Watching a River Basin from Orbit

A river gauge is a superb instrument with a narrow question. It tells you how much water passed one cross-section of one channel, at one moment. Multiply that by the handful of functioning gauges in a large West African basin and you have a few dozen numbers a day describing a landscape of tens of thousands of square kilometres, most of which contains no river at all.

My MSc research at the University of Energy and Natural Resources, completed in December 2022 under a World Bank Africa Centre of Excellence scholarship, asked whether earth observation could carry water budget monitoring in exactly that situation. The basin was the White Volta. The question has followed me into every hydrological project since.

What a water budget actually is

Strip away the software and a water budget is an accounting identity. Over some area and some period, precipitation comes in; evapotranspiration, discharge and change in storage account for where it went. Everything else — every model, every satellite product, every calibration argument — exists to put defensible numbers on those terms.

The identity is trivially true, which is precisely what makes it useful. If you can measure three terms well, the fourth is constrained. If you measure all four independently, the extent to which they fail to balance is a direct, quantitative statement about how much you do not know. That residual is not an embarrassment to be hidden. It is the most honest diagnostic in the discipline.

Why gauges alone cannot close it here

In a basin like the White Volta, the ground network struggles with the budget for reasons that are structural rather than fixable by maintenance.

Gauges measure discharge, and discharge is usually not the dominant outgoing term. In a semi-arid, agriculturally intensive basin, evapotranspiration is where most of the rainfall goes — and no gauge measures it. Point discharge tells you almost nothing about the term that dominates the balance.

The stations that do exist are sparse and unevenly distributed, and the records have gaps. Rating curves drift as channels scour and deposit, so the relationship between measured stage and reported discharge quietly decays between calibrations. The many small reservoirs and dugouts that hold and evaporate water across the landscape are largely unmetered, as is a great deal of abstraction.

And the basin is transboundary. Substantial upstream area lies outside the jurisdiction of any Ghanaian agency, which means that even a perfect Ghanaian gauge network would be measuring the downstream consequences of an unobserved upstream.

Every error in every term ends up in the closure residual. With enough unmeasured terms, the residual stops being a diagnostic and becomes a shrug.

What earth observation changes, and what it does not

Satellite observation replaces points with fields. Instead of a value at a station, you get a spatially complete grid: precipitation estimates, land surface temperature, vegetation indices, soil moisture, land cover, surface water extent — everywhere in the basin, on a repeating cycle, on both sides of an international border.

But an observation is not a flux. A vegetation index is not evapotranspiration; a rainfall estimate is not runoff. Earth observation gives you spatially complete state, and the water budget needs fluxes. Bridging that is what the model chain is for.

VegET: a water balance in every pixel

VegET is a satellite-driven daily soil-water-balance model. For each grid cell it takes precipitation, a reference evapotranspiration field and a vegetation-index-derived coefficient that scales reference ET to what the actual vegetation there is plausibly transpiring, together with the soil’s water-holding capacity. It then does the bookkeeping day by day: water in, water held, water lost to the atmosphere, water shed as runoff.

The output is the thing that has always been missing from the gauge-only picture — a spatially distributed estimate of actual evapotranspiration and runoff generation across the whole basin, including the parts nobody visits. It is also computationally light enough to run over very large domains, which turns out to matter enormously.

mizuRoute: turning a map into a hydrograph

VegET tells you how much water each cell shed. It does not tell you when that water arrives somewhere downstream, and it has no concept of a river network at all.

mizuRoute supplies both. It maps gridded runoff onto the catchments of an explicit river network, then routes that water downstream through the network topology, applying the delay and attenuation that a real channel imposes. Given a hydrography dataset defining the reaches and their connectivity, it converts a map of runoff generation into a simulated hydrograph at any reach you care to ask about.

That conversion is what makes the whole chain testable. You cannot validate a continental evapotranspiration field directly. But you can route it to the few locations where a real gauge exists and compare simulated discharge against measured discharge — and because that number is the integrated consequence of everything upstream in the chain, agreement is a meaningful test of the whole thing.

The scarce gauges become more valuable once you have a model to test against them, not less. They stop being the entire monitoring system and become its calibration and evidence base.

Doing it at continental scale

Since May 2025 I have worked as a research consultant with the International Water Management Institute’s West Africa office, and the basin-scale method has scaled up considerably.

On DIWASA V2, the World Bank-funded continental hydrology and discharge modelling programme, I supported MERIT-based VegET and mizuRoute baseline runs covering 1992 to 2024 — a record long enough that a trend statement carries weight rather than reflecting one wet decade. On top of that baseline sit validation and future-scenario discharge simulations to 2050, run through the same modelling chain so that historical and projected behaviour are produced by one consistent method rather than stitched together from two incompatible ones. My analytical contribution was the discharge trend analysis and visualisation, and the basin-scale summaries and figures that make a continental simulation legible to a basin authority.

The companion workstream, the Google-funded Lerma–Santiago study in Mexico, adds the piece that heavily managed basins demand. There I ran mizuRoute for naturalised discharge and mizuLake for reservoir-regulated discharge, so that the gap between what the river would do and what the river is permitted to do becomes an explicit, quantified object rather than an assumption buried in the calibration. The results were reduced to environmental-flow indicators for an e-flow dashboard.

What changes for management

Ungauged sub-basins acquire numbers. Allocation arguments that previously ran on competing anecdote now run against a shared budget with a stated uncertainty.

Change becomes attributable. When flows fall, a distributed budget can begin to separate declining rainfall from rising upstream consumption — a distinction that determines whether the appropriate response is adaptation or negotiation.

Transboundary discussion gains a neutral dataset. Satellite observation does not stop at a border and is not owned by either party, which makes it unusually well suited to conversations where each side distrusts the other’s figures.

And monitoring becomes continuous rather than campaign-based. A basin authority can maintain a budget without first funding a dense observation network it has no prospect of sustaining.

The honest limits

Products disagree. Different precipitation and evapotranspiration datasets will give you materially different budgets for the same basin, which is why multi-product comparison and deviation analysis are part of the method rather than a caveat at the end — the spread between products belongs in the result.

None of this removes the need for ground data. It changes what ground data is for: from the sole source of truth to the validation anchor for something far more spatially complete than it could ever be alone.

The White Volta work is now a manuscript under review with colleagues at UENR and IWMI. I have since facilitated Water Accounting Plus and VegET workshops in Ghana and Ethiopia, because the constraint on this method in most African basins is no longer the satellites. It is the number of people trained to run the chain and defend its numbers in a room full of stakeholders.