the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 20 Jul 2026
Ecosystem mobilization of subsurface water revealed by global active root-zone storage
Abstract. Subsurface water sustains vegetation between precipitation events and shapes ecosystem responses to hydroclimatic variability. Yet existing observations capture either near-surface soil moisture or bulk terrestrial water storage, leaving the dynamic component of subsurface water that vegetation actively mobilizes poorly understood at the global scale. Here we analyze a global reconstruction of active root zone water storage (aSrz), defined as the depth of subsurface water ecosystems actively mobilize for evapotranspiration, from 2001 to 2020. The global area-weighted mean aSrz is about 100 mm, with the largest values concentrated where precipitation supply and atmospheric demand are climatologically comparable. Trends in climatic water balance are most strongly reflected in aSrz in seasonally driven systems such as croplands, savannas, and seasonal forests, and more weakly in evergreen tropical forests and tundra. The turnover time of the active component distinguishes rapidly cycled storage in warm and seasonally dry regions from slowly cycled storage in boreal and Arctic regions. Comparison with satellite gravimetry shows that aSrz co-varies with bulk terrestrial water storage at monthly scales, but the coupling weakens and becomes more regime-dependent at interannual scales, especially in snow-dominated and dry regions. These patterns identify where bulk storage provides information on subsurface water accessed by ecosystem and where changes in bulk storage mainly reflect changes in other water stores. Overall, these findings establish ecosystem-accessed water storage as an observation-based dimension of the terrestrial water cycle, revealing patterns of ecosystem water access and change that bulk storage and climatic wetness indicators do not resolve, and providing a foundation for assessing ecosystem water vulnerability under hydroclimatic change.
- Preprint
(2419 KB) - Metadata XML
-
Supplement
(954 KB) - BibTeX
- EndNote
Status: open (until 09 Oct 2026)
-
RC1: 'Comment on egusphere-2026-3923', Anonymous Referee #1, 13 Aug 2026
reply
-
AC1: 'Reply on RC1', Shijie Jiang, 16 Aug 2026
reply
We thank the referee for the detailed and critical review. This response is intended as a preliminary clarification during the discussion stage. Because the main concern raised by the referee bears directly on the scientific premise of the study, we think it is important to clarify our position here. A preliminary point-by-point response and supporting analyses are attached.
First, we agree that our description of the deficit literature was too narrow. Dynamic deficit trajectories can represent recharge, drawdown, and antecedent storage memory (e.g., Lapides et al., 2022), including explicitly non-resetting formulations that allow multi-year deficit accumulation (Ehlert et al., 2024). Gao et al. (2024) describe root-zone water storage and water deficit as two sides of the same coin. We will revise this framing and consistently describe aSrz as an observation-constrained, model-derived storage state.
However, we respectfully disagree with the implication that a shared first-order water-balance basis is sufficient to treat aSrz as redundant with, or effectively replaceable by, a deficit-derived storage trajectory. In a deficit formulation, the storage state is defined directly from the cumulative imbalance between prescribed water input and ET. In our framework, that imbalance is not prescribed a priori for the ecosystem-accessed reservoir. Instead, liquid input is dynamically partitioned between an ecosystem-accessed reservoir and runoff/drainage pathways, with the fraction entering the accessible reservoir depending on its current state and inferred process parameters. The accessible reservoir is then depleted through ET, and aSrz is diagnosed from the realized range of this prognostic storage state. Thus, the two approaches share water-balance bookkeeping, but differ in what is treated as ecosystem-accessible input and, consequently, in how the storage trajectory itself is constructed.
