The charts on this page use synthetic, illustrative curves to demonstrate the method — not measured results from a real catchment. See our validation page for what is currently benchmarked against independent networks.
How much electricity is in the snow? A dam is fed by its catchment — and the catchment is visible from orbit.
A hydropower operator knows the reservoir well enough. What it doesn't know is the future inflow — how much water will melt out of the snowpack, how much evaporation will take, how much rain will add on top. We observe the whole catchment from orbit and turn that into an inflow forecast weeks ahead. This page is a method study on a representative mountain catchment, and the plotted curves are illustrative.
BRIEFINGAS OF 2026-06-30WEEKLY CADENCEANALYSIS DESK: ORBISPECTv1.0
READ ACROSS THE WHOLE CATCHMENT FROM ORBIT, THE INFLOW SIGNAL LEADS THE RIVER GAUGE BY 2–6 WEEKS — THE WINDOW IN WHICH TURBINES AND RESERVE LEVEL CAN STILL BE PLANNED.
The water that drives the turbine is observable before it arrives — unevenly. Catchment precipitation, evaporation and soil moisture read strongly from orbit. Deep-snow water content is the harder case: retrieval degrades above roughly 150 mm SWE, which is precisely the regime that matters most for a mountain reservoir, so it is fused with the other layers rather than trusted alone. MIXED — STRONG ON PRECIP/ET, LIMITED ON DEEP SNOW
The output is a banded forecast, not a single number. Every inflow estimate carries a P10–P90 uncertainty band; the spread is the product, mapping onto turbine scheduling, flood-reserve level and contracted-energy value. HIGH ON METHOD
Lead-time is conditional, and the conditions are known. The 2–6 week horizon is strongest where inflow is snowmelt-driven and the catchment is well observed. It shortens for flashy rain-driven events, and where persistent cloud thins the optical record. Radar carries the signal through cloud, but it does not close every gap. MODERATE
Outlook — the band narrows as the melt season is observed out. If the snowmelt regime holds, the inflow forecast band should tighten through the spring melt as more passes accumulate over the catchment; a flashy rain-driven event would widen it. The narrowing is checkable against national river gauges. FORECAST · MODERATE
Two independent layers converge here: snow and surface-water layers from passive-microwave, optical, radar and altimetry sensing, and catchment precipitation and evapotranspiration estimates. The inflow judgment sits where they agree — neither layer alone would carry it.
RESERVOIR CATCHMENT
whole basinthe full mountain catchment upstream of the dam — every pixel observed from orbit every few days
WHAT WE WEIGH
3 streamssnowpack and its water content · catchment precipitation and evaporation · soil moisture — from satellite constellations
HORIZON
2–6 weeksinflow forecast with an uncertainty band — enough to plan turbine operation and reserve level
The snow store becomes the inflow wave — visible before the gauge
SNOW (SWE)INFLOW
ILLUSTRATIVE CURVE (METHOD DEMO): SPRING MELT TURNS THE SNOW STORE INTO AN INFLOW WAVE — VISIBLE FROM ORBIT BEFORE A RIVER GAUGE REGISTERS IT.
Inflow forecast — P10–P90 band
P50P10–P90OBSERVED
ILLUSTRATIVE CURVE (METHOD DEMO): THE UNCERTAINTY BAND NARROWS AS THE CATCHMENT IS WELL OBSERVED — THE OPERATOR GETS SCENARIOS, NOT A SINGLE NUMBER.
How it works — without opening the engine
1. The reservoir is a battery; the snow is its charger. Through winter the catchment stores water in its snowpack, and from orbit we measure how much water actually sits in that snow, not just where it lies. So we get a read on how much energy is waiting in the mountains before any of it runs off. This is also where the method is honestly weakest. Passive-microwave retrieval underestimates once the snowpack passes roughly 150 mm of water content, because the physics it depends on inverts: a deep snowpack stops scattering the microwave signal and starts radiating its own, and dense forest canopy can add as much again to the uncertainty. Snowmelt supplies over 30% of annual generation in snow-fed systems. That's a big enough share that the limitation is worth naming plainly rather than smoothing over, which is exactly why the snow signal gets fused with precipitation, evapotranspiration and ground-truth gauges instead of being trusted on its own. Published validation against airborne and ground reference puts passive-microwave SWE retrieval at roughly 10–30% mean absolute percentage error even outside the deep-snow regime — which is also why NOAA's own operational system for this problem, the Snow Data Assimilation System (SNODAS) run by the National Operational Hydrologic Remote Sensing Center, never relies on the satellite signal alone: it blends passive-microwave and optical snow observations with airborne surveys and a physical snow model before the estimate reaches a forecaster.
2. A catchment water balance, not a backward-looking reading. A river gauge tells you what already happened. We estimate what's still coming — precipitation minus evaporation minus infiltration, worked out for each part of the catchment separately, every few days, even through cloud, where radar carries the signal.
3. The operator gets a decision, not data. An inflow forecast with a P10–P90 band maps fairly directly onto the turbine operating plan, the flood-reserve level, and the value of energy under contract. Method details get shared with clients, and results are validated publicly against national river gauges.
Schedule generation and reserve drawdown weeks ahead instead of reacting to today's gauge reading.
GRID / ENERGY TRADER
Price the water
Translate expected inflow into expected output, and hedge or contract energy with the uncertainty band attached.
FLOOD-RESERVE MANAGER
Hold the right margin
See the melt wave forming in the catchment early, so reservoir headroom is set before the water arrives.
REGULATOR
Audit the basis
Independent, catchment-wide observation that can be checked against national river gauges.
Method & limits
THE SIGNAL
Water stored in the snowpack, catchment precipitation and evaporation, and soil moisture — observed across the whole contributing area and combined into an inflow estimate, not a single sensor reading.
UPDATE CADENCE
Refreshed every few days as new passes come in; optical signals depend on clear sky, while radar continues through cloud.
WHERE LEAD-TIME HOLDS
The 2–6 week horizon is strongest when inflow is snowmelt-driven and the catchment is well observed; it shortens for flashy rain-driven events and where persistent cloud thins the optical record.
FOR CITATION
The charger, not the battery. The forecast reads the snowpack's water content and the catchment water balance, so inflow gets estimated weeks before it reaches the reservoir, well ahead of when a gauge would register it.
Scenarios, not a single number. Every inflow figure carries a P10–P90 uncertainty band. It narrows as the catchment is well observed and widens for rain-driven events or persistent cloud.
Checked against the ground. Method details are shared with clients, and results are validated publicly against independent national river gauges, scored with the standard hydrological efficiency metrics: NSE and KGE, not a bespoke number. So a hydrologist can judge the result against the same yardstick used across the field.
Inputs: passive-microwave and optical snow layers · satellite precipitation and evapotranspiration estimates · radar and altimetry reservoir surfaces · terrain models. Reference: national hydrological gauges. Illustrative charts — method study, not a publication of measurements.
The operator briefing carries the live inflow forecast for your catchment with its P10–P90 band, the melt-timing and low-flow signals, and the validation against your reference gauges. It's refreshed as new passes come in, with the underlying figures and their stated uncertainty.