Accuracy analysis of GNSS-IR snow depth inversion algorithms

In recent years, with the continuous development of Global Navigation Satellite System (GNSS), it has been applied not only to navigation and positioning, but also to Earth surface environment monitoring. At present, when performing GNSS-IR (GNSS Interferometric Reflectometry) snow depth inversion, Lomb-Scargle Periodogram (LSP) spectrum analysis is mainly used to calculate the vertical height from the antenna phase center to the reflection surface. However, it has the problem of low identification of power spectrum analysis, which may lead to frequency leakage. Therefore, Fast Fourier Transform (FFT) spectrum analysis and Nonlinear Least Square Fitting (NLSF) are introduced to calculate the vertical height in this paper. The GNSS-IR snow depth inversion experiment is carried out by using the observation data of P351 station in PBO (Plate Boundary Observatory) network of the United States from 2013 to 2016. Three algorithms are used to invert the snow depth and compared with the actual snow depth provided by the station 490 in the SNOTEL network. The observations data of L1 and L2 bands are respectively used to find the optimal combination between different algorithms further to improve the accuracy of GNSS-IR snow depth inversion. For L1 band, different snow depths correspond to different optimal algorithms. When the snow depth is less than 0.8 m, the inversion accuracy of NLSF algorithm is the highest. When the snow depth is greater than 0.8 m, the inversion accuracy of FFT algorithm is higher. Therefore, according to the different snow depth, a combined algorithm of NLSF + FFT is proposed for GNSS-IR snow depth inversion. Compared with the traditional LSP algorithm, the inversion accuracy of the combined algorithm is improved by 10%. For L2 band data, the results show that the accuracy of snow depth inversion of various algorithms do not change with the variations of snow depth. Among the three single algorithms, the inversion accuracy of FFT algorithm is better than that of LSP and NLSF algorithms.     

Snow depth retrieval by using robust estimation algorithm to perform multi-SNR and multi-system fusion in GNSS-IR

Global Navigation Satellite System Interferometric Reflectometry (GNSS-IR) technology provides a new means of snow depth detection. Multi-satellite and multi-Signal-to-Noise Ratio (SNR) provide more data for daily high-precision snow depth retrieval, but also face the problem of data fusion and effective utilization. Therefore, this study proposes a robust estimation algorithm based on multi-satellite and multi-SNR fusion applied to the observations of a GNSS station in Alaska. This study uses four solutions (S_avg, S_med, S_{RE_avg} and S_{RE_med}) to carry out multi-system fusion snow depth inversion and precision comparison research. The S_avg has more obvious disadvantages, which is not suitable for snow depth assessment. The S_{RE_avg} and S_{RE_med} have better snow depth retrieval effects in the snowy periods. The correlation coefficient (R), root mean square error (RMSE) and mean error (ME) of the calculated snow depth using the robust estimation algorithm with respect to the nearest in-situ measurements reached 0.759, 3.7 cm and ?1.4 cm, respectively. Compared with the S_med, the R is increased by 2.0 %, the RMSE corresponds to an improvement of 2.6 %. Moreover, the ME of the snow depth retrievals, as an indicator of the measurement bias, has significantly decreased by 6.7 %. The result also shows that the snow depth inversion by the robust estimation algorithm is more consistent with the in-situ measurements, further extending and advancing the optimal algorithm for snow depth retrieval.     

小波与BP神经网络联合反演GNSS IR高精度水库水位变化

GNSS-IR技术作为一种近地表遥感的新兴手段，在水库水位监测方面凭借其成本低、精度高等优势成为了研究热点。为了提高GNSS-IR技术反演水库水位变化的精度，提出了利用GPS、BDS双系统观测量基于小波分解与BP神经网络联合的方法反演水库水位变化。选取位于南水北调山东境内双王城水库GNSS变形观测站2017年10月1日—12月26日共87天的信噪比(SNR）数据为研究对象，分别利用小波分解和二阶多项式拟合两种方法消除其趋势项并生成SNR残差序列，然后进行Lomb-Scargle谱分析得到水库水位高度变化值，并与实测水位结果相比较。结果表明：各频段SNR利用小波分解和二阶多项式反演水位变化的平均均方根误差分别为0.1062m和 0.2245m，说明小波分解去趋势项的方法更优。最后，在小波分解去趋势项的基础上，利用阈值筛选出融合所用的频段(S1C、S2L、S5Q和S7I），分别采用均值算法、中值算法、随机森林算法和BP神经网络算法对GPS、BDS多频多模信号进行水库水位的融合反演。结果表明，对于水面较为平静的环境，以上算法均能实现厘米级的反演精度，其中BP神经网络算法用于水位反演的效果更优。