Upstream land use silently reshapes who gets the world’s shared river water

Every international water treaty on Earth shares a hidden assumption: that the amount of water crossing a national boundary is set by climate and geography, not by what an upstream country decides to do with its land. A new global study demolishes that assumption with satellite data. Researchers at Michigan State University have produced the first worldwide, spatially explicit map of what they call silent water redistribution — the unrecognized transfer of water availability across sovereign borders caused not by dams or diversions, but by upstream changes in land use and land cover. When a headwater nation expands irrigated farmland or plants new forests, more precipitation evaporates from the landscape and less runs off into the river. Downstream co-riparians lose water without a single physical structure touching the flow, and, according to the new analysis, without any legal provision recognizing the loss.

The study, published in Frontiers in Forests and Global Change, quantifies the scale of this governance blind spot at treaty-comparable resolution for the first time. Transboundary river basins encompass roughly 40 percent of global freshwater flow and sustain more than three billion people across over 150 nations. Yet the more than 800 international freshwater agreements cataloged in the Transboundary Freshwater Dispute Database regulate only extractive, infrastructure-based water use: dams, diversions, and pumping that physically remove water. An entire class of transboundary water transfer — one driven by upstream territorial decisions — has never been codified. The new research converts that qualitative concern into hard numbers, finding that upstream land-use change silently removes 31.66 cubic kilometers of water per year from the renewable water budgets of downstream territories worldwide.

To achieve this, the team built a basin-country-unit framework spanning 313 transboundary river basins, intersected with national boundaries to yield 796 units across 153 countries. Each unit was classified by hydrological position — upstream, midstream, or downstream — using flow accumulation data from the HydroSHEDS digital elevation product, with thresholds normalized per basin so that headwater nations in small catchments were ranked against their own basin’s gradient rather than a global reference. Across 413 source units, the researchers detected five land-use transition types between 2001 and 2023 using the MODIS Land Cover Type product at 500-meter annual resolution: deforestation, rainfed agricultural expansion, urbanization, reforestation, and irrigated agricultural expansion. Rigorous noise suppression — a five-year temporal majority filter, early- and late-epoch composites, and a three-year persistence rule — filtered out false positives, including a striking 70.4 percent of initially detected reforestation that consisted of pixels merely cycling between spectrally similar woody classes.

The water consequences of each transition were quantified through evapotranspiration attribution. The team used the gap-filled MODIS evapotranspiration product with a space-for-time substitution: the mean recent evapotranspiration of pixels that underwent a transition, minus that of class-specific control pixels within the same unit that remained in the original land cover. This methodological choice proved decisive. Using global controls instead produced a sign error for deforestation — a spurious gain of 90.8 millimeters per year rather than the physically required loss of 30.3 millimeters per year. To separate land-use signals from concurrent climate variability, the researchers applied a Budyko elasticity decomposition, deriving analytical sensitivities of evapotranspiration to precipitation and potential evapotranspiration from the Choudhury-Yang parametric Budyko equation, and computing climate inputs from CHIRPS satellite precipitation and ERA5 reanalysis temperatures. The climate correction retained 84 percent of the uncorrected net signal as land-use-attributed, and for irrigation the climate fraction was a mere 0.02 — confirming that irrigation-driven redistribution is an almost entirely anthropogenic signal.

The results reveal a consumptive signal dominated by two transitions that most people would never suspect of moving water between countries. Irrigated agricultural expansion, covering 69,336 square kilometers — less than a quarter of the 233,157 square kilometers of rainfed expansion — contributes the largest single term, minus 36.37 cubic kilometers per year, because supplemental water application raises evapotranspiration by 200 to 650 millimeters per year above rainfed baselines depending on climate zone. Reforestation ranks second at minus 23.76 cubic kilometers per year, as growing trees transpire more and intercept more rainfall than the vegetation they replace. Deforestation partially offsets the balance by releasing plus 25.65 cubic kilometers per year, with rainfed expansion and urbanization contributing small positive terms. The net global signal of minus 31.66 cubic kilometers per year proved insensitive to the screening rules: substituting observed MODIS contrasts for literature coefficients reduced the deforestation surplus and deepened the reforestation loss, and sensitivity analyses shifting the Budyko parameter between 1.5 and 2.5 bounded the total between minus 27.86 and minus 35.46 cubic kilometers per year.

Routing these signals downstream through the Transboundary Freshwater Dispute Database basin topology yielded 683 unique upstream-to-downstream pairs and a Silent Water Redistribution Index for each of 323 downstream units, expressed as a percentage of the receiving country’s total actual renewable water resources from FAO AQUASTAT. The index ranges from minus 12.89 to plus 5.02 percent, with 189 downstream units experiencing a net deficit and 134 a net surplus. Governance urgency concentrates in the tail: 15 units in 13 basins exceed the 1 percent significance threshold. The Jordan basin produces the two most extreme values — Israel faces a silent redistribution of minus 12.89 percent and the Palestinian Territories minus 3.10 percent, driven entirely by irrigation expansion in the Jordanian and Syrian headwaters, with their tiny renewable endowments of 0.75 and 0.80 cubic kilometers per year amplifying the relative signal to levels comparable to major allocation shares negotiated in bilateral treaties. The Dniester basin ranks second most severe, where Moldova loses minus 12.25 percent not to irrigation but to upstream Carpathian afforestation — a country whose entire national water endowment is 1.3 cubic kilometers per year.

