Global Human Presence Intensity Dataset (2017)

Overview Understanding global patterns of human presence is crucial for monitoring anthropogenic pressures on ecological integrity, optimizing tourism management, and informing decision-making in various domains. However, existing datasets on human presence are insufficient and primarily limited to local scales. We combined massive geotagged microblogs and multi-covariates to infer the global human presence in 2017 at a fine spatial resolution of 0.01 degrees. Specifically, we proposed a Human Presence Indicator (HPI) to quantify human presence. It categorizes the intensity of human presence at a location in four levels based on year-long statistics of geotagged data. HPI-0, HPI-1, HPI-2, and HPI-3 represent no human presence, occasional human presence, frequent human presence, and sustained human presence, respectively. The model achieved a macro-F1 score of 0.72 on a test set comprising over 1.9 million grids in China and 0.84 on the manually labeled samples available worldwide. Cross-validation with external datasets, including geotagged social media data from X and global human settlement and population data, corroborated the model's effectiveness. Data Directory Contents The complete dataset is organized into four main components, each available as a separate compressed .zip file for download: Gridded HPI Data: Contains the primary data product showing global human presence intensity (on a 0-3 scale) at a 0.01-degree resolution in GeoTIFF format. Covariate Layers: Contains all 76 predictor variables used for model training, resampled to the same 0.01-degree resolution. Validation Data: Contains the manually labeled samples that were used to assess model performance. Model Files: Contains the final trained random forest model in .joblib format, ready for use by Python practitioners. Citation Please cite both the dataset and the related article when using these data: Luo, P., Yi, J., Du, Y., Huang, S., Wang, N., Tu, W., Hu, D., & Wei, H. (2025). Global Human Presence Intensity Dataset (2017) (1.2) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.16499251 Luo, P., Yi, J., Du, Y., Huang, S., Wang, N., Tu, W., Hu, D., & Wei, H. (2025). Mapping global human presence for nature conservation using geotagged social media data. Biological Conservation, 311, 111404. https://doi.org/10.1016/j.biocon.2025.111404概述 明晰人类活动的全球分布模式，对于监测人类活动对生态完整性的胁迫压力、优化旅游管理，以及为多领域决策提供依据均至关重要。然而，现有人类活动相关数据集仍存在不足，且多局限于局地尺度。 本研究结合海量带地理标签的微博（geotagged microblogs）数据与多协变量，以0.01度的精细空间分辨率推演得到2017年全球人类活动分布情况。具体而言，本研究提出人类活动强度指标（Human Presence Indicator, HPI）以量化人类活动水平。该指标基于全年带地理标签数据的统计结果，将某一位置的人类活动强度划分为四个等级：HPI-0、HPI-1、HPI-2、HPI-3分别代表无人类活动、偶发人类活动、高频人类活动与持续人类活动。 该模型在包含中国境内超190万个网格单元的测试集上取得了0.72的宏F1得分（macro-F1 score），在全球范围的人工标注样本集上则达到了0.84的宏F1得分。通过包含X平台带地理标签社交媒体数据、全球人类聚居区与人口数据在内的外部数据集进行交叉验证，进一步证实了该模型的有效性。 数据集目录内容 完整数据集包含四个核心组成部分，各部分均以独立的压缩.zip文件形式提供下载： 网格化HPI数据：包含核心数据产品，以GeoTIFF格式（GeoTIFF）存储，以0.01度分辨率展示全球人类活动强度（等级范围0-3）。 协变量图层：包含模型训练所用的全部76个预测变量，均重采样至0.01度的统一分辨率。 验证数据集：包含用于评估模型性能的人工标注样本。 模型文件：包含最终训练完成的随机森林模型，格式为.joblib，可供Python开发者直接调用。 引用说明 使用本数据集时，请同时引用该数据集及相关研究论文： Luo, P., Yi, J., Du, Y., Huang, S., Wang, N., Tu, W., Hu, D., & Wei, H. (2025). Global Human Presence Intensity Dataset (2017) (1.2) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.16499251 Luo, P., Yi, J., Du, Y., Huang, S., Wang, N., Tu, W., Hu, D., & Wei, H. (2025). Mapping global human presence for nature conservation using geotagged social media data. Biological Conservation, 311, 111404. https://doi.org/10.1016/j.biocon.2025.111404