Prediction of soil organic matter content in cultivated land of Qilu Lake basin based on ZY1E remote sensing images

【Objective】 To study the relationship between the content of soil organic matter（SOM） in cultivated land within the Qilu Lake basin and the spectra of satellite images, clarify the spectral characteristics of SOM, establish an inversion model for the content of SOM, which could provide reference basis for precise fertilization of cultivated land soil in Qilu Lake basin. 【Method】 Based on the ZY1E hyperspectral remote sensing data of the Qilu Lake basin, the original reflectance（RF）, first order differentiation（FDR）, ratio soil index（RSI） and normalized soil index（NDSI） were selected as independent variables, and the correlation analysis with SOM content was carried out to screen out the sensitive bands, and the univariate linear regression model and multiple stepwise regression model were constructed respectively,and the SOM content of cultivated land in the basin was inverted.【Result】The spectral reflectance characteristics of soil with SOM contents in Qilu Lake basin had similar trend, and the spectral reflectance of soil had absorption valley at 900,1200, 1500 and 2000 nm. The correlation between SOM content and spectral reflectance could be improved by different mathematical transformations of the spectrum, and NDSI and RSI had the best improvement effects. By comparing the accuracy of each model, the stepwise regression model constructed with the reflectance RSI had the best fitting effect, its coefficient of determination（R²） was 0.80, and the root mean square error（RMSE） was 6.96, the R² of the measured and predicted values of the SOM in validation model was 0.82, and the RMSE was 3.65. Using GIS spatial analysis function,the SOM content of cultivated land in Qilu Lake basin was retrieved according to the optimal model, and most of the SOM content was concentrated in 35-45 g/kg, which was at a relatively high level. Moreover, the SOM content of cultivated land soil in Qilu Lake basin showed a low-high-low distribution from the southwest to the northeast.【Suggestion】Optimizing the extraction of characteristic wavelengths for the variation of SOM content in cultivated land soil can improve the accuracy of the prediction model; using the hyperspectral remote sensing data of ZY1E to monitor the nutrients in cultivated land soil can solve the problem of being affected by redundant information in traditional geostatistics, and provide data support for the precise fertilization management of soil, thereby enhancing soil fertility and crop yields.