the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 13 Nov 2025
DRYP 2.0: A hydrological model for local and regional scale across aridity gradients
Abstract. The increasing demand for freshwater resources due to population growth, economic development, and climate change, requires more accurate representation and quantification of the key components of the water balance at relevant scales that goes beyond catchment domains. To address this need, we present DRYP 2.0 (DRYland water Partition model), a new version of a parsimonious, process-based, spatially distributed hydrological model (DRYP). DRYP 2.0 introduces several new capabilities, including the hydrological representation of small ephemeral ponds and large lakes, multiple interacting hydrogeological domains within a single-layer groundwater model, and vegetation canopy interception and evaporation to better capture the effects of vegetation on hydrology across different climatic gradients. Computational performance has also been enhanced through more efficient algorithms that reduce simulation time for long runs and/or over large spatial domains. We demonstrate these advances using high-resolution (1 km, 1 h) simulations over the Horn of Africa Dryland region (2,000,000 km²) as well as through various synthetic numerical tests. The results highlight the ability of the model, even without calibration, to reproduce global remote sensing data such as soil moisture, actual evapotranspiration, and total water storage, while also significantly reducing computation time. Furthermore, the explicit inclusion of multiple hydrogeological domains reveals important impacts on water table depth, with implications for improving global-scale simulations of the water balance.
- Preprint
(9359 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 12 Jan 2026)
- RC1: 'Comment on egusphere-2025-5316', Anonymous Referee #1, 10 Dec 2025 reply
-
RC2: 'Comment on egusphere-2025-5316', Anonymous Referee #2, 16 Dec 2025
reply
The manuscript by Quichimbo et al. presents an overview of the newest version of the DRYland water Partition model – a gridded, process-based hydrologic model that simulates the water budget and relevant fluxes for arid regions. Updates to the model include the simulation of water partitioning for the vegetation canopy, groundwater-lake interactions, ponds, and more. The authors also state that the new version of the model has improved computational efficiency compared to previous versions. Two synthetic applications of the model are shown, including simulations illustrating groundwater-lake interactions, and variable transmissivity functions for subsurface water fluxes. Results from an uncalibrated realization of the model are also compared to remote sensing data for the Horn of Africa Drylands, and the authors evaluated the model to match the data well.
Evaluation:
The formulation and conceptualization of the DRYP 2.0 model as described in the manuscript are novel and may provide valuable insights for researchers interested in understanding hydrologic budgets in arid areas. The conceptualization is within the general scope of GMD. With this being said, I have three primary issues with the manuscript that should be addressed/clarified prior to the manuscript being accepted.
- Although the introduction situates the paper within the context of ephemeral river networks, the manuscript does not assess the model’s performance in capturing these processes. I would recommend reworking the introduction to avoid giving the impression that the paper evaluates the model’s ability to simulate ephemeral river networks.
- While I agree with the authors that synthetic numerical experiments are useful for understanding model behavior under hypothetical conditions, I would argue that it is necessary for the synthetic test to still demonstrate that the newly introduced model features adequately reflect expected physical behavior. Sometimes the literature will refer to these as ‘laugh tests,’ because they provide the most basic evaluation of whether a model behaves in a physically plausible manner (see Clark et al., 2021, https://doi.org/10.1175/JHM-D-20-0175.1). Common approaches include comparisons with analytical solutions to simplified forms of the governing ODEs, or benchmarking against an established numerical model. Here are some examples where this has been previously done: https://doi.org/10.2113/4.1.206, https://doi.org/10.1002/2016WR019672, https://doi.org/10.1002/hyp.11476, https://doi.org/10.1175/JHM-D-16-0284.1, and https://doi.org/10.1175/JHM-D-20-0175.1. While the qualitative evaluation of the model presented herein is informative, I would encourage the authors to include an experiment that more objectively evaluates the ability of the model to represent expected hydrologic behaviors.
