Assessing seasonal and interannual water storage variations in Taiwan using geodetic and hydrological data

We systematically investigate the spatiotemporal water storage changes in Taiwan using geodetic (GNSS and GRACE) and hydrological (precipitation, GLDAS and LSDM assimilation models, and insitu groundwater level) datasets. We use GNSS-observed vertical deformation to estimate water storage changes based on elastic loading theory and weighted least-squares inversion, correcting for contributions from global loads using GRACE. The mean annual water-thickness change inferred from GNSS across Taiwan is 0.53 ± 0.17 m and the largest seasonal change of up to 0.91 m is estimated in southwest Taiwan. Comparison of the geodetic and hydrological data shows that the spatial pattern of annual water storage change estimated from GNSS, GLDAS, and precipitation data are generally consistent, indicating significant seasonal water-load fluctuations in Taiwan. However, the GRACE solution significantly underestimates the amplitude of water mass change in Taiwan due to leakage effect, but temporally correlates well with GNSS estimates. Hydrological assimilation model GLDAS, dominated by shallow soil moisture variations, predicts that the average seasonal variation of water thickness is only about 17% of GNSS estimates. This value is about half of the mean annual LSDM water storage change of 0.18 m including an estimate of both soil moisture and surface water. The discrepancy suggests that the contribution of groundwater is substantial and the total water storage change in the hydrological assimilation model is underestimated in Taiwan. The spatiotemporal distributions derived using independent component analysis (ICA) are generally consistent between the geodetic and hydrological data. However, comparisons of seasonal amplitudes and phases between all data pairs reveal different response times to precipitation, reflecting the complex nature of transient water storage due to variable rainfall patterns, infiltration rate, soil saturation, and runoff. The peak rainfall occurs in June-July, which is one-to-two months before the peak GNSS subsidence. Water storage of the GLDAS model also reaches its maximum in August, suggesting the water storage is controlled by the infiltration rate and capacity and the total water recharge from rainfall is generally larger than discharge in the summer. The highest groundwater levels lag one and two months behind the peak GNSS subsidence in western and eastern Taiwan, respectively, indicating a higher infiltration rate in western Taiwan.