We have therefore tested directly whether a deficit-derived storage state can serve as a substitute for aSrz in the analyses considered here, by calculating the deficit-derived storage using the same data products as MOREDO. At the monthly timescale, correspondence between the two states is strongly regime-dependent, with particularly weak agreement in cold and dry regions and a global median pixel-wise correlation of r=0.39. Across grid cells, their 2001–2020 climatological means also show markedly different spatial patterns, with a spatial correlation of r = -0.31. These results do not establish that either representation is universally superior, but they do not support assuming that one can be substituted for the other. A preliminary functional comparison using GPP products that were not used as model training targets also shows different relationships with ecosystem functioning. After accounting for meteorological conditions, precipitation, LAI, and previous-month GPP, aSrz and deficit-derived storage show different spatial relationships with GPP anomalies, with positive partial correlations for aSrz extending more consistently across many vegetated regions. This provides an additional indication that the two storage representations carry different information in relation to ecosystem functioning. We will quantify these differences more fully in the revision.
Most importantly, the scientific contribution of this manuscript does not rest on claiming that dynamic plant-accessible storage is itself a new concept. The study addresses how the reconstructed active-storage state is organized globally, how it has changed, how rapidly it turns over, and when bulk terrestrial water storage can be informative of water accessed for ecosystem functioning. We will revise the Introduction carefully to avoid suggesting that the novelty relies on the ecosystem-accessible bucket and further highlight the concrete scientific questions. The studies cited by the referee provide important methodological and conceptual context, but they do not establish that these scientific questions have already been answered using dynamic deficit states. The major global deficit-based studies most directly relevant here have primarily used deficit trajectories to infer rooting-zone storage capacity from cumulative-deficit extremes or return periods (e.g., Wang-Erlandsson et al., 2016; Stocker et al., 2023). Stocker et al. (2023), for example, explicitly describe their global storage-capacity estimate as a snapshot and do not consider its temporal changes. We will revise the Introduction to position this literature more precisely. We will also include a direct comparison with deficit-based storage to clarify where the two representations agree and where they differ, while keeping the focus of the study on the global dynamics of active storage.
Overall, we greatly appreciate the referee's comments concerning the literature framing and the interpretation and evaluation of the model-derived state. The attached preliminary point-by-point response provides additional detail on how we are addressing the individual comments and includes the supporting analyses currently available. A complete response incorporating the corresponding manuscript changes and finalized analyses will be provided with the revised manuscript after the discussion phase. At the same time, we disagree that a shared first-order water-balance basis alone establishes redundancy of the scientific contribution. For methodological (or conceptual) overlap to make the present scientific analysis redundant, one would need to show either that the scientific questions addressed here have already been answered by existing deficit-based studies, or that a deficit-derived representation reproduces the findings at issue. The literature cited in the review does not establish the former, while our preliminary direct comparison does not support the latter.
References:
Ehlert, R. S., Hahm, W. J., Dralle, D. N., Rempe, D. M., & Allen, D. M. (2024). Bedrock controls on water and energy partitioning. Water Resources Research, 60(8), e2023WR036719.
Gao, H., Hrachowitz, M., Wang-Erlandsson, L., Fenicia, F., Xi, Q., Xia, J., ... & Savenije, H. H. (2024). Root zone in the Earth system. Hydrology and Earth System Sciences, 28(19), 4477-4499.
Lapides, D. A., Hahm, W. J., Rempe, D. M., Whiting, J., & Dralle, D. N. (2022). Causes of missing snowmelt following drought. Geophysical Research Letters, 49(19), e2022GL100505.
Stocker, B. D., Tumber-Dávila, S. J., Konings, A. G., Anderson, M. C., Hain, C., & Jackson, R. B. (2023). Global patterns of water storage in the rooting zones of vegetation. Nature Geoscience, 16(3), 250-256.
Wang-Erlandsson, L., Bastiaanssen, W. G., Gao, H., Jägermeyr, J., Senay, G. B., Van Dijk, A. I., ... & Savenije, H. H. (2016). Global root zone storage capacity from satellite-based evaporation. Hydrology and Earth System Sciences, 20(4), 1459-1481.
