GNSS-IR model of snow depth estimation combining wavelet transform with sliding window

Currently, GNSS interferometric reflectometry technology has become a high-precision method for monitoring land surface snow depth. Aiming at the problems of signal separation and random estimation biases, we developed a GNSS-IR refined model with multi-satellite fusion for snow depth estimation combining wavelet transform with sliding window. The common polynomial method was replaced by discrete wavelet transform to obtain the high-quality SNR sequences of the reflected signals which can calculate the reflected height of GPS antenna. Then, these reflected heights from SNR observations of multi-satellite were effectively selected and averaged using the sliding window under a constrained threshold. The refined model was established using GNSS observations for snow season from 2016 to 2017, and then the snow depth datasets of both PBO H2O and SNOTEL were regarded as reference to verify the performance of the refined model. The results show that there is a high agreement between snow depths derived from the refined model and in situ measurements, and the RMSE is 10 cm. Compared with the results of a single satellite, the accuracy and the stability of the refined model with multi-satellite fusion are obviously better. In terms of RMSE, the accuracy of the refined model has been improved by 50% when compared with PBO H2O dataset. In addition, taking into consideration that land surface roughness is an error factor, a relative RMSE value of snow depth estimations corrected by a new datum of the reflection height is approximately 4 cm, and the correlation coefficient between snow depth estimations and in situ measurements reaches 0.98. © 2020, Surveying and Mapping Press. All right reserved.