Selection Of Soil Moisture Sensitivity Bands And Estimation Of Soil Moisture In Plough Layer

Soil is the basis of land,and mastering the physical and chemical properties of soil is an important prerequisite for the rational use of land resources,the effective protection of cultivated land,the effective implementation of precision agricultural management,and the guarantee of agricultural production.Soil moisture is the basic premise and condition for sustainable survival of crops,and it is one of the key indicators for determining soil fertility and evaluating soil quality.At present,with the wide application of remote sensing technology,especially hyperspectral remote sensing,its advantages of high spectral resolution and multiple bands provide an accurate,rapid and non-destructive method and method for obtaining soil moisture content indicators.However,soil hyperspectral reflect the comprehensive information of soil physical and chemical properties.How to effectively extract spectral characteristics and sensitive information of soil moisture in spectra containing complex properties has become a problem that needs to be solved urgently.This has profound theoretical significance for reducing the complexity of soil moisture estimation model and improving the robustness of the estimation model,and it has important application value for realizing rapid monitoring of soil moisture at the regional or land level.Therefore,in this paper,the fluvo-aquic soil in Zhugentan Town,Qianjiang City,Hubei Province is taken as the research object,and the original spectral data is subjected to continuous removal(CR)transformation,centered on the laboratory and field measured soil spectral information acquired by the spectrograph.Analyze the spectral characteristics and response laws of soils with different water contents;use the variable projection importance(VIP)method,information-free variable elimination(UVE)coupled competitive adaptive adaptive weighted sampling(CARS)method,iterative and retained information variables(IRIV)The method of the method is to optimize the soil spectrum for variable,preferably for the soil moisture sensitive band;to perform the discrete wavelet transform(DWT)denoising and CR spectrum conversion for the field measured spectrum,based on the soil moisture sensitive band obtained in the laboratory,utilizing the linear The partial least-squares regression(PLSR)method and nonlinear support vector machine(SVM)method were used to establish the soil water content estimation model,and the full-band model was introduced for comparison to determine the best-performing field soil moisture estimation model.The findings are as follows:(1)By designing the soil air-drying process test in the laboratory,we obtained 9 sets of gradient data of different water content,and observed the spectrum change characteristics.It was found that the overall change of the original spectral reflection curve was relatively stable,and with the increase of soil moisture,the spectral reflectance was presented.With decreasing trend,the absolute change of reflectivity in the near-infrared band is greater than the visible light band.(2)Based on the continuum removal method,the spectral absorption characteristics of the original spectra acquired in the laboratory are highlighted.There is a strong spectral absorption valley around 480 nm,and there are weak spectral absorption valleys near 420 nm,640 nm,720 nm,and 920.(3)Using three variable selection methods VIP,UVE-CARS,and IRIV to optimize the variable gradient data obtained from the laboratory,the best method for obtaining the best variable is UVE-CARS.The 415～428nm,470～485nm,639～645nm,718～729nm,924～941 are soil moisture sensitive bands.(4)Based on field measured spectral data,PLSR and SVM models for estimating field soil moisture content were established based on the obtained soil moisture sensitive bands,and the full band model was introduced for comparison.The SVM model based on the soil moisture sensitivity band has the best comprehensive performance.Although the model accuracy is slightly lower than the full-band model,the model has low complexity and high robustness;the R2,RMSE,MAE,and ME of the model modeling set are respectively 0.86,4.12%,3.49%,and-1.03.The R2,RMSE,MAE,ME,and RPD of the validation set are 0.85,4.70%,4.03%,-1.13,and 2.23,respectively.The model can achieve rapid monitoring of soil moisture status in the field.In this study,laboratory tests were conducted to analyze the response laws of soil moisture content changes,and to provide accurate and effective test conditions for obtaining sensitive bands of soil moisture;and to apply the obtained soil moisture sensitive bands to soil moisture in field measurement environments.Estimation,the higher accuracy of the estimation results verify the necessity and effectiveness of the indoor analysis,improve the method of field estimation of soil moisture,and provide theoretical support for the implementation of regional-level field soil moisture monitoring.