Understanding factors that influence spatial and temporal variation in wildlife population trends 19 and distribution is crucial for effective conservation and management. Pronghorn (Antilocapra americana) are a wildlife species that exhibit substantial annual fluctuations in overall 21 abundance, density, and distribution – patterns often influenced by both climate and landscape 22 conditions. In this study, we sought to connect commonly collected survey data to spatial 23 ecological questions by applying annual Density Surface Models (DSMs) to four years of aerial 24 surveys conducted between 2015 and 2021 in northwestern South Dakota, USA. As a result, we 25 (1) derived design-based annual abundance estimates under an assumption of constant detection 26 and (2) identified relationships between environmental factors and variation in pronghorn density 27 across space (i.e., spatial abundance). Over the study period, pronghorn abundance nearly 28 doubled, increasing from a mean of 7,198 in 2015 (95% CI: 65,21 – 7,945) to 12,672 individuals 29 in 2019 (95% CI: 11,735 – 13,684). We also found a consistent relationship between the spatial 30 distribution of pronghorn density and the number of snow days in all survey years, indicating 31 shifts in pronghorn distribution toward areas with less snow. These findings highlight how 32 spatially explicit abundance models can provide insight into climate–habitat–density 33 relationships that are not apparent from abundance trends alone. Our results also help wildlife 34 managers develop adaptive strategies in the face of shifting climatic and landscape conditions by 35 identifying priority areas for actionable population and habitat conservation.

DAU 1, South Dakota (2015–2021)

Associated publication: McDevitt et al. (in press). Linking aerial survey data to spatial abundance: snow as a predictor of pronghorn distribution. Ecosphere.

Dataset DOI: https://doi.org/10.5061/dryad.dr7sqvbfn

Overview

This repository contains aerial pronghorn (Antilocapra americana) survey data and analysis code for South Dakota Game, Fish and Parks (SDGFP) Deer Administrative Unit 1 (DAU 1) across four survey years: 2015, 2017, 2019, and 2021. DAU 1 encompasses five wildlife management units in northwestern South Dakota (Harding, Perkins, and Corson counties): units 20A, 35A, 35B, 53A, and 53B.

Surveys were conducted as fixed-wing aerial line transect surveys. Observers recorded pronghorn group locations, group size, and sex composition for all groups detected within approximately 400 m of the flight path.

Spatial resolution of this archive

Detections are released at the transect-segment level, not as individual group locations. Each observation is linked by segment_id to its transect segment in pronghorn_dau1_segments.csv, where the segment centroid coordinates are given. Coordinates of individual detected groups were collected during surveys but are not included in this archive; they are not used by any analysis in the associated publication, and segment-level linkage is sufficient to reproduce all published results. Researchers needing finer spatial resolution should contact South Dakota Game, Fish and Parks.

Segment centroids describe the survey flight path rather than animal locations. Segments average approximately 2 km in length and are drawn from a standardized survey design covering a study area of roughly 21,000 km².

Analysis Summary

Data were analyzed using a two-stage density surface modeling (DSM) framework (Miller et al. 2013).

Stage 1 — Distance sampling detection function

A uniform key detection function with no adjustment terms was fitted using the Distance R package (Miller et al. 2019). Truncation distance: 400 m.

Important note on perpendicular distances: Exact perpendicular distances from the transect to each detected group were not recorded during surveys. For the uniform key function, the probability of detection is assumed to be equal at all distances within the truncation width. Accordingly, each observation is assigned a random distance drawn from Uniform(0, 400 m). The column perpendicular_distance_m in pronghorn_dau1_observations.csv reflects these randomly assigned values and is included for completeness; it does not represent a field measurement.

Stage 2 — Density surface model (DSM)

A generalized additive model (GAM) with a Tweedie response distribution was fitted to estimated abundance per segment as a function of geographic location and environmental covariates. Nine candidate models were evaluated per survey year using AIC. All models include a geographic smooth [s(latitude, longitude)]; candidate covariates were snow cover, drought index (PDSI), vegetation greenness (NDVI), and human modification.

