deer were most frequently recorded in Dumfries and Galloway, and the only report from West Dunbartonshire with a reliable species identification involved a sika deer.

The results suggest a significant bias towards DVCs involving female roe deer over males. This may be because female deer are more likely to seek out and utilise wooded ‘islands’ within complex road junctions and interchanges or woodland strips adjacent to roads whilst they are rearing young, as these areas are generally free of human and dog disturbance. These deer are then more likely to be involved in RTCs when crossing the adjacent roads to access their wider environment.

DVCs resulting in human injury or vehicle damage

There is an apparent declining trend in DVCs resulting in human injury or vehicle damage, according to the available STATS19 data. This is most noticeable in an annual decline in injuries regarded as slight, whilst the annual number of serious injuries remain broadly unchanged. Fatal and damage-only incidents have been rarely reported since 2008.

However, it is hard to draw firm conclusions, as the data may be incomplete. Even if complete data has been included, there may have been a declining tendency for police to record slight injuries, or perhaps the threshold used for a slight injury has changed. This should be explored with Police Scotland in a future contract.

There are also questions around how many RTCs where deer are implicated actually involve deer. It is easy for drivers to ‘blame’ deer for an incident, excusing bad driving and/or human error on the sudden presence of one in the road. There will always be an element of doubt unless dead or injured deer are found at the scene of the incident. How many of the approximately two-thirds of incidents where deer are implicated but not at the scene this scenario affects cannot be known.

Lack of British Deer Society data

The number of records per year previously obtained from DeerAware and the British Deer Society mobile app was low, but its absence from this study is nevertheless notable. Relatively few data sources provide accurate information on deer species, but these datasets were considered reliable. It is also unclear what currently happens to records made via the DeerAware website, which was still live and presumably receiving reports at the time of writing.

Funding should be made available for the British Deer Society to continue their work in this area, including collation of DeerAware data, update and release of the mobile app, and ongoing data management. Funding should be contingent on the supply of data to inform analysis such as this. 

Incident-based assessment of DVC risk

Analysis of DVC risk at a fine enough resolution to guide mitigation efforts relies upon the provision of positionally accurate DVC reports. Whilst DVCs more than 500 m from the relevant trunk road can be removed from the analysis, identification of records inaccurately located on the trunk road networks is non-trivial. Poorly located DVC records can result in artificially clustered records of DVCs that bias the results of the analysis.

The previous analysis identified potential bias in the position of records supplied by several TROCs, but these issues appeared largely resolved in this analysis. Only the North West Unit showed evidence of positional bias in the years analysed, and only in the first few months of data (Annex 2: Accuracy of incident locations). This may be partly due to the replacement of IRIS, which contained a known bug where locations defaulted to trunk road section start nodes, with AMPS. Whilst this marks an important improvement, positional inaccuracy may occur without biasing data. Positional accuracy remains a key goal for TROCs and other data suppliers.

Whilst all trunk road operators are meeting their DVC-related contractual obligations and providing high-quality data on DVCs, the only location information for some records was a section code or description. Ideally coordinates would be supplied, but it is not clear whether coordinates would reflect the actual locations of incidents. This may be true of TROC data more generally, as reports are often made after the incident has been attended and the process of locating an incident may preclude the recording of accurate locations.

Positional accuracy was also an issue in other datasets. SSPCA data only includes postcodes, which cover an area rather than a point location, so accuracy is inevitably variable. Also, manual review of Police Scotland’s STORM dataset, whilst being a significant contribution to the analysis and including coordinates, identified clustering of records and suggested that some were poorly located.

Manual review of the underlying data for lengths of road identified as high-risk was therefore necessary to account for the influence of potential positional accuracy issues. This was undertaken in 2019–2021 (Lush and Lush, 2023) and 2022–2024, and tended to be cautious, removing high-scoring road lengths from the shortlist if there were any potential issues. It was not necessary to review the full forty highest scoring lengths of roads to create the shortlist for 2022–2024, which demonstrates the improvements made since 2019–2021.

