全球5公里分辨率植物功能型网格化生根深度数据集（2015-2100）

本数据集提供了全球5公里分辨率植物功能型网格化生根深度数据（2015-2100）。基于在5公里网格重新整合后的13种植物功能型的1184个生根深度观测值，开发了结合随机森林算法、LPJ-GUESS 模型模拟的生态系统数据以及多种环境变量的混合建模框架，以预测全球根系深度。采用十折交叉验证选取最优模型并利用最优模型预测了2015-2100年全球13种植物功能型的生根深度以及不确定性，时间间隔为5年。其中2015年为历史时期，2020-2100年为未来时期，包含SSP119、SSP245、SSP585三种共享社会经济路径，每种情景下的结果是三种大气环流模式（EC-Earth3, GFDL-ESM4, and MRI-ESM2-0）平均得到的。LPJ-GUESS模型进行了参数敏感性分析和校准工作，模拟的生态系统变量可以准确刻画陆地生态系统状态，建模采用十折交叉验证提升模型精度并选取最优模型进行空间预测，建模精度和验证精度均符合要求。

This dataset provides global gridded root depth data for plant functional types (PFTs) at 5 km resolution, covering the period 2015–2100. A hybrid modeling framework combining random forest algorithm, ecosystem data simulated by the LPJ-GUESS model, and multiple environmental variables was developed based on 1,184 root depth observations across 13 plant functional types reaggregated at the 5 km grid scale, to predict global root depth. Ten-fold cross-validation was adopted to select the optimal model, which was then used to predict the root depth and its uncertainty for 13 global plant functional types at 5-year intervals over 2015–2100. Specifically, 2015 represents the historical period, while 2020–2100 is the future period, including three Shared Socioeconomic Pathways (SSP119, SSP245, and SSP585). Results under each scenario are averaged across three atmospheric general circulation models: EC-Earth3, GFDL-ESM4, and MRI-ESM2-0. The LPJ-GUESS model underwent parameter sensitivity analysis and calibration, and its simulated ecosystem variables can accurately characterize the state of terrestrial ecosystems. During modeling, ten-fold cross-validation was employed to improve model accuracy and select the optimal model for spatial prediction, with both modeling and validation accuracies meeting the required standards.

搜集汇总

背景与挑战

背景概述

本数据集是一个全球范围、5公里分辨率的植物生根深度网格化数据，覆盖2015年至2100年，基于13种植物功能型的观测数据，采用随机森林算法和LPJ-GUESS模型结合环境变量的混合框架进行预测。数据包括历史时期和未来三种共享社会经济路径（SSP119、SSP245、SSP585）下的结果，并提供了不确定性评估，以NetCDF格式开放获取，适用于全球植被和生态系统研究。