Since 1999, the Loess Plateau, China, has undergone one of the world's largest revegetation programs (Grain for Green Project, GfGP). Revegetation has profound impacts on hydrological cycle and water balance, especially in arid and semi‐arid areas. As an essential component of water balance, the long‐term change in water storage (ΔS) is generally treated as zero. However, it remains unclear how to define the time scale of “long‐term,” especially over regions undergoing dramatic vegetation restoration. In this study, we quantify ΔS over the Loess Plateau from 1982 to 2009 from a water balance perspective, with a particular focus on the impact of vegetation on ΔS at different time scales. Results show that a 44.78% increase in vegetation coverage could lead to a 37.72% decrease in ΔS. Moreover, the better the vegetation growth, the lower the ratio of ΔS to precipitation (ΔS/P) in the Loess Plateau. ΔS/P achieved a local minimum at the 4‐ to 5‐year scale and reached a plateau after 10 years. We further use the Budyko framework to verify the above conclusions. We find that errors of simulated streamflow over 11/16 of catchments become stable after 4–5 years, and all 16 catchments become stable after 10 years. Ignoring ΔS at a 10‐year scale has the least influence on water budget closure. These findings highlight the importance of considering time scales when ignoring ΔS in the water balance analysis, especially when the length of measurement is limited.

Profile soil moisture (SM) in mountainous areas is important for water resource management and ecohydrological studies of downstream arid watersheds. Satellite products are useful for providing spatially distributed SM information but only have limited penetration depth (e.g., top 5 cm). In contrast, in situ observations can provide measurements at several depths, but only with limited spatial coverage. Spatially continuous estimates of subsurface SM can be obtained from surface observations using multiple methods. This study evaluates methods to calculate subsurface SM from surface SM and its application to satellite SM products, based on a SM observation network in the Qilian Mountains (China) that has operated since 2013. Three different methods were tested to estimate subsurface SM at 10 to 20, 20 to 30, 30 to 50, and 50 to 70 cm, and, in a profile of 0 to 70 cm, from in situ surface SM (0 to 10 cm): the exponential filter (ExpF), the artificial neural network (ANN), and the cumulative distribution function (CDF) matching methods. The ANN method had the lowest estimation errors (RSR), while the ExpF method best captured the temporal variation of subsurface soil moisture; the CDF method is not recommended for the estimation. Meanwhile the ExpF method was able to provide accurate estimates of subsurface soil moisture at 10 to 20 cm and for the profile of 0 to 70 cm using surface (0 to 10 cm) soil moisture only. Furthermore, it was shown that the estimation of profile SM was not significantly worse when an area-generalized optimum characteristic time (T_opt) was used instead of station-specific T_opt for the Qilian Mountains. The ExpF method was applied to obtain profile SM from the SMAP_L3 surface soil moisture product, and the resulting profile SM was compared with in situ observations. The ExpF method was able to estimate profile SM from SMAP_L3 surface data with reasonable accuracy (median R of 0.65). Also, the combination of the ExpF method and SMAP_L3 surface product can significantly improve the estimation of profile SM in mountainous areas compared to the SMAP_L4 root zone product. The ExpF method is useful and has potential for estimating profile SM from SMAP surface products in the Qilian Mountains.