- There are several instances in the description of the model architecture where I would have appreciated more information about how several of the fluxes were parameterized. This is especially true for the description of the riparian UZ store and its respective fluxes. Note that GMD states that for model description papers, the expectation is that it should be theoretically possible for an independent scientist to construct a model that, while not necessarily numerically identical, will produce scientifically equivalent results. In the manuscript’s current form, I’m not sure that the reader is provided with enough information to support this level of reproducibility. I have called several of these out in the specific comments below.
Specific comments:
L68: simulating ‘small’ headwater catchments at a scale of 500-m actually sounds quite coarse to me.
Fig 1c: do the numbers shown here correspond with those in Fig 1a? It’s not clear. And if so, why aren’t all 11 processes included?
Fig 1d: Is there no capillary rise from the SZ to the RUZ? Is SZ not connected to the SW?
L123: Sometimes RUZ is described as ‘adjacent’ to streams, other times it is described as ‘below’ streams. I think the latter is correct if I understand correctly.
L127: the ‘bi-directional water exchange’ isn’t always clear, especially between different stores and the SZ.
L135: What is the routing method? Is it kinematic wave? Something more simplified?
L174: What about throughfall?
L180: It’s not clear how a predefined stream network simulates ephemeral channels.
L183: It seems to burry the lede to reference transmission losses in the subtitle and then refer the reader to another paper.
L187: this mentions that RUZ is beneath the stream, but other places mention “adjacent” to the stream.
L191: It seems like your schematic shows that the SZ can transport water to the UZ via capillary rise, but I'm not sure if this is reflected here.
L191: Many of the fluxes in the equation are not described in the manuscript. It seems important to include. From my understanding RUZ is a new component of DRYP 2.0. I especially would like to see how Q_BF is parameterized because it seems very important to streamflow. Same for recharge for riparian zones.
L199: Is this AET parameterization for RUZ and UZ? There’s also a typo at the end of this line.
L213: again, it’d be nice to include information on how these flux equations were parameterized.
Fig. 3: doesn’t seem to be referenced anywhere in the main body of the manuscript.
L281: Typo
Table 1: Why is there an asterisk next to 0.083? For regional stream networks, assuming a constant stream width of 10m doesn’t seem reasonable. How is K_ch incorporated in the model? Why is it a constant value?
L400: It feels like the model isn't evaluated against any datasets with a time step that reflects the 'temporally dynamic' conditions of arid systems, as mentioned in the introduction.
Citation: https://doi.org/10.5194/egusphere-2025-5316-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 248 | 58 | 25 | 331 | 19 | 25 |
- HTML: 248
- PDF: 58
- XML: 25
- Total: 331
- BibTeX: 19
- EndNote: 25
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
The manuscript presents the updated version 2.0 of the hydrological modelling software DRYP 1.0 described in GMD in 2021:
Quichimbo, E. A., Singer, M. B., Michaelides, K., Hobley, D. E. J., Rosolem, R., and Cuthbert, M. O.: DRYP 1.0: a parsimonious hydrological model of DRYland Partitioning of the water balance, Geosci. Model Dev., 14, 6893–6917, https://doi.org/10.5194/gmd-14-6893-2021, 2021.
The modeling software aims to represent important hydrological processes in the dry regions of the globe, including transmission losses in channels, local and non-local groundwater recharge, and streamflow intermittency, processes difficult to simulate, particularly in large-scale hydrological models. The update includes the consideration of additional water storage compartments (interception, ponds, and lakes) as well as an improvement in run-time due to more efficient algorithms and the use of Fortran for some software parts. The model is applied for a few synthetic cases and the Horn of Africa dryland (HAD) region (2 million km2 with a spatial resolution of 1 km2 and hourly time steps). The uncalibrated model, which is able to represent water flows between streams and groundwater as well as between lakes and groundwater based on hydraulic head gradients, is compared to remote sensing-based estimates of evapotranspiration, soil moisture, and terrestrial water storage anomaly.
