Multifeature GNSS-R Snow Depth Retrieval Using GA-BP Neural Network

Abstract

The Global Navigation Satellite System interferometric reflectometry (GNSS-IR) technique based on signal-to-noise ratio (SNR) data is widely used for snow depth retrieval. Since snow depth retrieval in a snow-free state is very important for meteorological monitoring and since many corrections are post-processed to improve the retrieval accuracy, we propose a GNSS-IR snow depth retrieval model based on a back-propagation neural network optimized by a genetic algorithm to detect the snow state and predict snow depth using the frequency, amplitude and phase of the multipath oscillation term as input features. GPS data collected from the P351 station of the PBO network and measured snow depth from the SNOTEL network were used to conduct the experiments. The accuracy of daily snow state detection for the experimental station exceeded 96%. Combined with the snow state detection results for snow depth regression prediction, the experimental results show that the root mean square error of the snow depth retrieval results for P351 station is 12.09 cm. Compared with the traditional model, the retrieval accuracy is improved by 29.1%, and the correlation coefficient also reaches 0.97, indicating that the proposed snow depth retrieval model not only has high accuracy but also has strong stability. In this study, snow state detection is proposed to improve the retrieval accuracy in snow-free conditions, and the possibility of snow depth retrieval without antenna height is provided.