A soil-thickness-constrained geospatial framework for reconstructing monthly shallow soil water storage across the Tibetan Plateau

Fixed-depth (FD) assumptions can substantially bias estimates of soil water storage (SWS) integrated over the 0–200 cm profile, hereafter referred to as shallow SWS, particularly in mountain regions where soil thickness (ST) varies markedly and shallow bedrock is widespread. In this study, we developed an ST-constrained framework to reconstruct monthly shallow SWS across the Tibetan Plateau (TP) at 1 km resolution over an 11-year window comprising 132 months (January 2009–December 2019) by integrating remote-sensing products, reanalysis data, and field observations. Using 1,615 in situ ST profiles and multi-source environmental covariates, we first mapped ST with a random forest model and then downscaled GLDAS soil moisture (SM) for four standard layers: 0–10, 10–40, 40–100, and 100–200 cm. Pixel-level SWS was estimated by truncating the vertical SM profile at the predicted ST, thereby preventing water from being allocated below the soil–bedrock interface. In spatial cross-validation, the ST model showed moderate spatial transferability (R 2 = 0.46, CCC = 0.62, and ubRMSE = 41.78 cm). Downscaled SM in the 0–100 cm profile was evaluated against Tibet-Obs measurements for the three upper layers. Station-based temporal Pearson’s R (mean R 2; CCC) values were 0.56 (0.40; 0.23), 0.28 (0.24; 0.06), and 0.28 (0.16; 0.02) for the 0–10, 10–40, and 40–100 cm layers, respectively. The corresponding ubRMSE (bias) values were 0.15 (−0.03), 0.12 (+0.04), and 0.15 (+0.03) m3/m3, respectively. Point-scale performance was comparatively stronger in the surface layer but weak in the 40–100 cm layer (R = 0.28; CCC = 0.02); therefore, layer-resolved SWS estimates in deep-soil regions should be interpreted cautiously as predictor-conditioned downscaled reanalysis rather than as fully in situ-validated estimates. Direct validation of the 100–200 cm layer was not feasible because in situ observations below 80 cm were unavailable; therefore, this layer should be interpreted as a profile-consistent reconstruction rather than a directly validated product. Because 81.30% of the estimated difference between the FD and ST-constrained approaches originated from the reconstructed 100–200 cm layer, the absolute magnitude of the reported storage difference and trend reduction should be interpreted with caution, although the direction of the structural bias was consistent across GLDAS and ERA5-Land. The ST-constrained SWS averaged 504.06 km3, whereas the conventional FD approach yielded 1,255.85 km3. Thus, ignoring soil–bedrock constraints caused an overestimate of 751.79 km3 (149.15%). Sensitivity analysis further showed that FD-based SWS increased nearly linearly with assumed profile depth, whereas ST-constrained estimates stabilized once the physically available storage capacity was reached. The reconstructed dataset captured a pronounced seasonal cycle and a significant wetting trend. Specifically, the ST-constrained framework indicated a moderate increase of 4.99 km3 yr−1 during 2009–2019, compared with 10.78 km3 yr−1 under the FD assumption. These findings show the importance of incorporating ST as a spatially explicit lower boundary when reconstructing shallow subsurface water storage in topographically heterogeneous mountain regions.