基于PSO_GA-RBF神经网络模型的元谋干热河谷区土壤水分反演  

Inversion of soil moisture in the Yuanmou hot-dry river valley area based on the PSO_GA-RBF neural network model

作　　者：杜金明 罗明良[1,2] 白雷超 吴秋声 DU Jinming;LUO Mingliang;BAI Leichao;WU Qiusheng(School of Geography,China West Normal University,Nanchong 637009,China;China Liangshan Soil Erosion and Ecological Restoration in Dry Valleys Observation and Research Station,Xide 616753,China)

机构地区：[1]西华师范大学地理科学学院,四川南充637009 [2]大小凉山干旱河谷土壤侵蚀与生态修复野外科学观测研究站,四川喜德616753

出　　处：《测绘通报》2024年第11期1-6,共6页Bulletin of Surveying and Mapping

基　　金：国家自然科学基金(41871324);西华师范大学博士启动项目(22kE001)。

摘　　要：土壤水分对水文和气候过程有重要影响,充分、准确地掌握土壤水分状态对水文模拟、生态治理等具有十分重要的研究价值。本文针对元谋干热河谷区土壤水分反演问题,利用PSO_GA组合优化的RBF神经网络构建了一种新的土壤水分反演模型。试验利用Sentinel-1雷达数据和Sentinel-2光学数据,首先采取适用于研究区低矮植被覆盖类型的水云模型校正植被散射影响;然后将得到的VV和VH极化的土壤直接后向散射系数及交叉极化差代入构建的模型中,实现了对云南省元谋县干热河谷区土壤体积含水量的遥感反演;最后将反演结果与实测的土壤体积含水量数据进行对比验证。结果显示,两者的均方根误差为0.55%m 3/m^(3),决定系数(R^(2))为0.855,对比传统RBF神经网络模型,精度提升明显。将反演结果与NDVI值进行相关分析,结果显示两者的决定系数(R 2)为0.5127。因此,基于Sentinel-1雷达影像数据,利用水云模型和PSO_GA组合优化的RBF神经网络反演的土壤体积含水量具有极高的精度,验证了在干热河谷区大范围土壤水分监测的可行性。Soil moisture has a significant impact on hydrological and climatic processes.A comprehensive and accurate understanding of soil moisture status is of great research value for hydrological simulation,ecological governance,and other related fields.In response to the soil moisture inversion issue in the Yuanmou hot-dry river valley area,a new soil moisture inversion model is constructed using the PSO_GA-combined optimized RBF neural network.The experiment utilizes Sentinel-1 radar data and Sentinel-2 optical data,and employs the water-cloud model suitable for low vegetation cover types in the study area to correct the vegetation scattering effects.The obtained VV and VH polarized soil backscattering coefficients and cross-polarization differences are incorporated into the constructed model,enabling the remote sensing inversion of soil volumetric water content in the hot-dry river valley area of Yuanmou county,Yunnan province.Comparisons and validation against measured soil volumetric water content data show a root mean square error of 0.55%m^(3)/m^(3)and a coefficient of determination(R^(2))of 0.855,demonstrating a significant improvement in accuracy compared to traditional RBF neural network models.Correlational analysis is conducted between the inversion results and NDVI values,revealing a coefficient of determination(R^(2))of 0.5127 between the two.This verifies the high precision of soil volumetric water content inversion based on Sentinel-1 radar image data,utilizing the water-cloud model and PSO_GA-combined optimized RBF neural network,validating the feasibility of large-scale soil moisture monitoring in hot-dry river valley areas.

关 键 词：土壤水分 Sentinel-1/2 径向基函数神经网络 干热河谷 土壤水分反演 

分 类 号：P237[天文地球—摄影测量与遥感]