土壤水分数据融合及其在旱涝灾害多维度评估中的应用  

Assessing drought and waterlogging in multiple dimensions using data fusion of soil water

作　　者：张蕾 郭安红 宋迎波 何亮 赵晓凤 赵运成 ZHANG Lei;GUO Anhong;SONG Yingbo;HE Liang;ZHAO Xiaofeng;ZHAO Yuncheng(National Meteorological Center,Beijing 100081,China)

机构地区：[1]国家气象中心,北京100081

出　　处：《农业工程学报》2024年第13期68-76,共9页Transactions of the Chinese Society of Agricultural Engineering

基　　金：国家重点研发计划专项(2022YFD2300200);中国气象局创新发展专项(CXFZ2023J057)。

摘　　要：获取高精度的土壤相对湿度对开展土壤墒情和旱涝精细化监测评估和预报预警有重要意义。该研究基于2020–2023年4–11月中国气象局陆面数据同化系统(China Meteorological Administration Land Data Assimilation System,CLDAS)逐日土壤相对湿度、全国土壤水分自动站逐小时土壤相对湿度以及土地利用类型、土壤属性、地理信息等数据,采用随机森林和支持向量机模型构建土壤水分自动站观测和CLDAS反演的土壤相对湿度动态融合订正模型,基于融合的土壤相对湿度构建土壤旱涝强度-面积-时间多维度评估指数,开展多维度旱涝监测评估。结果表明:1)采用随机森林模型融合后,0~10、0~20、0~50 cm土壤相对湿度与观测的土壤相对湿度的决定系数分别为0.79、0.81、0.80,相对均方根误差分别为13.81%、11.40%、9.50%,优于支持向量机模型。2)全国土壤缺墒日数百分率呈东南至西北增加趋势,内蒙古中西部、西北地区大部普遍在70%、甚至80%以上,内蒙古东南部、华北中北部、西南地区中西部为50%~70%,中东部大部在40%以下;土壤过湿日数百分率呈东南至西北减小趋势,华南东部和南部、西南地区南部、东北地区东北部多数在50%以上。3)基于融合土壤相对湿度数据构建的土壤缺墒、土壤过湿、墒情指数以及旱涝面积、持续时间指数,明显提升了2022年长江流域高温干旱、2023年台风“杜苏芮”和“卡努”等典型灾害性天气过程动态评估的定量化、精细化水平。土壤湿度融合数据及其旱涝评估指数可有效助力旱涝灾害多维度精细化定量评估,为防灾减灾提供重要支撑。Soil moisture is one of the most important components in the land-air coupling system.Acquisition of soil relative moisture in high precision can benefit to the finer monitoring and assessment,as well as prediction and warning for drought and waterlogging.This study aims to assess drought and waterlogging using data fusion of soil water in multiple dimensions.Daily soil relative moisture was derived from China Meteorological Administration Land surface data assimilation system(CLDAS).Hourly soil relative moisture was observed from automatic soil moisture stations during April and November from the year of 2020 to 2023.The dataset also included the land use,soil properties and geographic information.A dynamic fusion model was constructed to correct the bias of relative soil moisture from between automatic soil moisture stations and CLDAS.Random forest and support vector machine models were used to take the latitude,longitude,altitude,soil sand,soil silt and soil clay as the inputs.Given that daily fusion relative soil moisture was constructed,the intensity,area and duration index were proposed to monitor and assess drought and waterlogging disasters at the multiple dimensions.The results showed that the values of determination coefficient between observed and corrected soil relative moisture by random forest model were 0.79,0.81 and 0.80,respectively,and relative root mean square errors were 13.81%,11.40%and 9.50%,respectively,at the length of 0-10,0-20 and 0-50 cm.Comparatively,the determination coefficient was 0.56-0.57 and relative root mean square error was 17.72%-23.63%between observed and corrected relative soil moisture by support vector machine model.Thus,random forest model was accepted to effectively correct the bias of relative soil moisture between automatic soil moisture stations and CLDAS.Daily fused relative soil moisture was then generated as well.At the spatial scale,the days percent of water deficit increased from southeast to northwest in the period of April-November.Moreover,the days percent of

关 键 词：模型 土壤 相对湿度 旱涝指数 随机森林 支持向量机 

分 类 号：P49[天文地球—大气科学及气象学]