Shortlisted hotspots for targeted mitigation activities

The ten subjectively selected most significant hotspots for 2022–2024 were all within the central belt (Annex 3: Maps of DVC hotspots for 2022–2024; Table 24), as were most of those within the top twenty lists for 50–500 m and 500 m lengths (Tables 21 and 22). This broadly reflects the lists for 2019–2021.

However, only one of the 2019–2021 top ten hotspots, the A9 Dunblane West junction, was selected again. This may be due to the mobile and territorial nature of roe deer, which means that woody islands within road interchanges and road-adjacent habitat may be occupied in some years and not others. This makes it hard to create a reliable list of the locations with the highest risk, as the level of risk will change from year to year. This could be mitigated against by increasing the number of years included in the analysis, which may be possible in future years if the positional reliability of incident records continues to improve. Nevertheless, many of the longer lists of hotspots included locations identified previously.

One notable omission from the hotspot lists was the stretch of the A9 north of the Navidale Roundabout, which was a top hotspot in 2019–2021. This does not mean that deer are no longer a concern along this stretch of road, as the surrounding land continues to be well used by red deer (Langbein, 2025). Several factors may have meant that it was not identified as one of the most significant hotspots in 2022–2024, including:

• The removal of scrub that had been encroaching onto the verge will have increased visibility for drivers and deer, thus reducing the risk of DVCs (Langbein, 2025).
• Other mitigation may have reduced DVC numbers, such as the implementation of a mobile Variable Messaging Sign in November 2024 (Mitigation at Navidale).
• Low rates of reporting is a known issue in the more remote locations (Langbein, 2011) and is likely a factor here.
• The length of affected road is very long, compared to some of the more localised hotspots in the central belt. DVCs spread over a greater length of road are less likely to be identified as a hotspot than the same number over a shorter road length.
• Whilst the risk of DVCs may remain high, this does not mean that they will be temporally evenly spaced. The low numbers in 2022–2024 may be within the normal range of variation.

All of the shortlisted hotspots and most of those in the top twenty lists for 50–500 m and 500 m lengths (Tables 21 and 22) related to junctions. This supports previous conclusions that junctions are more likely to be DVC hotspots, especially in the light of improvements to the positional accuracy in the source data. The possible reasons for this given by Lush and Lush (2023) and Putman and Langbein (2024) are:

• Slip roads at road junctions tend to be curved and thus visibility is reduced, potentially increasing the chance of vehicles colliding with deer on the road.
• Slip roads are not in constant use, potentially making it more likely that deer will stray onto the road when cars are not visible.
• Road junctions often include wooded areas surrounded by roads with no public disturbance that may act as daytime refuges for deer from which they will move to feed. This is especially likely near built up areas where there are few alternative undisturbed resting up places for deer.
• Fencing along roads must end at junctions to permit vehicle access, also allowing access to animals otherwise excluded from the carriageway.

The shortlisted hotspots could be targeted for mitigation. In their review of mitigation measures to reduce DVCs, Putman and Langbein (2024) primarily recommended the use of fencing, Dynamic Warning Signs activated by Roadside Animal Detection Systems, and the removal of tree cover and shrubby vegetation along verges. They note that mitigation needs to be tailored to the specific location and typically involve a range of measures, including monitoring to determine its effectiveness. However, it may be appropriate to await results from the mitigation proposed in Langbein (2025), as this may indicate the most effective actions at each hotspot.

Improving DVC hotspot mapping

The current approach to DVC hotspot mapping has proven to be effective since its development by Lush and Lush (2023). For example, the outputs of this analysis were used to quickly assess the basis for concerns at one location raised by a member of the public. Four of the worst hotspots were also selected as study sites for Langbein (2025).

The approach is most similar to the Getis-OrdGi* statistic used by Garrah et al. (2015), but with differences, most notably the use of a moving window algorithm. However, other standard approaches have been applied elsewhere that may be more accurate or provide additional insight.

