多源驱动与季节性集成学习的SMAP表层土壤水分降尺度方法

摘要：

针对土壤水分主动被动卫星(soil moisture active passive,SMAP)土壤水分产品空间分辨率较粗、难以满足区域精细化应用需求的问题，本文以河南省为研究区，提出一种多源驱动与季节性集成学习相结合的1km日尺度表层土壤水分(surface soil moisture.SSM)降尺度方法.以 SMAP L4 SSM为目标变量，融合遥感植被与能量因子、气象降水因子、土壤属性及地形因子,采用“9 km训练-1 km推理”策略,构建基于Top-k优选与Stacking的季节性集成模型 SOBLE-Top5.结果表明，分季节建模整体优于全年统一建模;SOBLE-Top5在春、夏、秋、冬测试集上的R²分别为0.938 9.0.907 8,0.8977和0.9632.均方根误差(RMSE)分别为0.020 8,0.029 1,0.0222和 0.013 3 m³/m³.站点验证表明，降尺度结果能够较好表征不同地类下SSM时序变化.SHAP分析显示.SSM驱动机制具有明显季节差异，整体呈现“冷季土壤属性主导、暖季多因子协同驱动”的特征,研究可为区域农业干旱监测和水资源管理提供高分辨率土壤水分数据支持.

To address the problem that soil moisture active passive(SMAP) soil moisture products have coarse spatial resolution and are insufficient for regional fine-scale applications, this study proposes a 1km daily-scale surface soil moisture(SSM) downscaling method that integrates muli-source driving factors with seasonal ensemble leaming,sampling Henan Province as the study area. SMAP L4 SSM was used as the target variable, and remote sensing vegetation and energy factors.meteorological precipitation factors, soil properties, and topographic factors were integrated. A9 km training-1km inferencestrategy was adopted, and a seasonal ensemble model named SOBLE-Top5 was constructed based on Top-k selection and Stacking. The results show that seasonal modeling generally outperformed annual unified modeling. The R² values of SOBLE-Top5 on the spring. summer, autumn. and winter test sets were 0.938 9,0.907 8, 0.897 7 and 0.963 2. respectively, while the root mean square error(RMSE) values were 0.020 8, 0.029 1. 0.022 2, and 0.013 3 m³/m³, respectively. Site-based validation indi-cated that the downscaled results effectively captured the temporal variations of SSM under different land-cover types. SHAP analysis revealed pronounced seasonal differences in the driving mechanisms of SSM, generally characterized by soil-property dominance in cold seasons and multi-factor synergistic control in warm seasons. This study can provide high-resolution soil moisture data support for regional agricultural drought monitoring and water resource management.

作者：

张春芳，朱胜民，潘登，孟文民，于海洋

Zhang Chunfang, Zhu Shengmin, Pan Deng, Meng Wenmin, Yu Haiyang

机构地区：

河南省焦作水文水资源测报分中心；河南理工大学自然资源部矿山时空信息与生态修复重点实验室；河南省自然资源监测和国土整治院

引用本文：

张春芳，朱胜民，潘登等。多源驱动与季节性集成学习的SMAP表层土壤水分降尺度方法[J].河南师范大学学报(自然科学版),2026,54(5):62-71.(Zhang Chunfang.Zhu Shengmin.Pan Deng,et al.Downscaling SMAP sur-face soil moisture using multi-source drivers and seasonal ensemble learning[J].Journal of Henan Normal University(Natural Science Edition),2026.54(5):62-71.DOI:10.16366/j.cnki.1000-2367.2026.03.11.0001.)

基金：

国家自然科学基金；河南省水利科技攻关项目；河南省自然资源科研项目

关键词：

SMAP;表层土壤水分;降尺度;季节性建模;集成学习

SMAP; surface soil moisture; downscaling; seasonal modeling; ensemble learning

分类号：

P237

多源驱动与季节性集成学习的SMAP表层土壤水分降尺度方法.pdf