Abstract:Soil moisture content (SMC) is one of the most critical soil components for successful plant growth and land management, particularly in arid and semiarid areas. In existing researches, it was determined by a conventional method based on oven drying of samples collected from fields. The first derivative (FD), absorbance (Abs) and continuumremoval (CR) algorithm were brought into the preprocessing of hyperspectral data based on the initial Savitzky-Golay (SG) smoothing. With SMC data and unmanned aerial vehicle (UAV) platform derived imaging hyperspectral imagery collected from the cropland in Fukang Oasis, Xinjiang Uyghur Autonomous Region, China. Then, the raw hyperspectral reflectance data were transformed into five preprocessing, i.e., SG, SG-FD, CR, Abs and Abs-FD. In addition, the relationships between SMC and pretreated difference index (DI), ratio index (RI), normalization index (NDI) and perpendicular vegetation index (PVI) were discussed. The correlation coefficients between each spectral index and SMC were also computed. Based on the optimal spectral index and pretreatment scheme, the hyperspectral quantitative estimating model was constructed for the dictation of SMC in oasis cropland in arid area. The result showed that the correlation between pretreated spectral index and SMC was improved to some extent, and the PVI (R644, R651) based on Abs preprocessing was the best with correlation coefficient of 0788. The cubic fitting function was optimal. On the basis of noise elimination, the multivariable SMC estimation model based on different preprocessing schemes could detect much finer spectral information from reflectance data, reduce the error caused by the single spectral index, and further improve the quantitative estimation effect of the model. The prediction accuracy of the Abs model was the most prominent, with R2c of 0.84， RMSE of 2.16%， R2p of 0.91 and RMSE of 1.71%. The effect of the SMC estimation model constructed was based on the preprocessing and noise elimination. The constructed SMC estimation model could reduce the error of independent single variable; and further resolve the problem of over fitting. The model could be used for hyperspectral mapping and performance estimating. The research result could provide a novel perspective and scheme for the remote sensed detection of soil water condition, especially in the arid and semiarid areas.