Overall, the description of the model enhancement lacks clarity. Also, the three model applications are insufficiently described in the manuscript. The application for HAD lacks information about which lakes are taken into account in this application, and the resulting exchange flows between lakes and groundwater, and streams and groundwater. And what information was used to simulate interception and ponds, and what is the impact of these new components on the model results? I disagree with the statement in L10 of the abstract, “The results highlight the ability of the model, even without calibration, to reproduce global remote sensing data such as soil moisture, actual evapotranspiration, and total water storage”. Only correlations were assessed, and the correlations for soil moisture were mostly rather poor, with the majority of the area showing a Pearson correlation coefficient below 0.6. The manuscript does not show how meaningfully and successfully a model simulating hydraulic head gradients and including many parameters that are difficult to derive by model calibration can be applied at a large scale. The synthetic experiments only cover small regions, and the HAD application is not well described and evaluated enough to be convincing. I suggest that the HAD application is run with v1 and v2 of the DRYP model, and differences in the fit to the three remote-sensing-based variable values (plus streamflow) are analysed.
Main comments
Introduction
I suggest expanding the introduction regarding the following:
Methods
The description of the model (algorithms) is not sufficiently clear. Both the synthetic experiments and the application to HAD need a lot of clarification to make them well understandable by the reader (see my specific comments below). In particular, no information is provided on the integration of lakes and ponds in the HAD application in section 4.2, nor on how the newly introduced interception was parameterized in this application (was it even included?). The appendix provides detailed descriptions of the new model components, which are, however, difficult to follow. In particular, information on the values or sources of data is not provided for all the parameters of the algorithms. Just as an example, in A3, it is stated in L486 “Here, it is assumed that the parameters a, hmax, and Amax are known”. But no information is given in the application to HAD about the values/sources used for these three parameters. And this is exactly the challenge of large-scale modeling: How to find suitable values for, e.g. a or Amax? And how to decide which surface water bodies interact with the groundwater (here: lakes) and which not (here: ponds)?
Results
Also the information provided in the results section is not precise enough to be readily understandable (e.g., regarding the temporal resolution and extent of the evaluation data used for correlation analysis in HAD). Besides, it is methodologically incorrect to evaluate the correlation with GRACE TWSA for spatial units of 0.25° due to the inherent low spatial resolution of GRACE signals (see my comments to section 3.3 below).
Specific comments
The reference Quichimbo et al. (2021) should cite the GMD paper, not a Zenodo page.
L9: Write “about 2.1 million km2” not “2,000,000 km2”
L21: I think “streamflow intermittence” is a more suitable term than “stream intermittence”.
L31: In standard hydrological models, groundwater is assumed to be discharged to the river and maintain low flows. So what is described here as non-local hydrological connectivity is normally represented by hydrological models (and not omitted, as written in L25).
L54: Clarify why you use the term “process-based” here
Fig. 1 correct typos such as throughall, Developemement, .... Include lakes in Figs. 1 b and c.
L187: “River cells are characterised by the riparian unsaturated zone, while non-river cells are characterised by the hillslope unsaturated zone.” Does this mean that a cell either has at store UZ (in Fig. 1) or a store RUZ (in Fig. 1) but never both? I.e. there is no diffusion groundwater recharge in river cells? If so, please indicate this in Fig. 1, e.g. in Fig 1c by making whole cells either blue river cells or green hillslope/non-river cells.
L191: For easier understanding, I suggest having two equations instead of eq. 7, one for the river cells and one for hillslope/non-river cells.
L215: Is QPND routed as overland flow to the downstream cell or as streamflow? Please clarify in section 2.5.1 or before which type of surface water bodies is represented as “pond” by Eq. 10. What about the reservoirs and small lakes mentioned in the title of section 2.5.1? Hamonize with the title for A3.
L234 and Fig. 2: Please explain whether it is assumed that a lake does not have a surface water outflow? What precisely is Q? If you assume no surface water outflow of a lake, under what circumstances is such a representation suitable for representing reality?