Kernel Density Estimation (KDE) has been used by several similar studies (Bíl et al., 2015; 2019; Seiler et al., 2016; Laube et al., 2023). This is a non-parametric method for estimating the probability density function of a random variable by treating each point, in this case each DVC, as the median point on a symmetrical curve, i.e. a kernel. How wide each kernel spreads is determined by a bandwidth parameter. Summing the values of all kernels creates a surface that shows where the points are more densely clustered.

KDEs fall into two types:

• Planar KDE, which treats the world as an open, two-dimensional plane. The kernels therefore spread in all directions. Applied to DVCs, this means that a kernel from a DVC on one road can affect the density estimation of other roads within the distance determined by the bandwidth.
• Linear KDE, which are constrained to a linear network, such as a road network. The kernels therefore only spread along the road network, including following branches in the network that represent joining roads. Applied to DVCs, this means that kernels from DVCs can only affect roads that are physically connected, up to the distance determined by the bandwidth.

The main difference between KDE and the approach used in this study is that KDE produces a smooth surface. The size and type of kernel, which dictates the shape of the curve, dictates how smooth the surface is. KDE is also more statistically rigorous. They are otherwise compatible and likely to produce similar results.

Linear KDE has been used in most studies, but would be difficult to implement on Scottish trunk roads, as the available trunk road data does not form a cohesive network – there are gaps between some of the sections. This means that kernels would not be able to pass to physically joined road sections where they are not joined in the data, resulting in artificially abrupt changes in the surface. Linear KDE also fails to account for the increased risk indicated by DVCs on nearby roads, which the approach applied in this study attempts to include. This is particularly important in Scotland, since many DVCs occur at junctions where there are multiple road sections and dual carriageways are represented as parallel lines in the network.

Planar KDE has been used by (Laube et al., 2023) and overcomes some of the shortcomings of linear KDE, as DVCs on one road will affect nearby roads. However, planar KDE applied to a network assumes that the area between links in the network is uniformly penetrable. This may not be the case for DVCs, as fences and other obstacles could limit the movement of deer between nearby road sections. The nature of planar KDE also means that it produces uniform results only when applied to straight roads, as each kernel can intersect with a greater length of sinuous road. This may result in bias towards sinuous roads.

Attempts could be made to resolve the issues with the Scottish trunk road network data. However, automated approaches to repair networks rarely produce good results and manually joining road sections would be costly. This suggests that planar KDE may be the best method for analysing DVCs in Scotland, despite its shortcomings.

However, other solutions could be considered:

• Nearby road sections could be connected where they are within a defined distance, such as 50 m. This would mean that the original network data would be unmodified, but would connect them to allow kernels in linear KDE to spread to nearby roads. The connections could be weighted, thus adding a cost reflecting the uncertain permeability of the space between the roads to deer. However, the results would be influenced by the frequency of the connections. Whilst this could be resolved by increasing the frequency of the connections, the amount of processing required to adequately resolve the issue could be too costly.
• Implement a hybrid approach that combines linear and planar KDE. This would involve calculating both linear and planar KDE, and then combining them for the road network, summing weighted or unweighted values. This could produce results that account for the increased risk of DVCs on the road section where a DVC has occurred and, to a lesser extent, the likely increased risk to nearby roads.

Further exploration and application of KDE to DVC analysis in Scotland is recommended for a future contract.

Proximity of DVC hotspots and road signs

An analysis of the proximity of DVC hotspots and permanent signs indicating a risk of deer on the road was undertaken, but provided no evidence of a relationship. It considered two types of sign: the “Wild animals likely to be in road ahead” signs that feature a picture of a running deer and fixed location Variable Message Signs (VMS) that can programmed to display different messages to alert drivers to the risks on the road.

