Spectral Index Fusion for Salinized Soil Salinity Inversion Using Sentinel-2A and UAV Images in a Coastal Area

Abstract

The accurate and rapid inversion of soil salinity in regions based on the fusion of multisource remote sensing is not only practical for the treatment and utilization of saline soil but also the main trend in the development of quantitative soil salinization remote sensing. In this paper, the use of a numerical regression method to fuse spectral indexes based on high-spatial-resolution unmanned aerial vehicle (UAV) images and low-spatial-resolution satellite images was proposed to deeply assess the internal relationships between different types of remote sensing data. An inversion model of soil salt content (SSC) was constructed based on high-spatial-resolution UAV images, and the spectral indexes involved in the fusion were selected from the model. Then, a quadratic polynomial fusion function describing the relationship between the spectral indexes based on the two images was established to correct the spectral indexes based on the low-spatial-resolution satellite image (from Sentinel-2A). Then, scenario 1 (the best model based on Sentinel-2A used for the unfused Sentinel-2A spectral index), scenario 2 (the best inversion model based on UAV used for the unfused Sentinel-2A-based spectral index), and scenario 3 (the best inversion model based on UAV used for the fused Sentinel-2A-based spectral index) were compared and analyzed, and the SSC distribution map was obtained through scenario 3. The results indicate that the scenario 3 had highest accuracy, with the calibration R <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> improving by 0.078-0.111, the root mean square error (RMSE) decreasing by 0.338-1.048, the validation R <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> improving by 0.019-0.079, the RMSE decreasing by 0.517-1.030, and the ratio of performance to deviation (RPD) improving by 0.185-0.423. Therefore, this method can improve the accuracy of SSC remote sensing inversion, which is conducive to the accurate and rapid monitoring of SSC.