Spatial distribution of DFP-TWS and DFP-SSM
The DFP-TWS estimates and Nash-Sutcliffe efficiency (NSE) between TWSA from JPL RL06 v03 Mascon (JPLM) and reconstructed TWSA calibrated by JPLM (hereafter, JPLM-REC) for global 121 river basins are calculated using Eqs. 3, 4 (see Methods: Average daily fraction of P transformed into TWS (DFP-TWS)). The results are presented in Figs. 1a, b. The spatial distribution and specific result of DFP-SSM (definition can be seen in Methods) are also shown in Figs. 1c, d, and Supplementary Table 1. Please note that the SSM time series is anomalies relative to 2004-2009 mean baseline (hereafter, SSMA). The reconstructed SSMA time series is defined as SSMA-REC.
From Fig. 1a, large DFP-TWS is observed at high northern latitudes. In these basins, the daily mean snowfall (temperature) is generally large (Fig. 2a) (low, (Fig. 2b)), winter P is mostly stored as snow and ice until spring, and the ratio of ET to P (ET/P) is also generally smaller than that in low latitudes (Fig. 2c). Winter snowfall and freezing processes increase the average level of transformation of P into TWS, which is also consistent with the average concept of DFP-TWS. On the other hand, DFP-TWS is comparatively low in regions like Africa, Australia, and Central Asia, primarily due to the high ET/P as shown in Fig. 2c. This is because high ET negatively impacts the retention capacity of water storage. About 64% (area-weighted average) of global land P is transformed into TWS from 2002 to 2021, with 69% and 46% in humid and arid basins, respectively. However, it should be noted that smaller NSE values usually reveal large uncertainties in DFP-TWS estimates in north-central Africa (e.g., the Niger River Basin) and Central Asia (e.g., the Tarim River Basin) (Fig. 1b), these basins are in or around deserts with little P.
There is a dense vegetation cover in southern Asia, Central Africa, and northern South America (Fig. 2d). High NDVI generally increases ET, but ET/P is kept at a low level (ET accounts for a small proportion of P) in southeastern Asia and northern South America (Fig. 2c). Although some afforestation and ecological rehabilitation efforts are not conducive to the regional groundwater recharge in some arid and semi-arid areas31, extensive vegetation cover facilitates infiltration and the water retention capacity, allowing water to be fully absorbed by land and favoring P transformation32. However, there is a high proportion of clay in these three regions, as well as Australia (dark green basins) (Fig. 2e), and the smaller soil particle structure is not conducive to the transformation of P into TWS, which may have prevented these regions from having a larger DFP-TWS (Fig. 1a).
In Fig. 1c, DFP-SSM is generally large in Central Asia, northern Africa, Australia, and southwestern North America, which is spatially similar to Fig. 1 of Liu et al.11. This maybe resulted from low regional SSM (Fig. 2f) favoring the transformation of P into SSM (Fig. 1c). However, there are also significant differences in specific regions, such as northeastern Asia, southern Africa, and northern North America. In Eq. 2 of Liu et al.11, the increased SSM is considered as part of the P transformation only when P occurs. This may not be very accurate in high irrigation and capillary rise regions where the positive SSM increment may not be entirely due to P. Eq. 3 (replace TWSm with SSM) directly describes the fraction of P transformed into SSM and excludes other hydrologic processes that contribute to SSM. Therefore, differences in methods may result in some differences in the estimate. Although the stored P fraction calculated from different SM and P sources shared similar spatial distributions13,33, the results were still sensitive to model parameterization schemes and P information (e.g., CLM and GLDAS-Noah), even with the same soil depth34. Similarly, the method of Zhu and Yuan12 ignores other factors that can make ∆TWS positive, thus leading to some discrepancy between our estimates in Fig. 1a and their results. Differences in study period and other factors can also have some effect on the DFP-TWS estimates. In Fig. 1d, the NSE values are low in northeastern Asia and southwestern Hudson Bay, so the DFP-SSM should be interpreted with caution.
Correlation of DFP-TWS with drivers
DFP-TWS can be affected by a variety of factors. Here we use boxplots to investigate the correlation between DFP-TWS (DFP-SSM) and several drivers in global 121 river basins, and the results are presented in Fig. 3.
The mean snowfall and ET/P have a good correlation with DFP-TWS (Figs. 3a, b). In general, the higher the mean snowfall, the more P is stored in the form of snow, which has a positive effect on the P transformation into TWS. The larger the mean ET/P, the more P is evaporated, and the P stored on land will be less. This can also explain why DFP-TWS is usually larger in humid basins than in arid basins (see Results: Spatial distribution of DFP-TWS and DFP-SSM). In addition, there is a positive correlation between mean NDVI and DFP-TWS (Fig. 3c). Soil texture is an important factor influencing DFP-TWS. DFP-TWS decreases progressively with increasing surface clay (Fig. 3d), and it also exhibit similar characteristics for the entire clay layer (Fig. 3e). Neither the surface sand nor the entire sand layer has a clear relationship with DFP-TWS (not shown). Presumably, sand soils typically have a large porosity, so they can more easily infiltrate water into deeper soil layers or aquifers during the transformation of P into TWS. However, sand soils also have the disadvantage of poor water retention capacity, which can easily lead to TWS loss. Therefore, it is difficult to understand the transformation of P into TWS by sand, resulting in an insignificant correlation with DFP-TWS. Although clay has high water-holding capacity, it has a poor permeability to P. As the clay increases, more P is exposed to the ground, and lost through ET and other forms, resulting in a negative correlation with DFP-TWS. There is a positive correlation between mean SSM/SM and DFP-TWS (Figs. 3f, g), which may be related to groundwater recharge. In regions where SM is high (Fig. 2g), groundwater recharge and groundwater storage are both large35,36. In other words, the higher the SM, the easier it is for the vadose zone to reach the field capacity, and SM recharges groundwater when P occurs, thereby increasing TWS.
However, SSM shows opposite effects on DFP-TWS and DFP-SSM (Figs. 3f, h). This is because the DFP-SSM is calibrated with SSM, and characterizes directly the contribution of P to SSM without involving other recharge or transformation processes. As the mean SSM becomes more saturated, the DFP-SSM will be smaller, which is consistent with the conclusion in McColl et al.5. Typically, regions with dense vegetation tend to be wetter than those with sparse vegetation (Fig. 4), and the hydraulic conductivity of soil increases with increasing SM37. Therefore, P is largely transformed into high drainage and runoff, and water falling on the soil is difficult to recharge SSM by gravity, which results in a small P fraction. On the other hand, a large amount of P is intercepted by the vegetation canopies, and cannot reach the surface soil layer. Therefore, DFP-SSM decreases progressively as the mean NDVI increases (Fig. 3i), which is in agreement with the conclusion from Kim and Lakshmi13. TWS, unlike SSM, includes all water components (e.g., surface water and canopy water), and high vegetation cover is conducive to water retention, which makes NDVI have opposite effects on DFP-TWS and DFP-SSM (Figs. 3c, i).