Data Files

File	Description
`pronghorn_dau1_segments.csv`	Segment-level data. One row per transect segment per year. Contains effort, segment centroid coordinates (WGS84), detection counts, and all environmental covariates.
`pronghorn_dau1_observations.csv`	Observation-level data. One row per detected pronghorn group (size > 0 only). Segments with no detections appear only in the segments file (`n_groups_detected = 0`).
`dau1_study_area.geojson`	DAU 1 wildlife management unit boundaries (GeoJSON, WGS84). Units: 20A, 35A, 35B, 53A, 53B.
`data_dictionary.csv`	Column definitions, units, and source descriptions for all fields in the segment and observation CSV files.

Environmental Covariates

All covariates are measured at the segment level and were extracted via Google Earth Engine.

Covariate	Column	Source	Description
Snow cover	`snow_days`	SNODAS	Days with snow on ground, June 1 of the preceding year through May 31 of the survey year
Drought index	`pdsi`	GRIDMET (Abatzoglou 2013)	Palmer Drought Severity Index, segment median. Negative values indicate drought conditions.
Vegetation greenness	`ndvi`	Landsat 8 (USGS)	Normalized Difference Vegetation Index × 10,000, segment median, peak growing season
Human modification	`human_modification`	Kennedy et al. (2019)	Global Human Modification Index (gHM), segment median. Range: 0 (no modification) to 1 (maximum modification). Static layer representing cumulative human influence as of ~2016.

Coordinate Reference System

All geographic coordinates in the CSV files are in WGS84 (EPSG:4326, decimal degrees). The study area GeoJSON uses the same CRS. Internal analysis used UTM Zone 14N (EPSG:32614) for distance and area calculations.

Segment Length

Transect segments vary in length. Segments shorter than 1,600 m were excluded from analysis; retained segments range from approximately 1,630 m to 3,173 m (mean ≈ 2,027 m). Segment length is given per segment in segment_length_m and is used as survey effort in the detection and density surface models.

Missing Values

Missing values are coded as NA. In pronghorn_dau1_segments.csv, snow_days is missing for 24 of 17,589 segments (0.1%) and human_modification for 45 segments (0.3%); in pronghorn_dau1_observations.csv, snow_days is missing for 2 of 3,752 observations. These reflect gaps in the underlying remotely sensed layers at the segment centroid. Segments with missing covariate values are dropped by mgcv when fitting models that include the affected covariate, so candidate models differ slightly in the number of segments used.

R Code Pipeline

Scripts are numbered in execution order and located in the r/ directory. Run sequentially in a single R session, or source each in turn. Scripts 01–04 write intermediate .rds files to data/SavedOutputs/ and can be skipped on subsequent runs by loading those files directly.

Raw survey data (Excel/shapefiles)
        ↓
01_ReadRawData.R         — Read and spatially format survey observations
        ↓
02_DataForGEE.R          — Export shapefiles for GEE; read back covariate CSVs
        ↓ (one-time; outputs already in data/FromGEE/)
03_ReadCovariateData.R   — Consolidate GEE covariates → data/SavedOutputs/cov*.rds
        ↓
04_FormatSegObs.R        — Build Distance-package data structures
                            → data/SavedOutputs/dsdt*.rds
                            → data/SavedOutputs/segs*_fmt.rds
        ↓
05_RunDSM.R              — Fit detection functions + 9 candidate DSMs per year
                            → Figures/dsm*.rds (top models)
                            → Figures/Ndfm*.rds (abundance estimates)
                            → Figures/model_selection_table_complete.csv
                            → Figures/snow_effect_summary.csv
        ↓
06_Figure3.R             — Snow partial effects figure (Figure 3, 300 DPI PNG)
07_MakeTable.R           — Model selection table (Table 1, formatted Excel)
08_FigureBuild.R         — Supplementary abundance trend and covariate figures

