Hyperspectral Estimation Of Soil Salt And Moisture Content Based On Optimized Spectral Indices

Surface soil water and salt is an important boundary condition parameter of soil water and salt transport model.Accurate surface soil water and salt information can improve the simulation and prediction accuracy of water and salt transport model.Accurate soil water and salt information monitoring can help to further improve the accuracy of water and salt transport model simulation,and is of great significance for regional salinization prevention and control,agricultural sustainable development and ecological civilization construction.In order to find the best hyperspectral parameters for predicting soil water and salt,and to achieve efficient monitoring of soil water and salt information,this study used ASD FieldSpec3 hyperspectral instrument to perform indoor hyperspectral measurement of soil samples taken in the field,and the spectral index was performed by two-band optimization algorithm.Band optimization,screening the most sensitive hyperspectral parameters under different hyperspectral data(original hyperspectral reflectance and its corresponding three mathematical transformations)to establish a soil water and salt hyperspectral estimation model.The conclusions are as follows:(?)The soil salinity content and soil water content dispersion are high in the study area,and the soil acidity and alkalinity dispersion is weak.The soil ions are mainly sodium ions,chloride ions and sulfate ions,and the soil salt components are mainly sodium salts.The correlation coefficient between sodium ion and chloride ion is the largest.(?)The correlation between the optimized spectral index and soil water and salt is significant.Compared with the traditional one-dimensional correlation method,the optimized spectral index has obvious advantages.It can basically improve the correlation by 0.2 and play a great role in hyperspectral modeling.Whether in one-dimensional or two-dimensional level,appropriate hyperspectral data preprocessing methods are helpful to improve the correlation to a certain extent.(?)Variable Importance in Projection,which is applied in the case of strong correlation of independent variables,also plays a very significant role in this paper,and can improve the efficiency of the model.From the performance of the estimation model,it can be seen that the support vector machine model based on machine learning is superior to the partial least squares model,and the combination model based on multiple optimization spectral indices is superior to the model based on single optimization spectral index.When estimating soil salinity,the SVM model can reach the highest predictive determinant coefficient of 0.762.However,in the estimation model of soil moisture content,the PLSR model is the highest,the predictive determinant coefficient is 0.637,and the single optimized spectral index model is better than the combination model.The optimized spectral parameters obtained in this paper can provide a basis for quickly and accurately seeking the best wavelength of satellite sensors for monitoring soil water and salt content in arid and semi-arid regions.In addition,the optimization of the band can also provide a theoretical basis for designing active sensors in specific bands.Further reduce the workload of high-spectrum mass data processing,in order to achieve efficient monitoring of soil water and salt information.Finally,it can provide corresponding measures and countermeasures for the monitoring and prediction of soil salinization and ecological regulation,which is of great significance to the construction of regional ecological civilization.