Several issues were noted that meant a reliable statistical analysis was not possible, notably:

• The data showing the locations of “Wild animals likely to be in road ahead” signs was thought to be incomplete. The completeness of data on VMS was unknown.
• The direction of the sign, i.e. which way it was facing, was not known. This is important, as the sign might occur after a hotspot when accounting for direction of travel.
• Signs sometimes state the distance of road covered by the warning, but this data was not available.
• “Wild animals likely to be in road ahead” signs may not relate to a risk of deer on the road or necessarily be placed in response to strong evidence of a high risk. Additionally, if the signs were reducing DVC risk, then the likelihood of a nearby DVC hotspot would also be reduced.
• Straight-line distances were considered, rather than distances along the road. A network analysis was considered to give more accurate distances along a road, but was not undertaken due to the other issues noted.

A manual assessment of the ten most recently installed permanent “Wild animals likely to be in road ahead” signs up to the end of 2022 was therefore undertaken. The direction of each sign was established using Google Streetview and any distances associated with the signs were noted. These were compared with DVC risks along 500 m stretches of road (Table 21). The results of this assessment suggested that the reasons for the installation of many signs could not be related to a high incidence of DVCs. Where signs may have been installed due to high DVC incidence, there was little evidence to suggest that they had been effective. Only one location showed evidence of a decline in DVC incidence since the installation of signs. This aligns with other studies, which show no evidence to suggest that permanent signs reduce the incidence of Wildlife Vehicle Collisions (Putman and Langbein, 2024).

Analysis of the impact of road signs on DVC incidents may be possible, but this will need to be on a case-by-case basis. This could be achieved in two ways:

1. Identify DVC hotspots and either install signs or make use of mobile VMS, and compare DVC incidence before and after.
2. Analyse how the incidence of DVCs is affected if signs are removed due to damage or wear and tear, and not replaced.

Predicting DVC risk

Due to the relative sparseness of the few years of unbiased data included, the models described in Predicting DVC risk using environmental data did not converge reliably and should be interpreted cautiously. Nevertheless, it is worth considering the predictors that were retained, as these suggest patterns in the data.

The greatest effects on DVCs appeared to be due to road type. The model suggested that slip roads and roundabouts had several times more predicted DVCs than normal roads. This is unsurprising, given the patterns already observed in the data and the locations of the worst DVC hotspots in 2019–2021 and 2022–2024.

Interestingly, wider roads appeared to be associated with fewer DVCs. This may be due to the effects of slip roads, but may also suggest that deer are less likely to cross wider roads. Wider roads may also give drivers more space in which to see a deer and take successful measures to avoid a collision.

As expected, the model suggested that nearby habitat had an effect on DVC frequency. The area of woodland and urban habitats within 25 m appeared to have a positive relationship with the predicted number of DVCs. The apparent effect of woodland is unsurprising and is possibly heightened where urban habitats are also nearby. Urban habitats will reduce the availability of semi-natural habitats and increase potential public disturbance, thus making woodlands near roads more attractive to deer. However, the model suggested that large, contiguous areas of urban habitat nearby were negatively associated with DVCs. This may be because of a relative reduction in semi-natural habitat that can support deer, leading to a supressed deer population and fewer DVCs. It may also be that vehicle speeds near large urban areas were slower on average, due to slower maximum speed limits or volume of traffic, thus reflecting the risk of DVCs.

With the exception of road width, the aforementioned predictors most likely relate to roe deer. However, the model suggested that the proximity of moorland was positively related to DVC numbers, with predicted DVCs declining as moorland became more distant. This was most likely to reflect the likelihood of DVCs involving red deer. Moorland proximity was likely selected over moorland extent because moorland tends not to occur within 100 m of roads in the central belt where most DVC hotpots occur.

The model also suggested that increases in the number of vehicles on the road and in road speed increased the number of predicted DVCs. Higher volumes of traffic will increase the opportunities for DVCs to occur, whilst increased speed reduces stopping distances and therefore the number of DVCs that are averted. 

