Retrieval Of Soil Physico-chemical Properties Based On Hyper-spectral Dataand MODIS Images

Soil salinization is a typical phenomenon of land degradation. It reduces soil permeability, constrains crop growth and production, and restricts agricultural development.Getting information on a large area of soil salinization is essential for soil management and sustainable agricultural governance. Traditional methods of measuring soil physico-chemical properties are slowly in detection with pollution and poor real-time capability, and the sampling number is limited by human and material resources and other factors. Therefore,they cannot be applied to a large area of real-time dynamic monitoring and do not qualify the development of precision agriculture. Soil remote sensing information technology is fast,time-saving with no pollution, and not limited by space-time and topography, so it has been applied widely to the monitoring of soil properties.In this paper, soils in Manasi river basin, Xinjiang Uygur Autonomous Region, were taken as the research subject. Spatial distribution of soil physio-chemical properties was analyzed. Based on the measured spectral reflectance data and its multi-shapes of transformations, models of soil physical and chemical properties with the obtained spectral indices were established using stepwise linear regression(SLR) method, partial least squares regression(PLSR) and support vector machine regression(SVR). Spatial distribution of soil physico-chemical properties in Manasi river basin were also mapped based on MODIS images and prediction models. The main results are as follows:(1) Classical statistical analysis was conducted on soil physico-chemical properties in Manasi river basin including soil water potential repellent time(WDPT), clay(CL), silt(SI),sand(SA) percent, electrical conductivity(EC), organic matter(SOM), potassium([K+]),sodium([Na+]), calcium([Ca2+]), magnesium([Mg2+]) concentration and sodium adsorption ratio(SAR). Soils of 0-10 cm and 10-20 cm layers are not water-repellent because mean WDPT are less than 5s, and weak variation. CL, SI and SA are weak variation, for the coefficient of variation varies from 0.1 to 1. while soil cations, SAR, EC and SOM are all strong variation.correlation analysis between each two kinds of physical and chemical properties were made,WDPT is significantly correlated with CL, SI, and SA. EC, SOM are significantly correlated with part of CL, SI, SA and soil cation concentration. Mostly, SAR is significantly correlatedwith soil cation. Measured data was used to interpolated to predict spatial distribution of soil properties in Manasi river basin. CL, SI, SA and EC in 0-10 cm and 10-20 cm layers have a similar distribution, while WDPT, SOM, [K+], [Na+], [Ca2+], [Mg2+] and SAR not.(2) Models of log(EC), SOM and SAR vs. six spectral indices, i.e. R, SNV, NDVI, CR,LR and FDR, were established based on 221 samples in Manasi river basin using the SLR method, respectively. A comprehensive comparison of model establishment parameters,validation parameters and scatter plots of different EC models indicated that the log(EC)~R model is the best model for predicting soil EC in Manasi river basin, with the Radj2 for the model being 0.94, r of the model being 0.78 and RMSE of the model being 0.25. With similar procedures, the SOM~NDVI and the SAR~FDR models are selected to be the best models to predict SOM and SAR in Manasi river basin, respectively. By comparing the best hyperspectral models established for the three studied soil properties, the models for soil EC had the highest accuracy, the next is the SOM model, while the SAR model is of the most inaccurate.(3) [Na+], [Ca2+], [Mg2+]、CL、SI、SA、SOM、EC models of 78 soil samples in Anjihai irrigation district in Manas river basin established by SLR, PLSR and SVR methods. Though R2 of [Na+], [Ca2+], [Mg2+] models are large, while the RPD are less than 1.4, they cannot be used to predict. SAR~SNV model based on SVR is chose to be the best model to predict soil SAR in Manasi river basin with R2 and RPD being 0.87 and 2.13. CL~R, SI~LR, SA~R model based on SLR are chose to be the best models to predict soil CL, SI and SA in Anjihai irrigate district in Manasi river basin with R2 being 0.62, 0.77 and 0.71, RMSE being 4.90,3.49 and 7.98. SOM~SNV and log(EC)~LR model based on SLR are choose to be the best model to predict SOM and EC in Anjihai irrigate district in Manasi river basin, with R2 being0.67 and 0.57, and RMSE being 0.57 and 0.27(4)EC minimum value based on band reflectance and spectral indices including SI, BI,DVI and NDVI are closed to measured value, while the maximum value are smaller than the measured value, and that is the same for CL, SI, SA and EC. Spatial distribution of the measured soil properties vary widely, while inversion of soil properties has smaller changes.