Hyperspectral estimation of soil conductivity of different land use types in lakeside oasis of Bosten Lake in Xinjiang
To more accurately understand the soil salinity of different land use types in the lakeside oasis of Bosten Lake, Xinjiang, three methods of competitive adaptive reweighted sampling(CARS), successive projection algorithm(SPA), and competitive adaptive reweighted-successive projection algorithm(CARS-SPA) were applied to screen the characteristic bands of soil conductivity hyperspectral data of different land use types, based on the full band and characteristic bands. The estimation models of soil conductivity of the lakeshore oasis were constructed based on the full band and characteristic bands combined with BP neural network to compare the accuracy of estimation models in different ways. The results showed that: (1) the mean values of soil conductivity of cropland, forest land, wasteland, and overall land are 0.84, 5.43, 5.78, and 3.26 mS/cm, respectively; the overall soil conductivity of the lakeshore oasis is 2.42 mS/cm higher than the mean value of cropland, 2.17 and 2.52 mS/cm lower than forest land and wasteland. (2) The CARS-SPA method can reduce the number of bands input to the model and improve the efficiency of the model. wasteland significantly improves the estimation accuracy of soil conductivity in the study area. Among the three models of FDR-CARSBP, FDR-SPA-BP, and FDR-CARS-SPA-BP, the average R2 of soil conductivity modeling for cropland, forest land, and wasteland increased by 0.12, 0.14, and 0.15, respectively, compared with the overall soil modeling, the FDR-CARS-SPA-BP model is the optimal model for hyperspectral estimation of soil conductivity in the study area.