Finally, the model suggested that the longer lengths of barriers within 10 m of a road increased the predicted number of DVCs, though only slightly. This may be because these act as a barrier to deer exiting roads and, in the case of central barriers, reducing deer transit time, meaning that they are on the road for longer. It suggests that roadside barriers are not effective at reducing DVCs unless deer are completely excluded from entering the road in the first place. The data exported from AMPS suggested that there was only 6 km of roadside deer fence in Scotland.

These predictors are therefore consistent with ecological expectations and previous studies, but given the convergence issues and data sparseness, they should be regarded as indicative associations rather than statistically reliable effects.

It may be possible to produce a reliable model given sufficient data without positional bias. Repeating the analysis with additional years of unbiased data will allow the robustness of these patterns to be tested. Assuming the positional accuracy of DVC records continues to improve, or with improvements to data management to flag and exclude records with positional accuracy issues without otherwise biasing the model, three more years of data may adequate. Repeat analysis using 2022–2027 data is therefore recommended.

Ideally, future analysis could incorporate some potential predictors that could not be included here. Notable parameters include actual traffic speeds and landscape heterogeneity and patch size. The landscape measures may require better habitat data than was available for this study. The presence of wildlife corridors perpendicular to a road would also be a valuable addition, though it is unclear how this data could be derived from information available at present. The slope of the road may also be relatively easy to calculate given suitable data and may show a relationship to DVC frequency (Laube et al., 2023). Other potential predictors, such as visibility and the presence of embankments, could ideally be included, though this may be difficult without manual capture of suitable data. 

It may be possible to include information on the time of year and day. These do not affect where DVCs happen, but as DVCs are temporally unevenly distributed they could then be included as random effects in a mixed-effects model. However, including these in what is essentially aggregated data is likely to mask any effects and incident times are reliable in only a small proportion of cases.

Ideally, it would also be useful to include the species of deer implicated in a repeat analysis. In theory this could be achieved by including counts for each species, but in practice relatively few DVCs are accurately identified to species. This means that future models are likely to be biased towards incidents involving roe deer. A possible proxy for this information could be local population estimates for each species, but this information currently does not exist at a national level. In the absence of information on deer species, the eastings and northings of the road length could be included as a proxy, to reflect the different distributions of deer species in Scotland.

Addition of non-deer species to the database

Changes to the DVC database mean that it is now possible to include incidents involving species other than deer. Whilst DVCs remain the focus in Scotland, this opens up possibilities for other analyses in the future. The data on incidents involving non-deer species is currently incomplete, but as more data holders supply records of non-deer species the analysis of such data could be improved. This has three potential benefits for Wildlife Vehicle Collision (WVC) analysis in Scotland:

• Better data on WVCs in general can help calibrate DVC records by determining which species are most and least affected.
• The risks imposed by non-deer species may change over time, in which case data tracking this change would be beneficial. For example, collisions with badgers and feral pigs can damage cars, and the latter may increase in Scotland without management (Massei and Ward, 2022).
• Better knowledge of non-deer WVCs could improve our understanding of the impacts of traffic on rare or threatened species, such as pine marten, otter and red squirrel.

More comprehensive data on non-deer species would be possible with investment in working with data holders to obtain such data. However, issues will likely remain with obtaining reliable data on small animals, as injured specimens are often taken directly to wildlife rescue centres (Paul Reynolds, pers. comm.).

These changes to the DVC database also create the opportunity to undertake similar analyses of WVCs in other countries where deer are not the sole or main concern, whilst using the same database and analysis infrastructure.

Mitigation at Navidale

Analysis of vehicle speed data at Navidale indicated that the deployment of two mobile VMS warning of deer on the road had no detectable effect on vehicle speed. However, the reliability of this finding is uncertain because of how the data were collected and the specific road layout.

The speed detection strips were located approximately 422 m north of the northbound VMS and 551 m south of the southbound VMS. The northbound VMS was positioned at the Navidale roundabout, where vehicles were still accelerating towards their normal cruising speed. In this case, the absence of a measurable effect may reflect the fact that recorded speeds captured vehicles before they reached their maximum speed. If the VMS did influence drivers, this might only have been apparent further along the road, beyond the location of the speed detection strips.

