Site-specific correlation of GPS height residuals with soil moisture variability using artificial neural networks

Abstract

Abstract Historical time series generated from GPS sites reveal significant seasonal variations in the vertical direction. It is well known that continental waters (soil moisture, snow, ground water) mass redistribution is one of the potential contributors to these observed seasonal variations although their actual loading effects on GPS results are least well understood. A number of hydrology model outputs exist with a fair degree of uncertainty. Studies of interrelations between anomalous vertical variations observed at geodetic sites and hydrology model outputs are useful, in particular, as the hydrology models continue to be refined. In this paper, artificial neural networks is proposed for correlating GPS height residuals with the soil moisture variability. Time series from eight sites of the global GPS network are used to correlate with the soil moisture information from the US National Oceanographic and Atmospheric Administration (NOAA) Climate Prediction Center’s land data assimilation system (CPC LDAS). The results show the feasibility of the neural interpretation in terms of the correlation coefficients (∼0.6) and root mean square errors (about 15% of residual range). Other geodetic time series can be used for the same purpose, such as from SLR, VLBI, and absolute gravity.