Groundwater quality is becoming increasingly polluted and monitoring the content of groundwater ionic compounds is beneficial for dynamic groundwater management and accurate prevention. Little is known about the weak spectral response and inversion mechanisms of ionic compounds， and most existing studies have performed simple qualitative analyses of ionic compounds， with less use of mathematical and statistical methods for comprehensive estimation of their content. Based on the spectral mechanism of ionic compounds and the redundant nature of hyperspectral data， the spectral response mechanism of three ionic compounds in water， the optimal pre-processing method and the algorithm of feature band selection were investigated by measuring the visible-near infrared reflectance spectra （400~1 000 nm） of three ionic compound standard solutions with different concentrations of sodium， potassium and calcium in the laboratory. And based on the characteristic spectral bands， a BP neural network model is constructed to quantitatively invert the ionic compound content. It was found that （1） The overall reflectance of the three ionic compounds is inversely proportional to the content at wavelengths from 400 to 1 000 nm and proportional to the charge number and radius of the ions； （2） Compared with the continuous projection method， the multiple linear regression model constructed based on the characteristic spectral bands extracted by principal component analysis can better infer the content of ionic compounds in water bodies； （3） The preprocessing of the KCl optimal inversion model by SG filtering and the preprocessing of the CaCl₂ and NaCl optimal inversion models by SG filtering followed by reflectance normalization； （4） Compared with the traditional linear inversion model， the PCA-BPNN nonlinear model achieves the best inversion results， among which the inversion results of potassium ion compound content are the best， with the R² and RMSE of the training set reaching 0.996 4 and 248.77， respectively， the R² and RMSE of the test set reaching 0.998 8 and 156.89， respectively. This study can provide important theoretical and technical support for groundwater ionization inversion.