Southbound traffic was not constrained by the roundabout, and the distance between the VMS and detection strips should have been sufficient to detect any change in speed. However, speeds immediately north of the VMS were not monitored, so it is not possible to assess whether the VMS had an effect in that location. It is also possible that southbound traffic happened to be travelling faster overall during periods when the VMS was active and conditions were clear of snow, further complicating interpretation.

Future data collection of this type would benefit from closer consideration of the relationship between VMS placement and speed monitoring. Ideally, several sets of detection strips would be positioned at varying distances downstream of the VMS to capture potential changes in speed over distance. Additional strips placed upstream of each VMS on comparable stretches of road would provide useful control data – provided they are sufficiently far from the sign to avoid influence and are unaffected by confounding features such as the road layout, in this case the Navidale roundabout.

Other recommendations to improve data collection and analysis include:

• Recording exact VMS activation times rather than approximate deployment weeks. This would allow each vehicle record to be assigned precisely to VMS on/off periods, enabling finer-grained modelling.
• More detailed weather logging, ideally including rainfall, temperature and snow depth, to account more accurately for environmental influences on driver behaviour.

While survey design will always be constrained by cost, even modest improvements to monitoring could help reveal whether mobile VMS do influence vehicle speeds.

It is also worth noting that vehicle speed is not the only measure of DVC risk. The signs may have resulted in increased driver awareness that meant DVCs were averted, without affecting vehicle speeds. Determining whether this is the case when VMS are deployed in this manner is likely to be impossible, as it will require deployment for long enough to observe an effect on actual DVCs. This would likely suffer from driver complacency, where the VMS looses its effect on driver behaviour over time, thus negating the results.

TROC reports not resulting in a deer carcass uplift

Despite changes to the requests for data, it was not clear how many TROCs supplied records for DVC reports where a carcass was not uplifted. In many cases this was due to the systems already in place for managing and collating the data. Changes to existing systems will not be quick to achieve.

This likely means that there is an element of bias in the data, as those TROCs that supplied such ‘no trace’ records would have relatively more records in their area in the DVC database. It is not thought that this will have influenced overall trends, but may have had localised impacts on the results of this analysis.

Future research should continue to ensure that all relevant data is collected by and collated from TROCs, for inclusion in the DVC database. Over time the number of TROCs supplying ‘no trace’ records should increase.

The effect of changes to the road network

All analyses in this study have been conducted using the road network as it was in 2021, rather than attempting to reconstruct the network as it existed at the time of each DVC. This is both pragmatic and appropriate: identifying historic network layouts for every incident and adjusting the analysis accordingly would be impractical and would provide little additional value. More importantly, analyses should be valid at the point they are undertaken, and therefore should reflect the present-day road network.

One implication of this approach is that some historic DVCs may no longer fall on the current network, while others that occurred on one road type may now appear on another. To minimise this issue, many of the more detailed analyses described in this report have been designed to match DVCs to roads by trunk road section and road number. Any misclassified incidents are expected to be rare and unlikely to materially affect results or their interpretation.

The use of the 2021 trunk road network data for this report will have had a negligible impact on the results of the latest analyses, as the trunk road network had only changed substantially in one location, which had few DVCs. However, as the trunk road network continues to evolve, it will be desirable to use updated trunk road network mapping for future DVC analyses, when this is available.

Because the trunk road network is evolving, analyses assigning DVCs to roads or sections of roads should be regarded as a snapshot in time, rather than a fully consistent comparison across years. Thus, direct year-to-year comparisons or specific locational comparisons may not always be reliable, but given the limited scale of the changes to the trunk road network compared with the extent of the network itself, the general conclusions remain robust.

