Abstract: The rapid monitoring of soil organic matter (SOM) content in large-scale salinized wheat fields can provide data for promoting research in saline soils and carbon cycle. Based on field sampling and remote sensing images of unmanned aerial vehicle, we established remote sensing prediction models of regional SOM using three methods, i.e., multiple linear regression (MLR), partial least squares (PLSR), and support vector machine regression (SVR) for bare land and wheat field, respectively. The models were validated and compared to identify the optimal inversion model of SOM. Moreover, the SOM in the area was inverted using the optimal model, with the inversion results being compared with the data by interpolation. The results showed that the spectrum after the filtering of 5×5 median was best related to surface SOM. Among the three models, the SVR model had the highest prediction accuracy, followed by the PLSR, while the MLR lowest. The SVR model was the best one for estimating wheat field, with coefficient of determination (R²) and root mean square error (RMSE) of 0.89 and 0.20, respectively, and the validated R² and RMSE were 0.82 and 0.24, respectively. The bare land SOM was also best fitted by the SVR model, with R² and RMSE were 0.63, 0.26, respectively, and the verified R² and RMSE were 0.61, 0.25, respectively, but without statistical significance. The inversion of the optimal model revealed that SOM content in this region ranged from 17.51 to 22.53 g·kg^-1, with an average of 19.51 g·kg^-1, which was generally consistent with the field measurement. Compared with the inversion results, the interpolation data were limited in accuracy. Overall, our study suggested that the unmanned aerial vehicle-based multi-spectral analysis could be applied to quick and accurate estimation of SOM content in saline soil at the jointing stage of winter wheat.

Key words: unmanned aerial vehicle, multispectral, saline soil, winter wheat, jointing stage, soil organic matter