Section 2.2.7 In Eq. 11., “Kaqh” is included, and not transmissivity. Please clarify under what circumstances the aquifer transmissivity is described by a time-dependent K*h and when by the transmissivity options described in section 2.2.7. Besides, I understand that there is only one layer of saturated grid cells. So what does a depth-dependent transmissivity as shown in Fig. 3 mean? Do you mean a depth-dependent hydraulic conductivity of the aquifer that results in different T of the saturated layer dependent on the C, L or E function? This needs to be explained much better. Please provide a suitable equation with transmissivity.
Sections 2.3.1, 2.3.2: Please indicate all parameter values used for the synthetic experiments (maybe in a supplement). The requirement is that the reader could recreate these experiments. Also the boundary conditions are not clearly indicated in the figures and the text.
L271: “Initial conditions assumed dry soil, with water content equal to the wilting point. For groundwater, the initial water table was specified as a flat surface located 1 m below the lake bottom elevation (95 m above the reference datum).” I cannot recognize this in Fig. 4. Also, Fig. 7 does not show a flat gw table at t-0, and the water table is exactly at the lowest point of the lake. Very confusing! It seems that Figure 4 b is incorrect regarding the initial water table; also indicate the elevation of the constant head boundary in m (<84 m). And indicate what the role of the maximum lake extent (Fig. 4) or maximum lake depth is in the modeling. It is not reached at t-4, will it be reached at steady-state?
Figure 5: I do not see the flow boundary. And what is indicated with “model domain”?
Line 281: Figure number missing.
Figure 6: Indicate Shabelle River as well as the Ogaden area in Fig. 6
Table 1: hlake: what exactly is meant by hlake; and what about the location of extent of lakes? Were ponds considered? Please describe how hlake is derived by Khazai et al. (2022). What is the source of Kch = 10.9 mm/h?
Section 2.4.2: Also compare to in-situ observed streamflow, e.g. the data available at GRDC, at least roughly, even if observed data are only available before 2000. If only correlations between the ESA CCI soil water content and the DRYP value are analyzed, is a normalization necessary? Provide information on the temporal resolution used for the correlation analysis in 3.3 (or provide this information in 3.3.)
Section 2.4.3, L330 Indicate with what forcing the 20 years were driven.
Section 3.1: Please provide the water balance of the model domain, in particular, which fraction of the precipitation becomes evaporation from the lake and the soil, and which fraction discharges via the constant head boundary. Please explain whether groundwater table elevations are shown in Fig. 7b, c, and d. If so, what is the water table of the lake? And why is the (groundwater)table in the lake lower than at P2? Is the lake a gaining lake, and under what circumstances would the lake discharge water into the saturated zone and thus towards the constant head boundary? And explain better the relevance of the model experiment for the simulation of the real world.
Section 3.3
While GRACE terrestrial water storage anomalies in the form of CSR-M mascons (Save et al., 2016) are provided with spatial resolution, evaluation can only be done for spatial aggregates of at least about 100,000 to 200,000 km2 (see also https://www2.csr.utexas.edu/grace/RL06_mascons.html). Therefore, 0.25° grid cell values cannot be compared to the DRYP output (Fig. 9c), but both CSR-M and DRYP values have to be first aggregated to regions of such a size. Save et al. (2016) write “While these mascon solutions are estimated on geodesic grid roughly 120 km wide, that is not the resolution of these GRACE solutions. These GRACE mascon solutions are limited by the band-limited nature of GRACE, with an approximate resolution of around 250–300 km near the equator.”
Fig. 9. It is the temporal correlation that is shown. Provide, in the figure caption, the temporal resolution and time period of the correlation analyses.
L401: While correlation is high, the temporal variability of AET is underestimated by DRYP and it looks like the long-term mean evapotranspiration is underestimated, too.
Appendix A4: I do not see the relevance of the synthetic experiment. In addition, the applied precipitation amount is unrealistically high for almost all regions of the globe.
Code and data availability: To make all described modeling exercises reproducible, scripts, input data (in particular parameter values) and output data files have to be provided. It is not enough to provide the code and the sources of the input and validation data.