全球石冰川发育区

石冰川是含冰的寒冻风化岩屑物或冰碛物，由于内部冰的形变驱动，在重力和冻融作用下沿山谷或坡面向下缓慢蠕动的舌（叶）状堆积体，是山地地区的典型冰缘地貌，具有独特的地貌特征，包括明显的脊-沟结构、鲜明的流动特征、陡峭的前缘等，在古气候、冻土建模、地质灾害、生态水文等多方面都具有重要价值。近年来在遥感等技术推动下石冰川编目数量激增，并在全球范围内的多年冻土分布建模中显示出良好的应用前景。但是目前的石冰川的识别和编目仍依赖于劳动密集型的目视解译，只有很少的研究估算了大尺度的石冰川分布，且这些编目数据集没有被综合利用，全球范围内的石冰川分布仍然处于相对空白的状态。各种高分辨率环境数据集的出现和机器学习算法的不断进步，为预测石冰川分布的研究提供了坚实的数据和方法基础。本研究整合现有的石冰川编目数据集，选取其中的完整型，利用1km×1km的网格对石冰川进行裁剪，利用随机森林算法，通过建立网格内的石冰川面积占比与环境变量的联系，在全球范围内进行了石冰川发育区模拟（rock glacier development area, RDA），空间分辨率为1km，投影坐标系为Krasovsky_1940_Albers。其中所用到的环境变量包括海拔、坡度、年平均气温、正负积温、土壤物理属性（密度，砂含量，粉砂含量，碎石含量，黏土含量）。预测结果表明，目前全球潜在的石冰川发育区范围（RDA≥0.01涉及的栅格面积）为204万平方千米，石冰川发育面积（RDA≥0.01的总面积）为14.1万平方千米，验证表明具有较高的精度（R²=0.60，RMSE=0.05）。Rock glaciers are tongue-shaped (lobate) accumulations composed of frost-weathered rock debris or moraines. Driven by internal ice deformation and slowly creeping down valleys or slopes under the combined effects of gravity and freeze-thaw processes, they are typical periglacial landforms in mountainous areas, featuring distinct geomorphic characteristics including prominent ridge-gully structures, clear flow signatures, and steep frontal margins. Rock glaciers hold important scientific values in multiple fields such as paleoclimatology, permafrost modeling, geological hazard research, and eco-hydrology. In recent decades, driven by advancements in remote sensing and related technologies, the number of published rock glacier inventories has increased dramatically, and rock glaciers have demonstrated excellent application potential in global permafrost distribution modeling. However, current rock glacier identification and inventory work still rely heavily on labor-intensive visual interpretation. Only a small number of studies have estimated large-scale rock glacier distributions, and these existing inventory datasets have not been comprehensively utilized. As a result, the global-scale distribution of rock glaciers remains relatively poorly characterized. The proliferation of high-resolution environmental datasets and the rapid progress of machine learning algorithms have provided a robust data and methodological foundation for research predicting rock glacier distributions. This study integrates existing rock glacier inventory datasets, selects complete and valid inventory subsets, and crops the identified rock glaciers using 1km×1km grids. Employing the Random Forest algorithm, we established the relationship between the proportion of rock glacier area within each grid cell and a set of environmental variables, and then simulated the global rock glacier development areas (RDA) with a spatial resolution of 1km under the Krasovsky_1940_Albers projected coordinate system. The environmental variables utilized in this study include elevation, slope, mean annual air temperature, positive and negative accumulated temperatures, and soil physical properties (density, sand content, silt content, gravel content, and clay content). The prediction results indicate that the current global potential rock glacier development area (total grid cell area where RDA ≥ 0.01) is 2.04 million square kilometers, and the total rock glacier development area (total accumulated area where RDA ≥ 0.01) is 141,000 square kilometers. Validation results show high prediction accuracy with a coefficient of determination (R²) of 0.60 and a root mean square error (RMSE) of 0.05.

搜集汇总

背景与挑战

背景概述

该数据集为全球石冰川发育区模拟数据，基于整合的石冰川编目和环境变量，使用随机森林算法预测生成，空间分辨率为1km，时间范围覆盖2000年至2024年。它提供了全球潜在石冰川发育区的面积估算（约204万平方千米）和发育面积（约14.1万平方千米），验证精度较高（R²=0.60），适用于冻土建模、气候研究和地质灾害评估等领域。