Mapping herbivore-accessible biomass across a heterogeneous mountain landscape using multisensor high-resolution UAV data

This repository contains data and code used for analysis of the publication Annika M. Zuleger, Martina M. Viti, Luise Quoss, Filipe S. Dias, Luís Borda-de-Água, Miguel N. Bugalho, Henrique M. Pereira (2025): Mapping herbivore-accessible biomass across a heterogeneous mountain landscape using multisensor high-resolution UAV data. Science of Remote Sensing. DOI: https://doi.org/10.1016/j.srs.2025.100302 Full Changelog: https://github.com/AMZuleger/Mapping_herbivore-accessible_biomass_UAV.git Abstract Herbivore-accessible biomass (HAB), defined as aboveground biomass under 2 m, including leaves and soft branches, is a key metric for understanding ecosystem function, but remains poorly quantified. We estimated HAB across diverse habitats in the Peneda-Gerês National Park using high-resolution NDVI, LiDAR, topography and field data. Generalized Additive Mixed Models (GAMMs) revealed habitat-specific effects of NDVI and vegetation height, as well as terrain, and structural metrics across plant types. Models were evaluated using hold-out cross-validation on a 20 % subset of the field data. The total HAB model performed well (Deviance Explained = 0.77, RMSE20 = 172.38 g/m2), while the shrub model performed slightly worse (Deviance Explained = 0.71, RMSE20 = 410.21 g/m2), and the herbaceous model exhibited a moderate fit and accuracy (Deviance Explained = 0.69, RMSE20 = 34.25 g/m2). Average total HAB was 1.31 ± 0.83 tons/ha, dominated by shrubs (1.02 tons/ha) compared to herbaceous HAB (0.14 tons/ha). HAB density varied by habitat, highest in shrublands (up to 1.83 ton/ha) and lowest in oak forests (0.85 tons/ha), while agricultural areas supported the most herbaceous HAB (0.68 tons/ha). These values are substantially lower than shrub biomass estimates reported in other studies (e.g., up to 30 tons/ha), reflecting our focus on live biomass <2 m. Prediction uncertainty was low (CV: 22–34 %), improving on other studies reporting up to 190 %, and highlighting the strength of combining spectral and structural data for fine-scale forage estimation. This study provides the first spatially explicit HAB estimates for the area, supporting herbivore ecology and management. Data structure: Mapping_HAB.Rmd - R Notebook document for the entire analysis. Code runs all models and produces all results and figures (except for maps, which were created using QGIS). Mapping_HAB.html - HTML document created from R Mapping_HAB.Rmd Data.zip - Contains datasets required for the analysis: biomass_data_2024.csv - Field measurements of biomass collected in May, 2024 and remote sensing metrics obtained through Zonal Statistics in QGIS on the plot level grid_data.csv - Remote sensing metrics across the entire study area on a 10 x 10 m grid for predictions Figures_main.zip - Contains all figures of the main document: Figure_1_Study_Location_and_Sampling_Design.png - Study Location and Sampling Design. Figure_2_Variable_Importance.png - F-Statistics per variable obtained from GAMMs for total, shrub and herbaceous HAB Figure_3_Cross_validation_results.png - Predicted vs. observed biomass across total (top), shrub (bottom left), and herbaceous HAB (bottom right) using hold-out cross-validation. Figure_4_HAB_Predictions_CV_Maps.png - HAB predictions and coefficient of variation of the predictions across the entire study area for total, shrub and herbaceous HAB. Figures_supplemental.zip - Contains all figures of the supplemental document: Figure_S2_Height_distribution.png - Histogramms of mean and maximum height measurements from field data per vegetation type. Figure_S4_Remote_Sensing_Metrics.png - Maps of all remote sensing metrics used in the final models. Figure_S5.1_Correlation_plot.png - Correlation plot between all variables. Figure_S2.2_Dissimilarity_tree.png - Dissimilarity tree of all variables. Figure_S6_Field_measurements.png - Field measurements of herbivore-accessible AGB per biomass type and habitat class. Figure_S8_Variable_importance_forb_grass.png - F-Statistics per variable obtained from GAMMs for forb and grass HAB. Figure_S9_Cross_validation_results_forb_grass.png - Predicted vs. observed biomass across forb (left), and grass HAB (right) using hold-out cross-validation. Figure_S11_HAB_Predictions_CV_Maps_herb.png - Biomass prediction maps and Coefficient of variation (CV) across the entire study area for forbs and grasses Figure_S12.1_Observed_pred_biomass_distribution.png - Observed and predicted biomass distribution per habitat and vegetation type. Figure_S12.2_Pred_biomass_CV_distribution.png - Predictions and Coefficient of Variation (CV) per habitat and vegetation type. Figure_S12.3_Historgramms_HAB_distribution.png - Histograms of HAB predictions, standard errors and coefficients of variation across vegetation types. Authors Annika M. Zuleger*, Martina M. Viti, Luise Quoss, Filipe S. Dias, Luís Borda-de-Água, Miguel N. Bugalho, Henrique M. Pereira *Corresponding author: Annika Mikaela Zuleger, German Centre for Integrative Biodiversity Research (iDiv) Halle-Jena-Leipzig, Puschstraße 4, 04103 Leipzig, Germany Email: [e-mail]