| 22 Jul 2026
A Novel High-Resolution Water Resources Assessment Model (HyDRA version 1.0) for Industrial and Domestic Water Use: A regional application for the Gulf Cooperation Council (GCC) Countries and Jordan
Abstract. Large-scale water models often generalize the methodology for developing regions by relying on coarse assumptions that fail to capture regional socioeconomic dynamics and the constraints of water-scarce systems, resulting in limited accuracy when evaluating water use across diverse economic and resource conditions. This gap is particularly evident in the Middle East, where data scarcity and underrepresentation hinder detailed assessments despite the region’s critical importance in a water-dependent economy. To address this, this study develops HyDRA (“High-Resolution Water Resources Assessment Model”) model, a model designed to improve the estimation of industrial and domestic water use through dynamic socioeconomic drivers and regionally available data. The model disaggregates industry into ten subsectors nearly covering all the industrial activities—electricity generation, petro-refining, chemical and petrochemical industries, cement, basic metals, food and beverage, paper and pulp, oil extraction, mining , and other manufacturing—each linked to distinct production indicators and corresponding water use intensities, alongside a domestically adaptive module that captures variations in per capita water use driven by demographic, economic, and infrastructure changes. Implemented at a fine spatial resolution of 3 arcseconds (~90 m), the model enables detailed, location-specific estimates of water withdrawal and consumption at both national and subnational scales. Its application to the GCC countries and Jordan over the period 1990–2025 provides a long-term reconstruction across hydro-economically contrasting systems, where the GCC represents one of the world’s most concentrated hubs of water-intensive industrial activity, while Jordan reflects a structurally distinct yet equally water-scarce system. Together, these regions form a critical testbed for analyzing water economy dynamics under extreme scarcity. By explicitly resolving sectoral production structures and socioeconomic adaptation, the model advances beyond conventional top-down scaling approaches, improving cross-country comparability while preserving regional specificity. Although demonstrated for the GCC and Jordan, the model structure is transferable and can be extended to other regions, given appropriate socioeconomic inputs and harmonized statistical data.
This paper builds a subsector-resolved model of industrial and domestic water use and runs it for the Gulf states and Jordan over 1990 to 2025. The region is a blind spot for the global models, and the authors have gathered more national data for it than I have seen in one place. So let me say up front that I think there is a useful model and dataset here. My problem is the distance between what the paper shows and what it says it shows. Several of the big claims are never tested, and parts of the accounting do not add up yet. I would recommend major revision. Most of what follows are questions.
Main comments
What water is being counted in the oil module? Section 2.1.8 says the external supply can be “seawater, freshwater, or desalinated water.” Section 3.1.3 then calls the result “industrial freshwater withdrawal.” Which is it, country by country? If the split isn’t known, why show the total as freshwater at all? For a region that runs on desalination this may be the most consequential number in the industrial module.
The oil intensities do not reproduce. Putting the page 17 parameters into Eqs. (11) and (12), I get roughly 1.84 and 0.69 m³ per m³; the paper reports 1.96 and 0.83. Which set went into the runs? And a question of convention while we are here: water that is injected and stays in the reservoir never returns to the sea or the aquifer it came from, so why isn’t it consumed under your own definition?
Eq. (8) puts all of non-revenue water into consumption, while water that reaches urban users is consumed at 10 percent. You note yourselves that the Qatar NRW figures include apparent losses as well as real ones. What is in those apparent losses, and why would any of it be consumptive? Then the UAE: its NRW was estimated “assuming a trajectory similar to Qatar’s improvements.” If that decline is prescribed, how can the falling UAE consumption-to-withdrawal ratio be read as evidence of better network efficiency?
Better than what? The abstract says the model “advances beyond conventional top-down scaling approaches,” and the Discussion ends by saying it can “substantially improve” on aggregated approaches. I looked for the conventional model that was run alongside and could not find one. Would a constant-intensity reconstruction, a single lumped industrial sector, or a plain GDP regression actually do worse against the same observations? Even a simple baseline would tell us what the extra complexity buys.
The evaluation is visual from start to finish. No bias, no RMSE, no slope, no R². Figure 3 is normalized to 2020, Figure 4 to a reference year the caption does not name, and the domestic series are pinned to the observed 2020 value, so agreement near that year is guaranteed. Why not report skill with the anchor year left out, with absolute values next to the normalized ones? And what would a failed test look like? Section 3.1.1 keeps explaining disagreement as reporting problems. That may well be true in places, but if agreement counts for the model and disagreement counts against the data, the model cannot lose.
Figure 3 is the clearest case. The paragraph beneath it opens with the comparison “highlights the model’s ability to capture long-term efficiency improvements.” Then come the panels: Saudi Arabia and Oman do not move the same way in the two series, and the UAE observations fall far more than the model does. Why open with the success statement? Say where it fits and where it doesn’t, then draw whatever conclusion the panels support.
TEF carries most of the industrial time series, so its form matters. Why average total GDP with two per capita quantities, in countries whose populations grew several times over? Why equal weights, and why 0.5 on the GDP term? These are choices, but the text presents them as properties of the system. It also says all three terms are “designed to exhibit a saturating, plateau-like response,” when only the GDP term is transformed. And one thing I kept wondering: when GDP contracts, does the model read that as intensity going up?
