Prediction of soil moisture content based on coupled hyperspectral data
1.Key Laboratory for Geographical Process Analysis & Simulation， Hubei Province，
Central China Normal University， Wuhan 430079；2.College of Urban & Environmental Science， Central China Normal University， Wuhan 430079；3.School of Geographic and Oceanographic Sciences， Nanjing University， Nanjing 210023
1.Key Laboratory for Geographical Process Analysis & Simulation， Hubei Province，
Central China Normal University， Wuhan 430079；2.College of Urban & Environmental Science， Central China Normal University， Wuhan 430079；3.School of Geographic and Oceanographic Sciences， Nanjing University， Nanjing 210023
Abstract：It becomes a hot spot in precision agricultural research to monitor soil moisture content (SMC) situation by using hyperspectral data in the field. However， the soil undisturbed spectral reflectance is affected by the soil surface properties such as surface roughness， texture and microcapsules as well as other environmental factors， and SMC spatial variation is not significant in smallsale region， making it more difficult to extract spectral information about SMC， which leading to the lower SMC estimation accuracy. In addition， the model based on hyperspectral data of soil samples by the laboratory preparation has higher accuracy， but the experimental samples have anthropogenic changes in soil structure and tightness which are not able to show the actual status of SMC in the field. Therefore， the present work attempts to propose a new method by coupling undisturbed spectral data to standard spectral data to estimate SMC of farmland. Undisturbed spectral reflectance (Rund) and drying spectral reflectance (Rdry) of fluvoaquic soil samples were collected； the standard spectral reflectance (StdR) in the dry state(SMC is 0%) was determined based on Rdry； Rund and StdR were coupled by the algorithm of subtraction， division and normalization. Then the coupled spectral reflectance(CplRS， CplRD and CplRN) is generated； the moisture sensitive bands spectral reflectance(MoeRS， MoeRD， MoeRN) was extracted from coupled spectral reflectance； SMC estimation model is established using partial least squares regression (PLSR) method. The results showed that standard spectral reflectance had excellent representation， which could provide a uniform and stable background value for coupling spectral methods； the coupled spectral reflectance reduced the influence of other factors except soil moisture on soil hyperspectral observations； the model based on coupled spectral reflectance of moisture sensitive bands had a noticeable promotion compared to the model based on Rund (Rc increased from 0.46 to 0.61， Rp advanced from 0.49 to the highest 0.71， and RPD improvedfrom 1.39 to 1.72)， whichwasmorestablebutlesscomplex. ThesimplemethodwillbeeasytoapplytotheeffectiveandaccurateassessmentofSMCinfield.
