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]本仓库包含用于分析以下论文所用的数据与代码：Annika M. Zuleger、Martina M. Viti、Luise Quoss、Filipe S. Dias、Luís Borda-de-Água、Miguel N. Bugalho、Henrique M. Pereira（2025）：《利用多传感器高分辨率无人机（UAV）数据绘制异质山地景观中草食动物可获取生物量分布图》，发表于《遥感科学》（Science of Remote Sensing），DOI：https://doi.org/10.1016/j.srs.2025.100302 完整变更日志：https://github.com/AMZuleger/Mapping_herbivore-accessible_biomass_UAV.git 摘要 草食动物可获取生物量（Herbivore-accessible biomass, HAB）指高度低于2米的地上生物量，包括叶片与柔软枝条，是理解生态系统功能的关键指标，但目前其量化程度仍较低。本研究借助高分辨率归一化差异植被指数（NDVI）、激光雷达（LiDAR）、地形数据与野外实测数据，对佩内达-热尔国家公园（Peneda-Gerês National Park）不同生境中的草食动物可获取生物量进行了估算。广义加性混合模型（Generalized Additive Mixed Models, GAMMs）揭示了归一化差异植被指数、植被高度、地形及不同植物类型的结构指标对草食动物可获取生物量的生境特异性影响。模型通过保留20%野外数据进行交叉验证评估。总草食动物可获取生物量模型表现优异（偏差解释率=0.77，RMSE₂₀=172.38 g/m²），灌木模型表现略逊（偏差解释率=0.71，RMSE₂₀=410.21 g/m²），草本模型拟合与精度处于中等水平（偏差解释率=0.69，RMSE₂₀=34.25 g/m²）。总草食动物可获取生物量的平均值为1.31±0.83 吨/公顷，其中灌木类生物量占主导（1.02 吨/公顷），草本类生物量为0.14 吨/公顷。草食动物可获取生物量密度因生境而异：灌丛地最高（可达1.83 吨/公顷），栎林最低（0.85 吨/公顷），而农业区域的草本类草食动物可获取生物量最高（0.68 吨/公顷）。上述数值远低于其他研究报道的灌木生物量估算值（例如最高可达30 吨/公顷），这反映了本研究聚焦于高度低于2米的活体生物量。预测不确定性较低（变异系数：22%~34%），优于其他研究报道的最高190%的不确定性，凸显了结合光谱与结构数据进行精细尺度饲草估算的优势。本研究首次提供了该区域的空间显式草食动物可获取生物量估算结果，可为草食动物生态学研究与管理提供支撑。 数据结构 Mapping_HAB.Rmd — 用于完整分析的R笔记本文档。该代码可运行所有模型并生成全部结果与图表（地图除外，地图通过QGIS制作）。 Mapping_HAB.html — 由Mapping_HAB.Rmd生成的HTML文档。 Data.zip — 包含分析所需的全部数据集： biomass_data_2024.csv — 2024年5月采集的野外生物量实测数据，以及通过QGIS分区统计得到的样地级遥感指标。 grid_data.csv — 整个研究区10×10米网格上的遥感指标，用于模型预测。 Figures_main.zip — 包含主文档的全部图表： Figure_1_Study_Location_and_Sampling_Design.png — 研究区域与采样设计图。 Figure_2_Variable_Importance.png — 基于广义加性混合模型得到的总、灌木及草本类草食动物可获取生物量的各变量F统计量图。 Figure_3_Cross_validation_results.png — 通过保留交叉验证得到的总（上）、灌木（左下）及草本类（右下）草食动物可获取生物量的预测值与实测值对比图。 Figure_4_HAB_Predictions_CV_Maps.png — 整个研究区总、灌木及草本类草食动物可获取生物量的预测结果与预测变异系数图。 Figures_supplemental.zip — 包含补充文档的全部图表： Figure_S2_Height_distribution.png — 不同植被类型野外实测的平均高度与最大高度直方图。 Figure_S4_Remote_Sensing_Metrics.png — 最终模型中使用的全部遥感指标的分布图。 Figure_S5.1_Correlation_plot.png — 所有变量间的相关系数图。 Figure_S2.2_Dissimilarity_tree.png — 所有变量的相异树状图。 Figure_S6_Field_measurements.png — 不同生物量类型与生境类别的野外实测草食动物可获取地上生物量数据。 Figure_S8_Variable_importance_forb_grass.png — 基于广义加性混合模型得到的杂类草与禾草类草食动物可获取生物量的各变量F统计量图。 Figure_S9_Cross_validation_results_forb_grass.png — 通过保留交叉验证得到的杂类草（左）与禾草类（右）草食动物可获取生物量的预测值与实测值对比图。 Figure_S11_HAB_Predictions_CV_Maps_herb.png — 整个研究区杂类草与禾草类草食动物可获取生物量的预测图与变异系数图。 Figure_S12.1_Observed_pred_biomass_distribution.png — 不同生境与植被类型的实测与预测生物量分布图。 Figure_S12.2_Pred_biomass_CV_distribution.png — 不同生境与植被类型的预测生物量与变异系数分布图。 Figure_S12.3_Historgramms_HAB_distribution.png — 不同植被类型的草食动物可获取生物量预测值、标准误差与变异系数直方图。 作者 Annika M. Zuleger*、Martina M. Viti、Luise Quoss、Filipe S. Dias、Luís Borda-de-Água、Miguel N. Bugalho、Henrique M. Pereira *通讯作者：Annika Mikaela Zuleger，德国综合生物多样性研究中心（iDiv）哈雷-耶拿-莱比锡分中心，德国莱比锡普施大街4号，邮编04103，电子邮箱：[e-mail]