地球信息科学学报 ›› 2023, Vol. 25 ›› Issue (8): 1669-1681.doi: 10.12082/dqxxkx.2023.221013

• 地理空间分析综合应用 • 上一篇    下一篇

基于随机森林的西辽河流域CCI土壤湿度降尺度研究

曹煜(), 方秀琴*(), 杨露露, 蒋心远, 廖美玉, 任立良   

  1. 河海大学水文水资源学院,南京 210098
  • 收稿日期:2022-12-29 修回日期:2023-03-21 出版日期:2023-08-25 发布日期:2023-07-14
  • 通讯作者: *方秀琴(1978— ),女,安徽池州人,教授,研究方向为地表参数遥感反演、分布式水文模型及山洪灾害防治。 E-mail: kinkinfang@hhu.edu.cn
  • 作者简介:曹 煜(1998— ),女,山东济南人,硕士生,研究方向为地表参数遥感反演。E-mail: caoyu98@hhu.edu.cn
  • 基金资助:
    国家自然科学基金项目(42071040);国家自然科学基金项目(U2243203);国家重点研发计划项目(2019YFC1510601)

Downscaling of CCI Soil Moisture in the Xiliaohe River Basin based on Random Forest

CAO Yu(), FANG Xiuqin*(), YANG Lulu, JIANG Xinyuan, LIAO Meiyu, REN Liliang   

  1. College of Hydrology and Water Resources, Hohai University, Nanjing 210098, China
  • Received:2022-12-29 Revised:2023-03-21 Online:2023-08-25 Published:2023-07-14
  • Contact: *FANG Xiuqin, E-mail: kinkinfang@hhu.edu.cn
  • Supported by:
    National Natural Science Foundation of China(42071040);National Natural Science Foundation of China(U2243203);National Key Research and Devel opment Program of China(2019YFC1510601)

摘要:

土壤湿度是气候系统中的关键因子,对农业管理、水资源管理和生态系统监测与评估等具有重要应用价值。遥感土壤湿度产品虽能提供大尺度范围的土壤湿度分布,但受限于较低的空间分辨率,难以满足实际应用的要求,对遥感土壤湿度产品的降尺度研究成为当前的热点之一。本文以0.25°分辨率的欧空局ESA CCI日土壤湿度为主要数据源,结合1 km分辨率的 MODIS下垫面数据、地形数据、气象数据等环境因子,构建随机森林降尺度模型,对我国西辽河流域2013—2020年CCI日土壤湿度产品进行降尺度,得到1 km分辨率的土壤湿度时空分布数据。研究发现:① 环境因子重要性分析表明,相对湿度和白天地表温度是影响土壤湿度变化最重要的2个因素,地形与位置因子的影响次之;② 利用研究区内22个站点的实测数据序列对随机森林降尺度模型性能进行验证,结果表明考虑多种环境因子(地表、地形和气象)的降尺度结果比仅考虑地表参数的降尺度结果的精度要高,每个站点的RMSE都在0.048 8 m3/m3以下,平均相关系数为0.497 3,BIAS绝对值在0.003 0~0.033 3 m3/m3,降尺度后的土壤湿度与原始CCI遥感土壤湿度的R2是0.52~0.84;③ 降尺度后的土壤湿度比站点实测土壤湿度时间序列的数值波动小,但二者有着相近的时间变化趋势。本研究构建的降尺度方法在提高遥感土壤数据空间分辨率的同时保留了原数据集的空间分布特征,能够满足实际应用对高分辨率土壤水分数据的需求,并为土壤湿度降尺度研究的环境因子选取提供参考。

关键词: 遥感, 土壤湿度, ESA CCI, 降尺度, 随机森林, 机器学习, 西辽河流域, 环境因子

Abstract:

Soil moisture is a key factor in the climate system and has important application in agricultural management, water resource management, and ecosystem monitoring and assessment. Although remote sensing-derived soil moisture products can provide soil moisture distribution on a large scale, they usually have coarse spatial resolution, making it difficult to meet the requirements of practical applications. Thus, downscaling of remote sensing-derived soil moisture products has become one of the hot topics recently. In this paper, the ESA CCI daily soil moisture at 0.25°resolution is used as the main data source, combined with the MODIS underlying surface data, topographic data, meteorological data, and other environmental factors at 1 km resolution. A random forest downscaling model is constructed to generate downscaled CCI daily soil moisture products at 1 km resolution in the Xiliaohe River Basin of China from 2013 to 2020. Results show that: (1) The analysis of the importance score of environmental factors shows that relative humidity and daytime surface temperature are the two most important factors influencing soil moisture, followed by topography and location factors; (2) The performance of the random forest downscaling model is verified by using the measured data of 22 stations within the study area, and the results show that the downscaled results considering multiple environmental factors (underlying surface, topography, and meteorology) are more accurate than that considering only surface elements. The RMSE of each site is below 0.048 8 m3/m3, the average correlation coefficient is 0.497 3, the absolute value of BIAS is 0.003 0~0.033 3 m3/m3, and the R2 of soil moisture after downscaling is 0.52~0.84 compared with the original CCI remote sensing soil moisture; (3) The downscaled soil moisture has similar temporal trends with the site-level measured soil moisture with less fluctuation in values. The downscaling method proposed in this study improves the spatial resolution of remote sensing-derived soil moisture data while preserving the spatial pattern of the original data set, which can meet the demand for high-resolution soil moisture data in practical applications and provide a reference for the selection of environmental factors in soil moisture downscaling studies.

Key words: remote sensing, soil moisture, ESA CCI, downscaling, random forest, machine learning, Xiliaohe River Basin, environmental factors