-
AC1: 'Reply on RC1', Shijie Jiang, 16 Aug 2026
reply
Data sets
Global active root-zone water storage Shijie Jiang and Georgios Blougouras https://doi.org/10.5281/zenodo.21136948
Model code and software
MOREDO: A Multi-Observation Root-zone Ecohydrology DiagnOstic model Georgios Blougouras and Shijie Jiang https://doi.org/10.5281/zenodo.20692925
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 149 | 61 | 19 | 229 | 34 | 15 | 13 |
- HTML: 149
- PDF: 61
- XML: 19
- Total: 229
- Supplement: 34
- BibTeX: 15
- EndNote: 13
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Review of "Ecosystem mobilization of subsurface water revealed by global active root-zone storage"
Â
SummaryÂ
Â
This manuscript analyzes a 2001-2020 global reconstruction of active root-zone water storage, aS_rz, generated by the MOREDO model. It maps the mean and trends of this modeled state, relates those patterns to climate and land-surface attributes, defines a storage-to-ET timescale, and compares aS_rz with GRACE terrestrial water storage anomalies.
The global analysis is potentially useful, but the conclusions exceed what the reconstruction presently supports. The central quantity is an unobserved internal state of a conceptual hydrological model, referenced to its simulated record minimum. The manuscript does not establish that this state contains information beyond what the simpler water-deficit calculation provides, which is what is critiqued and used as the motivation for this work. The claimed knowledge gap and novelty also overlook extensive work on time-varying plant-accessible storage and non-resetting root-zone deficits.Â
I recommend reject/resubmit. The manuscript should be reframed as an analysis of a model-derived relative storage index; meaningfully build on prior literature on this topic, include sufficient model evaluation to be reviewed independently of the companion preprint; and at a minimum compare the MOREDO output with the much simpler and widely used deficit calculation, using the same forcings/training targets.
Â
Major comments
Â
The distinction between aS_rz and existing deficit approaches is overstated. Deficit methods generate a time-varying state; a capacity estimate is a subsequent summary of its extremes. Recharge, drawdown, persistence, and multi-year carryover therefore do not fall outside that framework. Storage and deficit have already been described explicitly as complementary (Gao et al., 2024. https://doi.org/10.5194/hess-28-4477-2024). Non-resetting deficits have also been used as dynamic states to explore bedrock water use and post-drought runoff (Ehlert et al., 2024. https://doi.org/10.1029/2023WR036719; Lapides et al., 2022. https://doi.org/10.1029/2022GL100505).
MOREDO adds learned partitioning, multiple conceptual reservoirs, and basin-to-grid parameter transfer. Those are model developments: subtracting the record minimum from a modeled storage trajectory is not a new observation of active storage. The manuscript should state the incremental methodological contribution and test whether it adds information beyond the established deficit water-balance calculation (Gao et al., 2014. https://doi.org/10.1002/2014GL061668; Wang-Erlandsson et al., 2016. https://doi.org/10.5194/hess-20-1459-2016; Stocker et al., 2023. https://doi.org/10.1038/s41561-023-01125-2). Let me put it another way. Comparable minimum-referenced storage trajectories could be extracted from many hydrological or land-surface models containing vegetation-accessible reservoirs. The essential question is therefore what independent evidence demonstrates that the MOREDO state is better identified, or contains more hydrological information, than existing modeled storage states or a plain deficit calculation. An important virtue of the stand-alone deficit approach is its transparent bookkeeping based on externally estimated precipitation and ET fluxes, which limits dependence on the internal architecture and latent states of a particular hydrological model. These flux products have their own uncertainties, but those uncertainties are more readily exposed and propagated than the structural assumptions governing an unobserved internal reservoir.
Â
Although Appendix A summarizes the conceptual structure and Section 2.1 describes the principal training data and design, the detailed model architecture and the decisive sensitivity, identifiability, and product-evaluation evidence are deferred to a non-peer-reviewed companion manuscript (Blougouras et al., 2026. https://doi.org/10.22541/essoar.15005507/v1). Neither S_a nor aS_rz is a training target. ET constrains the accessible bucket only because the architecture assigns ET to that bucket; runoff constrains the complementary pathway; and GRACE constrains the sum of modeled stores, not their decomposition. Reproducing these targets does not establish a unique internal partition.
