基于机器学习算法模拟不同土壤深度含水率  

Simulation of Soil Moisture at Different Depths Based on Machine Learning Algorithms

作　　者：邓红艳 吴宗俊 崔宁博[2] 徐郁葭 高颖 DENG Hong-yan;WU Zong-jun;CUI Ning-bo;XU Yu-jia;Gao Ying(Xianyang Water Conservancy Team,Xianyang 712000,Shaanxi Province,China;State Key Laboratory of Hydraulics and Mountain River Development and Protection,Sichuan University,Chengdu 610065,China;Renshou Huarui Agricultural Investment Co.LTD,Renshou 620500,Sichuan Province,China)

机构地区：[1]咸阳市水利工作队,陕西咸阳712000 [2]四川大学水力学与山区河流开发保护国家重点实验室,成都610065 [3]仁寿铧锐农业投资有限责任公司,四川仁寿620500

出　　处：《节水灌溉》2023年第3期38-47,60,共11页Water Saving Irrigation

基　　金：“十四五”国家重点研发计划项目(2021YFD1600803-1);国家自然科学基金委优秀青年科学基金项目(51922072);国家自然科学基金项目(52279041,51779161,51009101);四川省科技项目(2022YFN0021,2022YFQ0028,2022YFQ0066,2022YFQ0082)。

摘　　要：准确模拟土壤水分(SM)对农田精准灌溉、优化区域水资源配置和提高农业水资源利用效率具有重要意义。采用逐日气象和土壤含水量数据,探究随机森林(RF)和支持向量机(SVM)模拟不同深度(4 cm、10cm和20 cm)土壤含水量的适用性,采用太阳辐射(Rs)、空气温度(T)、降水量(P)、风速(U2)、空气湿度(RH)和土壤温度(ST)数据的4种输入组合,探讨不同输入变量对SM模拟的影响。结果表明:RF模型在4种输入组合下模拟不同深度土壤含水量精度最高,SVM模型模拟精度较低,R2范围分别为0.871~0.914和0.710~0.814,RMSE范围分别为0.050~0.069 cm^(3)/cm^(3)和0.080~0.098 cm^(3)/cm^(3),MAE范围分别为0.030~0.051 cm^(3)/cm^(3)和0.060~0.077cm^(3)/cm^(3)。考虑模型精度和数据易获取性,RF模型在Rs、T、P、ST和RH输入下精度较高,R2范围为0.884~0.914,RMSE范围为0.050~0.064 cm^(3)/cm^(3),MAE范围为0.030~0.043 cm^(3)/cm^(3)。因此,建议采用RF模型在Rs、T、P、RH和ST输入下模拟不同土壤深度含水量。Accurate simulation of soil moisture (SM) is of great significance for precision irrigation of farmland,optimal scheduling of regional water resources and improving agricultural water utilization efficiency.In this study,daily meteorological and soil moisture data were used to explore the applicability of random forest (RF) and support vector machine (SVM) to simulate SM at different depths (4 cm,10 cm and 20 cm).Four input combinations of solar radiation (Rs),air temperature (T),precipitation (P),wind speed (U2),air humidity (RH) and soil temperature(ST) data were used to explore the impact of different input variables on SM simulation.The results showed that RF model had the highest accuracy in simulating SM at different depths under four input combinations,while SVM model had lower accuracy,with R^(2) ranging of 0.871~0.914 and 0.710~0.814,RMSE ranging of 0.050~0.069 cm^(3)·cm^(3)and 0.080~0.098 cm^(3)·cm^(3),and MAE ranging of 0.030~0.051 cm^(3)·cm^(3)and0.060~0.077 cm^(3)·cm^(3),respectively.Considering model accuracy and data accessibility,RF model had high accuracy under Rs,T,P,ST and RH input,with R~2 ranging of 0.884~0.914,RMSE ranging of 0.050~0.064 cm^(3)·cm^(3),and MAE ranging of 0.030~0.043 cm^(3)·cm^(3).Therefore,RF model is recommended to simulate SM in different soil depths under Rs,T,P,RH and ST input.

关 键 词：土壤水分 随机森林 支持向量机 土壤深度 

分 类 号：S152.7[农业科学—土壤学]