Hydro-reservoir water-level and storage-volume monitoring
Radar altimeters and imaging spectrometers can track water-surface elevation and inundated area independently, then combine them into storage-volume estimates that dam operators rarely publish themselves. Accuracy hinges on bathymetry you almost certainly do not have.
Sensors
- Sentinel-6 Michael Freilich (Poseidon-4 radar altimeter): Ku- and C-band nadir altimeter with a 10-day exact-repeat cycle and along-track surface-height precision of roughly 2–3 cm over open ocean; over inland water bodies the figure degrades to 10–30 cm depending on target width and surrounding topography. Only measures elevation where the ground track crosses the reservoir.
- SWOT KaRIn (Ka-band Radar Interferometer): Launched December 2022, SWOT maps water-surface elevation across a 120 km swath (with a 20 km nadir gap) at roughly 50–100 m posting. Science-phase results suggest height accuracy of approximately 10 cm for water bodies larger than 1 km², improving on conventional altimeters for mid-sized reservoirs. Nominal revisit is 21 days at mid-latitudes.
- ICESat-2 ATLAS (photon-counting lidar): 532 nm green lidar with six beams arranged in three pairs. Over calm inland water, surface-height precision reaches 2–5 cm along track. Revisit at any given point is roughly 91 days, limiting its operational use to periodic calibration or cross-validation of radar altimeter virtual-station records.
- Sentinel-2 MSI: 13-band multispectral imager at 10–60 m spatial resolution, 5-day revisit (combined Sentinel-2A/2B). Near-infrared and shortwave-infrared bands discriminate open water from wet soil with high reliability in cloud-free conditions. Used to map inundated surface area, which is then paired with altimeter-derived elevation to estimate storage volume.
What a water-level record actually measures
Conventional nadir altimeters, including Sentinel-6 and its predecessors in the TOPEX/Poseidon lineage, fire a microwave pulse straight down and time the return. The instrument measures the distance from the satellite to the nearest reflecting surface. Over a reservoir, that means water-surface elevation relative to a geodetic datum, typically EGM2008 or a local geoid model. The precision is genuinely impressive for open-ocean work. Over a narrow reservoir, it is considerably less so.
The fundamental constraint is geometry. A nadir altimeter samples only the narrow strip directly beneath the satellite. If the ground track crosses your reservoir at an angle, or clips only one arm of a branching impoundment, the virtual station captures a single transect of the water surface, not a spatially averaged level. For large reservoirs such as Lake Nasser or the Three Gorges impoundment, this is manageable. For a 2 km² mountain reservoir, the track may miss the body entirely on most repeat cycles, or the returned waveform may be contaminated by surrounding terrain and vegetation. Published virtual-station databases, including the USDA G-REALM and the Copernicus Global Land Service records, document hundreds of reservoirs where this limitation is clearly visible in the scatter of individual observations.
SWOT changes the geometry, not the physics
The Surface Water and Ocean Topography mission addresses the along-track sampling problem by using interferometric SAR rather than a nadir pulse. Two antennas separated by a 10 m boom measure the phase difference of Ka-band returns across a 120 km swath, converting phase to height with a posted resolution near 50–100 m. This means a reservoir of 1 km² or larger can, in principle, be mapped in its entirety during a single overpass rather than sampled at a single crossing point.
The caveats are real. The 20 km nadir gap means a reservoir sitting directly beneath the satellite centreline is not observed. The 21-day revisit at mid-latitudes is coarser than Sentinel-2's optical cadence, so SWOT elevation and Sentinel-2 area observations will rarely be simultaneous. Wind roughening of the water surface degrades height retrievals. And Ka-band is attenuated by heavy rain, which is precisely when reservoir inflows are highest and monitoring is most urgent. SWOT is a step forward, not a complete solution.
From elevation to volume: the bathymetry problem
Storage volume is not directly observable from space. What satellites measure is water-surface elevation and inundated area. Volume is inferred by integrating the hypsometric curve, the mathematical relationship between water level and the area enclosed at each elevation. That curve depends entirely on the shape of the reservoir basin below the current waterline.
Bathymetric surveys are expensive, infrequent, and in many jurisdictions treated as commercially or strategically sensitive. Dam operators rarely publish them. Where no survey exists, analysts use one of two approximations: they reconstruct the hypsometric curve from historical satellite imagery spanning a wide range of water levels, tracing the shoreline at multiple known elevations; or they apply empirical power-law relationships between area and volume that are calibrated on surveyed reservoirs of similar morphology. Both approaches introduce uncertainty that compounds directly into volume estimates. A 10% error in the hypsometric curve translates to a 10% error in storage volume at any given level. For drought early warning, where the question is how many weeks of generation remain, that uncertainty matters.
Sedimentation makes the problem worse over time. A bathymetric survey from 2005 may no longer represent a reservoir that has accumulated significant sediment since impoundment. Satellite turbidity monitoring can track sediment dynamics at the surface, but inferring below-surface deposition from optical data alone is not currently feasible.
Combining sensors in practice
The most defensible operational approach pairs a radar altimeter virtual-station record (for elevation continuity and cloud immunity) with Sentinel-2 surface-area mapping (for spatial detail and shoreline geometry). ICESat-2 passes, when they coincide with the reservoir, provide high-precision spot checks that can recalibrate altimeter biases introduced by geoid model errors or waveform retracking artefacts.
Sentinel-2 surface-area mapping is straightforward in cloud-free conditions. The Modified Normalised Difference Water Index (MNDWI), computed from green and shortwave-infrared bands, reliably separates open water from surrounding land at 10–20 m resolution. The practical floor is around 0.5–1 ha for confident delineation; below that, mixed pixels at the shoreline dominate the area estimate. Cloud cover is the main operational constraint. Tropical reservoirs in the wet season, when storage is most dynamic, can go weeks without a usable optical acquisition. Sentinel-1 SAR backscatter can substitute for surface-area mapping in persistent cloud, though the shoreline precision is lower, particularly where the surrounding terrain is rough or vegetated.
What the numbers can and cannot support
A well-configured virtual-station record on a large reservoir can detect water-level changes of 20–30 cm between repeat cycles. On a reservoir with a hypsometric curve that translates 1 cm of level to 5 million cubic metres of storage, that detection threshold is operationally useful for weekly dispatch planning. On a smaller, steeper reservoir, the same 20 cm threshold may represent a much smaller absolute volume change, and the noise in the altimeter record may swamp the signal.
For drought early warning, the relevant question is usually not the current storage volume but the rate of drawdown relative to historical percentiles at the same point in the seasonal cycle. Satellite-derived time series going back to the early 1990s through the TOPEX/Poseidon, Jason-1, Jason-2, Jason-3, and Sentinel-6 lineage provide three decades of level records for larger reservoirs. That archive is long enough to compute meaningful anomalies. For smaller reservoirs that only became observable with SWOT, the archive is currently less than two years old.
Satellize runs storage-volume analytics on open altimetry and Sentinel-2 data, applying hypsometric reconstruction where operator bathymetry is unavailable. The Tonga crop-estimation programme demonstrated the same multi-sensor integration logic at national scale; the same pipeline applies to reservoir networks. Uncertainty bounds are always reported explicitly, because a volume estimate without an error bar is not an estimate, it is a guess.
Typical figures
| Altimeter water-level precision (large reservoir) | 10–30 cm (Sentinel-6 nadir); ~10 cm (SWOT KaRIn, bodies >1 km²) |
| Lidar water-level precision (ICESat-2) | 2–5 cm along track, calm water |
| