Geochemical inversion study of potassium and phosphorus in soil based on neural network and ZY1-02D hyperspectral data.

In response to the agricultural demand for improving the quality and efficiency of the unique agricultural product "Zhefang Gongmi" in Yingjiang County, Yunnan Province, this study aims to uncover the relationship between soil potassium (K) and phosphorus (P) content and hyperspectral data, and to develop a precise inversion model based on hyperspectral remote sensing. The study innovatively uses AHSI hyperspectral data (166 bands, 400-2500 nm) from the ZY1-02D satellite, combined withY1-02D satellite, combined with geochemical data from 856 soil sampling points. Through Savitzky-Golay filtering, Minimum Noise Fraction (MNF) transformation, continuum removal, and third-order differential transformation to enhance spectral features, inversion models for K/P elements using Extreme Learning Machine (ELM) are constructed separately for vegetation-covered and bare soil areas. The key findings of the study are as follows: (1) The correlation of potassium content was significantly higher in the vegetated area compared to the bare area, reaching up to 0.55. After continuum removal, significant correlations were observed in the vegetated area at 979 nm, 1031 nm, 1929 nm, and 2334 nm, all with correlation coefficients above 0.50. In contrast, the bare area showed significant correlations in the third-order differential spectrum at 1014 nm, 1677 nm, 1880 nm, and 2216 nm, with a maximum correlation of 0.47. Phosphorus showed a higher correlation in the bare area than in the vegetated area. (2) The optimal prediction models for potassium and phosphorus in both the vegetated and bare areas were based on the ELM model. In the vegetated area, the coefficient of determination for potassium was 0.654, with a mean square error of 22.686 g/kg; in the bare area, the model for potassium yielded a coefficient of determination of 0.617 and a mean square error of 9.102 g/kg. (3) A novel method has been proposed for analyzing the geochemical element content of soil, designed to accurately assess potassium geochemical information and provide a basis for predicting phosphorus content. The "Vegetation - Bare Land" zonal inversion paradigm proposed in this study achieves high-precision inversion of soil potassium (K) content in the highland agricultural areasal inversion paradigm proposed in this study achieves high-precision inversion of soil K content in the highland agricultural areas, providing an expandable technological pathway for improving the quality of Yingjiang rice and enhancing soil fertility. This approach offers a theoretical foundation for precision agricultural fertilization management.