Phosphate content prediction models of Spartina alterniflora based on hyperspectral data

Phosphorus is an important element for pasture growth. Taking the Dafeng Elk Nature Reserve as the research object, the prediction models of phosphate content are established to find the best modeling method and provide a reference basis for physiological and ecological monitoring of Spartina alterniflora. Based on correlation coefficient, principal component analysis, and variable importance for projection, the sensitive bands of reflectance, first derivative, and continuum removal were screened, and the multiple linear regression, partial least squares regression, and BP artificial neural network prediction models were established. The correlation and significance between spectral data and phosphorus content of Spartina alterniflora were enhanced after the first derivative and continuum removal transformation. The accuracy was sorted as first derivative > reflectance > continuum removal by comparing the inversion models established by different spectral data forms,. The accuracy and stability of the three models were comprehensively compared, and the accuracy of the models was ranked as BP artificial neural network > partial least squares regression > multiple linear regression. The best inversion model was the first derivative-BP model based on full bands spectral data, with a modeling accuracy of R² = 0.920, RMSE = 0.059, and RPD = 2.949.