The study took the Yellow River source area as the research area，the degraded alpine meadow was selected asthe research samplesite，the soil vegetation and nutrient characteristics of degraded alpine meadow were analyzed throughfield survey sampling，combined with indoor soil nutrient and soil hyperspectral data measurement，the mathematical transformation of soil hyperspectral data was performed and the correlation coefficient with soil nutrient content was calculated.The inverse models of soil organic carbon，total nitrogen and total phosphorus contents were established by partial leastsquares regression(PLSR)and back propagation neuralnetwork(BPNN)methods，respectively. The results showed that the soil organic carbon and soil total phosphorus content prediction models established by PLSR were better than the BPNNmodels，it had better predictioneffects，in which the modeling set R2=0.9585 and RMSE=0.1079 for the PLSR prediction model of soil organic carbon content and the validation set R2=0.9493 and RMSE=0.1210 had higher model accuracy and could be accurately estimated. The modeling set R2=0.7497，RMSE=0.2391，validation set R2=0.5977，RMSE=0.2445 ofthe PLSR prediction model for total phosphorus content met the basic estimationrequirements；The prediction model for total soil nitrogen content established by BPNN was better than the PLSR model with modeling set R2=0.8772，RMSE=0.7663， validation set R2=0.6887，RMSE=0.8556，the model accuracy was better and met the basic estimation requirements.