A novel retrieval model for soil salinity from CYGNSS: Algorithm and test in the Yellow River Delta

Abstract

Soil salinization, which occurs mainly in arid and coastal regions, hampers agricultural development, and threatens food security in these regions. Therefore, acquiring the spatial distribution of soil salinization with high accuracy is paramount. However, obstacles continue to hinder attempts to use satellite remote sensing. Cyclone Global Navigation Satellite System (CYGNSS) provides a new alternative opportunity for soil salinization retrieval. In this study, soil electrical conductivity (EC) was chosen as the proxy of soil salinity. A CYGNSS-based model for soil EC retrieval (CIG) was proposed, and has been tested in the Yellow River Delta, China. The CIG model consisted of two modules. First, we modified the geometrical optics model to derive the surface’s Fresnel reflectivity from the incoherent scattering signal of the land surface with vegetation attenuation, and horizontal and vertical surface roughness calibrations. Second, soil EC was retrieved based on the gradient boost regression tree (GBRT) algorithm with the inputs of the surface’s Fresnel reflectivity and other ancillary variables. The in-situ measurements were used to compare the CIG retrieved soil EC against the retrievals from other classical machine learning approaches and the coherent model using CYGNSS data, and the optical SVIs model. CYGNSS signal was found to be highly sensitive to in-situ soil EC, and the proposed CIG model outperformed other models, with R = 0.730, RMSE = 1.318 mS/cm, and MAE = 0.570 mS/cm. In addition, several issues related to the model were also discussed, including the interactions between the incidence angle of the CYGNSS signal and soil EC, and the advantages and limitations of CYGNSS. The result suggested that the proposed CIG model, along with new CYGNSS data, can provide a promising method for monitoring land salinization on a large scale.