With the continuous development of GNSS (Global Navigation Satellite System) technology, GNSS-IR (GNSS Interferometric Reflectometry) has become a research hotspot in the field of snow surface monitoring, and the accuracy and reliability have been initially verified. We focus on the reasons for the low accuracy of the existing GNSS-IR snow depth inversion. Therefore, we use the P351 station data in the PBO (Plate Boundary Observatory) network in the United States to monitor the snow depth during the four years from 2011 to 2015. The actual measured snow depth at station 490 in the SONTEL network is used as the true value for accuracy verification. We studied the relationship between the inversion error caused by the slope and the slope angle and the satellite elevation angle, and proposed a slope correction method. The results show that the RMSE (Root Mean Square Error) of snow depth inversion after slope correction is reduced from 12.1 cm to 10.7 cm, the accuracy is improved by 11.6 %. In addition, it is found that there is an apparent correlation between the retrieved snow depth and the inversion error. With the increase of snow depth, the error gradually changes from positive to negative, and the absolute value of error still increases with the increase of snow depth after the error changes to negative. To this end, we introduce BPNN (Back Propagation Neural Network) to train the inversion snow depth and inversion error of the three snowfall periods from 2011 to 2014, then predicts and corrects the snow depth inversion error during the snowfall period from 2014 to 2015. The results show that the RMSE of the corrected GNSS-IR snow depth inversion is reduced from 10.7 cm to 5.7 cm, and the accuracy is increased by 46.7 %. The overall accuracy of the GNSS-IR snow depth inversion is improved by 52.9 % after the slope correction and the BPNN error prediction is performed, which further verifies the accuracy and effective of the approach that proposed by us.