The La Plata system carries the world’s largest absolute upstream deficit, minus 19.97 cubic kilometers per year, driven by Brazil’s 19,449 square kilometers of new irrigated cropland in the headwaters. Yet downstream Argentina and Paraguay register only moderate relative values of minus 1.64 and minus 1.97 percent because their large renewable water denominators absorb the signal — a demonstration that governance urgency is determined by the signal-to-endowment ratio, not absolute volume alone. Two basins show positive values: the Gambia at plus 5.02 percent and the Mekong in Laos at plus 2.28 percent. These surpluses, the authors stress, are not governance successes but transient liabilities whose eventual reversal, when upstream land use changes again, carries no treaty mechanism to anticipate or manage.

The study’s central governance finding emerges from a systematic cross-referencing of every significant basin against the full treaty corpus — 3,812 treaty-basin-country records drawn from more than 800 internationally recognized freshwater agreements spanning 1820 to 2023, each encoding up to 156 issue-area variables. The verdict: no agreement in force contains a provision that explicitly governs upstream land-use-driven evapotranspiration change as a mechanism of transboundary water redistribution. Eleven of the 13 significant basins are classified as Governance Gap, where a treaty exists but is wholly silent on the mechanism, and two — the Medjerda and Rann of Kutch — as No Governance, where no agreement of any kind applies. Together these basins account for a redistributed flux of minus 24.85 cubic kilometers per year, equivalent to 78.5 percent of the total flux across the significant basins. An estimated 70.3 million people reside within downstream territories subject to hydrologic redistribution exceeding the 1 percent threshold, lacking any formal, legally binding mechanism to recognize, negotiate, or mitigate these systemic shifts.

Perhaps the most uncomfortable finding is what the treaty chronology forecloses: the intuitive excuse of historical ignorance. Six of the eleven Governance Gap basins are regulated by instruments ratified after 2000, including the technically sophisticated 2018 Mekong River Commission agreement and the 2015 Tigris-Euphrates framework. Their mean treaty age is just 15.7 years, far younger than the global dataset mean of 31.2 years. Contemporary transboundary water diplomacy, in other words, continues to treat water availability at international boundaries as a function of climate and geography rather than as jointly determined by upstream land decisions — a conceptual architecture embedded in the 1997 UN Watercourses Convention, which defines watercourse use in terms of physical extraction rather than evapotranspiration change. This creates the reforestation paradox: climate finance under the Paris Agreement’s Article 6 and REDD+ mechanisms actively incentivizes upstream afforestation that reduces downstream water availability, with no requirement to assess downstream hydrological consequences — a direct, currently unresolved collision between SDG 13 and SDG 6, quantified here for Moldova, Chad, and Namibia, which lose 12.25, 1.56, and 1.12 percent respectively to upstream re-greening.

Independent physical verification came from the GRACE and GRACE-FO satellite gravity missions, which measure terrestrial water storage change directly. Across the 13 significant basins, the sign of the storage change agreed with the SWRI prediction in 8 of 13 cases — 62 percent — with the five mismatches individually explicable by documented confounders such as the post-1970s precipitation increase in the Rio de la Plata watershed and the post-1980s Sahelian rainfall recovery. Across all 57 affected downstream units, the continuous association was statistically significant, and the global mean storage anomaly across all 323 downstream units was minus 7.60 centimeters, confirming a broad pattern of water storage decline in the downstream territories of the world’s transboundary basins. The authors propose three incremental reforms that require no new institutions: incorporating SWRI monitoring into existing basin-organization reporting as an early-warning system, extending Espoo Convention prior-notification principles to upstream land-use transitions such as irrigated expansion exceeding agreed thresholds, and conditioning climate finance disbursements for afforestation on downstream water impact assessments. Land use, the study concludes, is no longer merely food, urban, industrial, and ecological policy — it is also water management, and the institutional architecture of transboundary water governance must now reflect that reality.

Subject of Research: Upstream land-use change and transboundary river water redistribution: a global mapping

Article Title: Upstream land-use change and transboundary river water redistribution: a global mapping

Article References: Khajavigodellou, Y., & Qi, J. (2026). Upstream land-use change and transboundary river water redistribution: a global mapping. Frontiers in Forests and Global Change, 9, Article 1951830. https://doi.org/10.3389/ffgc.2026.1951830

Image Credits: AI Generated

DOI: 10.3389/ffgc.2026.1951830

Keywords: Upstream, land-use, change, transboundary, river, water, redistribution, global, mapping, scientific research