Marker post datasets

For some DVC records, reference to the nearest marker post is the only available location information. The previous study began compiling a marker-post dataset, correcting existing records and adding previously unmapped locations (Lush and Lush, 2023). This work was continued during the present study.

To date, updates to the dataset have been piecemeal, meaning that coverage remains incomplete. Where marker posts are the sole location reference, this lack of comprehensive data is a barrier to effective DVC collation. Adding marker posts on an ad hoc basis also slows the overall process.

A more systematic method has now been developed to make completion of the dataset more efficient. This involves using Google Street View to record the latitude and longitude of marker posts visible on the roadside. These are then added directly to the dataset. Mapping whole road sections in this way – recording the number and position of each marker post sequentially – would be more effective than continuing to add posts individually as required. This approach also resolves cases where marker post numbers are illegible, as numbers can be inferred from neighbouring posts, and even allows approximate placement where posts are obscured, as they are typically 100 m apart.

Developing the dataset in this systematic manner would be significantly more efficient than the current incremental approach. Moreover, a complete marker-post dataset would likely have wider value for Transport Scotland and the relevant TROCs. It is therefore recommended that funding be secured to support the creation of a comprehensive, national marker-post dataset using this method.

The potential of animal rescue centres to collect DVC data

The SSPCA operates nationally in Scotland and has provided records of incidents reported and attended by staff. These data have been consistent and invaluable for DVC analysis since the 2008.

However, several other independent animal rescue centres operate within Scotland that do not regularly provide data. New Arc Wildlife Rescue provided valuable data for 2023 and 2024, which included many additional records not occurring in other sources. This highlights the already recognised under-recording of DVCs and demonstrates the potential value of data from animal rescue centres.

One of the obstacles to obtaining DVC data from wildlife rescue centres is the lack of a standardised system for recording such data. Most wildlife rescue centres operate on a limited budget and are therefore unlikely to develop a suitable system themselves. There would also be benefit in having a standardised central system where information from all wildlife rescue centres could be held, making it more readily available for analysis.

Options for a system range from basic with limited functionality, through to complex and functionality rich. A basic solution could involve creating a QField project that could be deployed to interested wildlife rescue centres, with data being fed back to a central body and perhaps published via QGIS Cloud. This would have no license costs, meaning that only development, ongoing maintenance and data management would need funding.

A more complex solution could involve a bespoke mobile application for use in the field linked to a centralised database and a web-based interface for managing and interrogating the data. This could include the use of user logins to ensure that staff only saw records related to their organisation. It could also encompass functionality for recording admissions, treatment and rerelease where appropriate. However, this would require significant investment in development and maintenance, which would need to be weighed against the number of users, and the volume and value of the data captured.

Further work on mobilising data from animal rescue centres is recommended. This should initially involve exploratory work to identify the relevant organisations, their interest in a solution, their requirements, and potential sources of funding for development, maintenance and management. This should also aim to gauge the volume of data that could be mobilised through the solution.

Increasing data mobilisation and improving quality

Several datasets obtained during this contract had issues that limited their inclusion in the analysis, notably lack of dates and accurate locations. This is unfortunate, as it reduces the value of the data. Other data could not be obtained or used, either because they were not supplied or due to licensing issues.

Future contracts should aim to continue to both improve data quality and mobilisation. Where DVCs are recorded by organisations with internal data management processes, efforts should be made to influence data capture and management, to improve record quality and increase record usefulness.

For other, less structured recording of DVCs, the use of an established mobile app is recommended. This should allow the date and location of the incident to be accurately recorded, along with other key information constituting a valid biological record. Data verification mechanisms are also valuable, especially where this results in usable species-specific data. The Mammal Society’s Mammal Mapper app is strongly recommended, as this specifically allows for recording of roadkill and the data is supplied for use in this analysis. The iRecord and iNaturalist apps can also be used, though the comment on any DVC records made should state that it is roadkill, the data should be licensed for commercial use and the coordinates should be unobscured, to ensure that the data can be used in this analysis.

Annexes