<b>SMRFR: </b><b>A global long-term multilayer soil moisture dataset (2000-2023)</b><b> generated using machine learning</b>

Accurate and continuous monitoring of soil moisture (SM) is crucial for a wide range of applications in agriculture, hydrology, and climate modelling. In this study, we present a novel machine learning (ML) based framework for generating a continuously updated, multilayer global SM dataset, and we introduced SMRFR (e.g., Soil Moisture from Random Forest Regression). Utilizing publicly available reanalysis and remote sensing data for generation, SMRFR provides daily SM estimates at five soil layers (0-5, 5-10, 10-30, 30-50 and 50-100 cm) with a fine spatial resolution of 9-km over the period 2000 to 2023. The evaluation results demonstrate SMRFR's potential to capture the spatial and temporal variability of SM. SMRFR exhibits robust performance in transferring knowledge across continents in capturing transient and seasonal SM dynamics after rainfall events, with medium ubRMSE of 0.0339 m³/m³. Our novel SM dataset offers basis and scientific reference for agricultural, hydrological, and ecological studies. Due to the whole dataset for the global scale is too big (391.59GB) to deposit at once，we uploaded the data of 2000 to figshare. The whole dataset can be accessed at https://doi.org/10.11888/Terre.tpdc.301526.精准且连续的土壤湿度（Soil Moisture，SM）监测对于农业、水文学及气候建模领域的诸多应用至关重要。本研究提出了一种新颖的基于机器学习（Machine Learning，ML）的框架，用于生成可持续更新的多层全球SM数据集，并将其命名为SMRFR（Soil Moisture from Random Forest Regression，即随机森林回归反演土壤湿度）。该数据集依托公开可用的再分析与遥感数据生成，可提供2000年至2023年间、空间分辨率达9公里的每日SM估算值，共涵盖5层土壤（0-5 cm、5-10 cm、10-30 cm、30-50 cm及50-100 cm）。评估结果表明，SMRFR能够有效捕捉SM的时空变异性。SMRFR在跨大洲迁移知识以捕捉降雨事件后的瞬态与季节性SM动态方面展现出稳健的性能，其无偏根均方误差（unbiased root mean square error，ubRMSE）的中位值为0.0339 m³/m³。本研究构建的新型SM数据集可为农业、水文学及生态学研究提供基础数据与科学参考。由于全球完整数据集体量过大（391.59 GB），无法一次性上传存储，我们已将2000年的数据上传至figshare平台。完整数据集可通过以下链接获取：https://doi.org/10.11888/Terre.tpdc.301526。