The present manuscript must contain enough information to evaluate the reconstruction independently. It should compare $aS_{rz}$ with a transparent deficit calculation using identical precipitation, ET, and snow inputs/targets.Â
Â
This is more of a comment about Moredo than this manuscript, but it matters for this manuscript. Pixel-scale deficit formulations generally do not include observed runoff directly: excess water input after a deficit has been replenished is treated implicitly as drainage or runoff, while discharge observations are generally available only at the integrated catchment scale. In MOREDO, each gauged basin is treated during training as one lumped unit: forcings and attributes are aggregated over the basin, and simulated runoff is compared with outlet discharge. during global inference, each 0.25-degree cell is run independently, without routing, upstream accumulation, lateral exchange, or a cell-scale runoff observation. Outlet discharge therefore does not directly constrain the partitioning or water balance of individual grid cells. It constrains an attribute-to-parameter relationship at catchment scales, whose validity at grid-cell support is assumed.
This matters directly for aS_rz because the learned partition between accessible water and runoff controls its magnitude. Effective catchment parameters may absorb subcatchment heterogeneity and partitioning of precip into runoff vs storage avaailble to plants.Because both runoff generation and storage dynamics are nonlinear, parameters inferred by applying the model to basin-mean forcing need not reproduce the area-average behavior obtained by applying it to heterogeneous grid-cell forcingsÂ
Line-level comments
lines 10-13: the stated observational gap is too broad. existing observations include profiles of soil moisture and water potential, lysimeters, boreholes, geophysics, rock-moisture measurements, groundwater wells, sap flow, eddy covariance, isotopes, and catchment water balances. the defensible gap is the absence of a direct, continuous, globally gridded observation of the entire plant-accessible domain.
line 30: "unsaturated stores," "weathered substrate," and "shallow groundwater connections" are overlapping categories of hydrologic state, material, and pathway. In other words, you’re contrasting apples and orangesÂ
lines 42-64: the discussion of observations is restricted almost entirely to satellite products and land-surface models. field and critical-zone observations need to be represented, including work that distinguishes plant-accessible capacity, available water, and observed drawdown (Klos et al., 2018. https://doi.org/10.1002/wat2.1277).
lines 65-70: deficit approaches do retain recharge, drawdown, persistence, and non-resetting multi-year behavior. revise this paragraph around the actual difference in model structure.
lines 70-78: the companion study introduced this notation and model formulation not the general concept of dynamic ecosystem-accessible storage. apparent or lower-bound accessible storage was already articulated directly (e.g. Hahm et al., 2024. https://doi.org/10.1029/2023WR035362).
figure 1: this relates to the claims of novelty. compare the time series with e.g. the deficit time series and bucket schematic in lapides et al. (Lapides et al., 2022. https://doi.org/10.1029/2022GL100505). the two represent the same first-order cumulative balance (moredo just constrains input differently).Â
lines 90-100: the first study question substantially overlaps the global analysis of climate, vegetation, topography, and rooting-zone storage in other papers, e.g. stocker et al. (Stocker et al., 2023. https://doi.org/10.1038/s41561-023-01125-2). clarify which questions are genuinely new.
lines 115-118: convergence of biome-mean values with increasing window length does not test sensitivity to the timing of an extreme multi-year drought.Â
Â
Because GRACE TWSA enters the training loss, the subsequent aS_rz–TWSA coupling analysis is partly circular(?) and cannot independently validate the inferred relationship. This limitation should be acknowledged or tested through a GRACE-withheld experiment
Â
Please maintain a consistent distinction between an observation-constrained model output and an observed quantity; several statements in the abstract and conclusions blur this distinction.
Â
Throughout this was super confusing: the prefixed a in aS_rz and subscript a in S_a are difficult to distinguish while denoting "active" and "accessible." use clearly distinct symbols.