Which subsectors use TEF? Eq. (2) gives it a subsector index; Eq. (3) does not. Page 10 says it applies “for all the subsectors,” page 18 says it “was not considered” for oil, and the power-plant equation has no TEF in it. If it is a single national multiplier, what is validated in a section called “Validation of Sub-Sector Efficiency Changes”? Subsector-specific baseline intensities and subsector-specific trends are different things, and only the first seems to exist.
Figure 4 tests four subsectors on countries outside the region. Fair enough, that is where the data are. But how do the ranges of the GDP, electricity and energy-intensity ratios in those countries compare with what the model meets in the Gulf? If they hardly overlap, the test checks interpolation in one set of economies while the model extrapolates in another. And why extend the conclusion to subsectors as a whole when four of ten were tested, and two of the untested ones don’t use TEF anyway?
Why 2020? It was a pandemic year, and everything is scaled from it. Do the trajectories change with 2019, or an average of a few years, as the reference? If not, say so. If they do, the reference year is a real uncertainty and should be treated as one.
The uncertainty in the inputs is far wider than anything in the figures. Table 3 has intensity ranges spanning a factor of three or more, Table 4 gives WOR as 0 to 20+, and single values carry through regardless. What produces the “Predicted Range” in Figure 3 and the “median ± standard deviation” in Figure 9? Parameters, input datasets, something else? What I would want to know is whether the country rankings and the trends survive the ranges you report yourselves.
The Introduction faults aggregated models for masking “water policy, such as water tariffs,” and promises a model that accounts for “urban form, housing structure.” Eqs. (5) to (8) contain GDP per capita, household size, connection rate, population, NRW and an urban/rural factor. Where are housing and urban form? Why make tariffs the headline omission of other models and then leave them out? And what in this driver set could produce the declines in Bahrain and the UAE seen in Figure 9a? Either the model needs those variables or the Introduction needs to promise less.
On the sigmoid in Figure 2: five countries, post-2010 data only, bounds fixed at 50 and 500, and the upper end appears to rest largely on Qatar. What is the fitted equation, how many points, how good is the fit, and were the years from one country treated as independent? The reason given for dropping earlier data is that it would “obscure the GDP–DWW relationship.” Why is that a reason to exclude it rather than a sign that GDP alone doesn’t explain domestic use? And which countries are actually estimated from this curve?
The text says the model “captures” the impact of the 1991 Kuwait War. Figure 9a has almost nothing for Kuwait, and Kuwait is absent from Figure 8a. Captured against what? If there is an observation that shows it, show it. Otherwise this is the GDP input doing what the GDP input does.
“High-resolution” needs pinning down. The 90 m domestic fields are Eq. (7), per capita withdrawal times gridded population; industrial use is allocated to plants or spread across industrial areas by their size. So what in the model is actually known at 3 arcseconds? “Allocated to a 3-arcsecond grid” would describe it accurately. Grid spacing, information resolution and validation scale are three different things here, and the paper treats them as one.
Several results restate the model. “Fluctuations in withdrawals closely follow economic growth and changes in household size” (p. 30) is Eq. (5) in words. Elsewhere, changes are credited to diversification policy, modernization and infrastructure development, none of which the model contains. Why not use the structure you already have and split the industrial change into production growth, composition shifts and intensity change? That would be a result nobody could get by reading the equations, and I suspect it would be the best thing in the paper.
There are very few numbers. The figures carry the trajectories, but I could find almost no national totals or sector shares stated in the text or in a table, and industrial consumption gets far less attention than withdrawal. One compact table with representative-year totals, sector shares and the fresh/saline split would make the reconstruction much easier to judge and to reuse.
The food and beverage intensity is given as “1 – 17 m³/$ value added.” Taken literally, even the low end gives absurd withdrawals for a modest sector. Is a scaling factor missing, or is the denominator in thousands of dollars? And why electricity per capita as the predictor for the countries without data? How well does that relationship actually work?
Table 3 should be the single source of truth, and right now it isn’t. The refining text adopts a value that does not sit inside the range shown for the crude class it names, and the basic-metals section gives a mean and a median without saying which is used. Could one table list, per subsector, the value used, its unit, the period it represents, the source and where it applies? Given these mismatches, I would also want confirmation that the archived code reproduces the figures from the equations as written.
Where is the novelty relative to the nearest work? Table 1 itself lists Fujimori et al. (2017) with ten industrial subsectors, and two high-resolution industrial datasets for China. What does HyDRA do that they don’t? Is it the pairing of industrial detail with domestic infrastructure variables, the regional harmonization, the spatial allocation? Say it in the Introduction rather than leaving the reader to work it out from the table.
Smaller points
Figure 6 is never referred to in the text; why include it? The Figure 9 caption says “industrial” for a domestic figure, and the y-axes of Figures 5 and 6 say m³ where the values must be millions. Equation numbers repeat. Figure 7 has a placeholder date in the caption and no key for circle size; can the Google imagery be reused under the journal’s license? Page 8 says “water use” will stand for both withdrawal and consumption, a strange choice when telling them apart is the point. How sensitive are the industrial totals to the 5 percent “other manufacturing” term? What did the Google Earth Engine “verification” of cooling types consist of? The Introduction ends with a list of advantages rather than a question, the Discussion’s limitations mention none of the issues above, and there is no Conclusions section. And why “Novel” in the title?