Script Descriptions

Script	Description
`01_ReadRawData.R`	Reads raw aerial survey data from Excel (2015, 2017) and shapefile (2019, 2021). Filters to DAU 1 counties/units, standardizes column names, fills zero-detection transects, and creates spatial observation objects. Loads DAU 1 boundary and computes unit areas.
`02_DataForGEE.R`	Exports observation-level point shapefiles for covariate extraction in Google Earth Engine. Also reads back snow covariate CSVs (one-time step; outputs already saved).
`03_ReadCovariateData.R`	Reads per-segment covariate CSVs from GEE (snow, PDSI, NDVI, human modification) and joins them into annual covariate objects saved as `.rds`.
`04_FormatSegObs.R`	Loads raw segment shapefiles and observation data, extracts coordinates, filters short segments (< 1,600 m), joins covariates, and builds the combined segment–observation data frames (`dsdt`) required by the `Distance` and `dsm` packages.
`05_RunDSM.R`	Core analysis script. Fits a uniform-key detection function for each year, then fits 9 candidate DSMs per year (null geographic model + single and additive combinations of snow, PDSI, NDVI, and human modification). Compares models by AIC, extracts snow effect sizes, and checks covariate collinearity.
`06_Figure3.R`	Generates Figure 3: a 2×2 panel of GAM snow smooths (partial effect on log density) from each year's top model, with 95% CI shading and observed-value rug marks. Saved as a 300 DPI PNG.
`07_MakeTable.R`	Generates Table 1: a styled Excel workbook with model selection statistics (AIC, ΔAIC, R², deviance explained) for all 36 candidate models.
`08_FigureBuild.R`	Supplementary figures: total DAU 1 abundance trend (2015–2021) and unit-level abundance trends by wildlife management unit.

Key Raw Inputs

File	Description
`data/Archive Aerial Pronghorn Survey Data.xlsx`	Raw 2015 & 2017 survey observations
`data/AerialPronghornSurveyArchive_2019.shp`	Raw 2019 observations
`data/AerialPronghorn_DAU1Observations_2021.xlsx`	Raw 2021 observations
`data/SD_Spatial_EN.gpkg`	South Dakota wildlife unit boundaries
`data/FromGEE/`	Covariate CSVs extracted from Google Earth Engine
`data/SavedOutputs/segs*.rds`	Pre-built segment shapefiles (one per year)
`data/SavedOutputs/obs*.rds`	Pre-built observation files (one per year)

Key Outputs

File	Description
`data/SavedOutputs/dsdt*.rds`	Formatted distance sampling data (input to `ds()` and `dsm()`)
`Figures/dsm*.rds`	Top-ranked DSM per year
`Figures/Ndfm*.rds`	Unit-level abundance estimates
`Figures/model_selection_table_complete.csv`	Full model selection table (all 36 models)
`Figures/snow_effect_summary.csv`	Snow effect sizes (density ratios) by year
`Figures/Figure3_snow_partial_effects.png`	Manuscript Figure 3
`Figures/Table1_ModelSelection.xlsx`	Manuscript Table 1

Software

R ≥ 4.4
Key packages: tidyverse, sf, Distance, dsm, mgcv, openxlsx, terra

References

Abatzoglou, J.T. (2013). Development of gridded surface meteorological data for ecological applications and modelling. International Journal of Climatology, 33(1), 121–131. https://doi.org/10.1002/joc.3413

Kennedy, C.M., et al. (2019). Managing the middle: A shift in conservation priorities based on the global human modification gradient. Global Change Biology, 25(3), 811–826. https://doi.org/10.1111/gcb.14549

Miller, D.L., et al. (2013). Spatial models for distance sampling data: Recent developments and future directions. Methods in Ecology and Evolution, 4(11), 1001–1010. https://doi.org/10.1111/2041-210X.12105

Miller, D.L., et al. (2019). Distance sampling in R. Journal of Statistical Software, 89(1), 1–28. https://doi.org/10.18637/jss.v089.i01

Contact

Molly McDevitt, PhD
Blackfoot Challenge

molly@blackfootchallenge.org

Data from: Linking aerial survey data to spatial abundance: Snow as a predictor of pronghorn distribution

Data files

Abstract

DAU 1, South Dakota (2015–2021)

Overview

Spatial resolution of this archive

Analysis Summary

Stage 1 — Distance sampling detection function

Stage 2 — Density surface model (DSM)

Data Files

Environmental Covariates

Coordinate Reference System

Segment Length

Missing Values

R Code Pipeline

Script Descriptions

Key Raw Inputs

Key Outputs

Software

References

Contact

Data from: Linking aerial survey data to spatial abundance: Snow as a predictor of pronghorn distribution

Data files

Abstract

README: Data from: Linking aerial survey data to spatial abundance: Snow as a predictor of pronghorn distribution

DAU 1, South Dakota (2015–2021)

Overview

Spatial resolution of this archive

Analysis Summary

Stage 1 — Distance sampling detection function

Stage 2 — Density surface model (DSM)

Data Files

Environmental Covariates

Coordinate Reference System

Segment Length

Missing Values

R Code Pipeline

Script Descriptions

Key Raw Inputs

Key Outputs

Software

References

Contact