Species distribution models associtaed with journal article The utility of dynamic forest structure from GEDI lidar fusion in tropical mammal species distribution models

The dataset includes species distribution models (SDMs) associtaed with published journal article The utility of dynamic forest structure from GEDI lidar fusion in tropical mammal species distribution models (https://doi.org/10.3389/frsen.2025.1563430). We modeled the distribution of 47 mammal species found in Borneo and/or Sumatra. Not all species are naturally occurring on both islands. Furthermore, some species, which may be naturally occuring on an island, did not have sufficient observations for distribution modeling on one island or the other. SDM prediction maps for critically endangered species (Bornean orangutan, Sunda pangolin, and Sumatran tiger) will be curated by the relevant IUCN taxonomic authority (i.e., specialist group) and may be made available upon request. See Table 1 in the journal article for details regarding species observations and modeling domains. For analytical purposes we provide cloud optimized GeoTIFF files corresponding to SDM predictions from the Hindcast with GEDI scenario (see journal article for details). Predictions corresponding to the probability of the presence class (1) from both Random Forest (RF) and Generalized Linear Model (GLM) algorithms are included in each file archive (.tar.gz). There are separate files for Borneo and Sumatra predictions since we fit separate models for these regions. For each species and region we made 10 predictions based on a shuffling of the model training/testing data (i.e. bootstrapping). The file archives contain GeoTIFF files which are aggregates of the (up to) 10 predictions per species. We used the mean and standard deviation to aggregate the bootstrap model predictions per pixel. The band order in each file isband1: rf_p1_mn - RF per-pixel meanband2: rf_p1_sd - RF per-pixel standard deviationband3: glm_p1_mn - GLM per-pixel meanband4: glm_p1_sd - GLM per-pixel standard deviation The spatial reference system is EPSG:4326. The spatial resolution (i.e. pixel size) is 0.00080848 decimal degrees which is equivalent to ~90 m. The prediction year (either 2001 or 2021) is included in the file name. All pixel values are scaled by 10000 and stored as Int16 to decrease file size. A value of 0 corresponds to 0.0 probability of occurrence while a value of 10000 corresponds to 1.0 probability of occurrence. Lakes were set to 0 probability of occurrence using the HydroLAKES lake polygons shapefile (https://www.hydrosheds.org/products/hydrolakes). The nodata value is -9999. SDM performance was assessed using the area under the receiver operator characteristic curve (AUCROC) metric and is summarized in Table1.xlsx of the journal article Supplementary Material. For quick visualization we also provide lower resolution, non-georeferenced PNG maps in the file archive sdm_maps_png.tar.gz. These maps show the mean probability of the presence class, as well as the coefficient of variation (standard deviation divided by the mean). There are separate files corresponding to SDM predictions for each region, modeling algorithm (RF and GLM), and year (2001 and 2021).本数据集包含与已发表期刊文章相关的物种分布模型（Species Distribution Models, SDMs），该文章DOI为https://doi.org/10.3389/frsen.2025.1563430，标题为《GEDI激光雷达融合动态森林结构在热带哺乳动物物种分布模型中的应用价值》。我们针对婆罗洲和/或苏门答腊的47种哺乳动物开展了分布建模。并非所有物种均在两个岛屿上自然分布；此外，部分在某一岛屿自然分布的物种，因观测数据不足，无法在该岛屿完成分布建模。极危物种（婆罗洲猩猩、巽他穿山甲、苏门答腊虎）的SDM预测图将由国际自然保护联盟（IUCN）相关物种分类权威机构（即物种专家小组）审核整理，可根据申请提供。有关物种观测与建模范围的详细信息，请参阅期刊文章中的表1。 为便于分析，我们提供了与GEDI情景回溯预测（详见期刊文章）得到的SDM预测结果对应的云优化GeoTIFF文件。每个.tar.gz归档文件均包含随机森林（Random Forest, RF）和广义线性模型（Generalized Linear Model, GLM）两种算法生成的物种存在类别（1）的概率预测结果。因婆罗洲与苏门答腊的模型为独立构建，故两者的预测结果分文件存储。针对每个物种与区域，我们通过对模型训练/测试数据进行随机重抽样（即自助法（bootstrapping））生成了10组预测结果。归档文件中的GeoTIFF文件为每个物种最多10组预测结果的聚合产物，我们通过逐像素计算均值与标准差完成了自助法预测结果的聚合。每个文件的波段顺序如下：波段1：rf_p1_mn — 随机森林逐像素均值；波段2：rf_p1_sd — 随机森林逐像素标准差；波段3：glm_p1_mn — 广义线性模型逐像素均值；波段4：glm_p1_sd — 广义线性模型逐像素标准差。 本数据集的空间参考系统为EPSG:4326，空间分辨率（即像素大小）为0.00080848十进制度，约合90米。文件名中包含预测年份（2001年或2021年）。 所有像素值均经过10000倍缩放并以Int16格式存储以减小文件体积。像素值0对应物种存在概率为0.0，像素值10000对应存在概率为1.0。我们通过HydroLAKES湖泊多边形矢量文件（https://www.hydrosheds.org/products/hydrolakes）将湖泊区域的存在概率设为0。无数据值为-9999。 SDM模型性能采用受试者工作特征曲线下面积（Area Under the Receiver Operating Characteristic Curve, AUCROC）指标进行评估，相关结果汇总于期刊文章补充材料的Table1.xlsx中。 为便于快速可视化，我们还在归档文件sdm_maps_png.tar.gz中提供了低分辨率、未进行地理配准的PNG地图。这些地图展示了物种存在类别的平均概率，以及变异系数（标准差除以均值）。每个文件分别对应不同区域、建模算法（RF与GLM）以及年份（2001年与2021年）的SDM预测结果。