Methodology & Data
HMM DecodingHidden Markov Models are statistical models used to predict an unobserved state (behavior) based on observable data (step length and turning angle).
Movement trajectories were segmented using Hidden Markov Models, which infer latent behavioral states from the joint distribution of step lengths and turning angles.
- Sleeping: Nocturnal clusters, minimal displacement.
- Low-energy: Maintenance behaviors, low tortuosity.
- Foraging: High tortuosity, moderate displacement.
- Movement: High-speed relocation, directional persistence.
RSF AnalysisResource Selection Functions are models that estimate the probability of use of a resource unit (like a pixel of habitat) by an organism.
Resource Selection Functions modeled the probability of selection for specific environmental features (water, NDVI, terrain) across behavioral states.
- Covariates: MODIS NDVI, Sentinel-2 landcover, DEM.
- Framework: Logistic Regression (Used vs. Available).
- Output: 30m resolution selection intensity maps.
BACI DesignBefore-After-Control-Impact design is an experimental setup used to measure the effect of an event (like fence removal) by comparing data before and after it happens.
The study follows a Before-After-Control-Impact (BACI) design to evaluate how the removal of the Kariega West fence impacted spatial behavior.
- PRE - HOME RANGE (2022-23): Physical constraint baseline.
- INTERIM (2024): Behavioral lag / Ghost Fence effect.
- POST - NOVEL RANGE (2024-25): Full territory expansion/adaptation.
Data Downloads
Behavioral Points (CSV)
HMM-decoded movement data with behavioral classifications
data/behavioral_points/
RSF Rasters (GeoTIFF)
Habitat selection intensity maps for each elephant, period, and behavior
results/RSF/rasters/{period}/
Note: Rasters are extremely large files (>2MB each). To download specifically selected rasters, please use the interactive map export in the RSF Comparison page or access the results directory.
Study Workflow
Comprehensive analytical pipeline integrating GPS tracking data, behavioral state classification, and habitat selection modeling
Data Collection & Preprocessing
GPS collar data from six elephants across three BACI periods (PRE - HOME RANGE, Interim, POST - NOVEL RANGE). Coordinate transformation from UTM Zone 35S to WGS84, temporal alignment, and quality control filtering.
Hidden Markov Model (HMM) Decoding
A multi-state HMM framework classifies trajectories based on intrinsic movement parameters. States include Sleeping (nocturnal inactivity), Low-energy (passive maintenance and nursing), Foraging (searching), and Movement (relocation).
Behavioral State Space Visualization
Classification based on joint distribution of movement parameters
1. Step Length (m)
Gamma Dist.2. Turning Angle
von Mises Dist.Resource Selection Function (RSF) Analysis
GLM-based habitat selection modeling comparing used vs. available locations. Environmental covariates include vegetation type, distance to water, terrain ruggedness, and elevation at 30m resolution.
Spatial Prediction & Raster Generation
Prediction of habitat selection intensity across study areas (KW and HV) for each elephant, period, and behavioral state. Output as georeferenced GeoTIFF rasters with green-to-red color scale.
Interactive Web Visualization
Web-based platform for exploring trajectories, behavioral patterns, and RSF comparisons. Features include animated playback, density heatmaps, temporal analysis, and side-by-side period comparisons.
Environmental Variable Catalog
The iSSA and RSF models utilize 39 high-resolution environmental covariates. All datasets were orthorectified and resampled to a consistent 30m spatial resolution and projected to UTM Zone 35S.
Topography & Terrain
Procured from NASA SRTM (30m)
- • Elevation.tif (DEM)
- • Slope.tif (Gradient)
- • TRI.tif (Ruggedness Index)
- • Cos_aspect.tif (Cos-Aspect)
Primary Productivity
MODIS MOD13Q1 / Sentinel-2
- • NDVI_2024.tif (Veg. Index)
- • EVI_2024.tif (Enhanced Veg.)
- • LST_2024.tif (Land Surf. Temp)
- • LandCover.tif (LULC Classes)
Bioclimatic Records
WorldClim v2.1 (Downscaled)
- • bio1.tif to bio19.tif (Temp/Precip)
- • (e.g., Annual Mean Temp, Precipitation seasonality, etc.)
Distance Covariates
Euclidean Distance (GIS Layers)
- • dist_water_Sources.tif
- • dist_Valley_Thicket.tif
- • dist_Albany_thckets.tif
- • dist_Grassland_Thicket.tif
- • (+ 7 other vegetation distances)
Exact Layer Manifest (39 Layers)
Funding & Acknowledgments
This study was carried out within the Space It Up project funded by the Italian Space Agency, ASI, and the Ministry of University and Research, MUR, under contract n. 2024-5-E.0 - CUP n. I53D24000060005.
References
Buderman, F. E., et al. (2024, December). Integrated movement models for individual tracking and species distribution data. Methods in Ecology and Evolution 16 (2), 345–361.
de Knegt, H. J., et al. (2010, November). The spatial scaling of habitat selection by African elephants. Journal of Animal Ecology 80 (1), 270–281.
DiStefano, J., et al. (2023, January). Editorial: Mechanistic, machine learning and hybrid models of endocrine regulatory systems. Frontiers in Endocrinology 14.
Elith, J. and J. R. Leathwick (2009, December). Species distribution models: Ecological explanation and prediction across space and time. Annual Review of Ecology 40 (1), 677–697.
Getz, W. M. (2023, June). An animal movement track segmentation framework for forecasting range adaptation. Frontiers in Ecology and Evolution 11.
Gilman, E. and M. Chaloupka (2024, February). Evidence from interpretable machine learning to inform spatial management. Ecosphere 15 (2).
Guisan, A., et al. (2013, October). Predicting species distributions for conservation decisions. Ecology Letters 16 (12), 1424–1435.
Guisan, A. and N. E. Zimmermann (2000, December). Predictive habitat distribution models in ecology. Ecological Modelling 135 (2–3), 147–186.
Jeltsch, F. and V. Grimm (2020, May). Editorial: thematic series "integrating movement ecology with biodiversity research". Movement Ecology 8 (1).
Karp, M. A., et al. (2025, March). Applications of species distribution modeling to support marine resource management. ICES Journal of Marine Science 82 (3).
Lu, W.-X. and G.-Y. Rao (2024, June). Integrated framework combining eco-evolutionary data and SDMs to predict range shifts. MethodsX 12, 102608.
Pohle, J., et al. (2024, February). How to account for behavioral states in step-selection analysis: a model comparison. PeerJ 12, e16509.
Schmitt, S., et al. (2017, August). ssdm: An R package to predict distribution based on stacked SDMs. Methods in Ecology and Evolution 8 (12), 1795–1803.
Signer, J., et al. (2023, December). Simulating animal space use from fitted integrated step-selection functions (iSSF). Methods in Ecology and Evolution 15 (1), 43–50.
Stasinos, S., et al. (2025). Biocube: A multimodal dataset for biodiversity research.
Sullivan, L. L., et al. (2018, October). Mechanistically derived dispersal kernels explain species-level patterns. Ecology 99 (11), 2415–2420.
Talluto, L., et al. (2015, October). Cross-scale integration of knowledge for predicting species ranges. Global Ecology and Biogeography 25 (2), 238–249.
Talluto, L., et al. (2018, June). Multifaceted biodiversity modelling at macroecological scales using Gaussian processes. Diversity and Distributions 24 (10), 1492–1502.
Tuia, D., et al. (2022, February). Perspectives in machine learning for wildlife conservation. Nature Communications 13 (1).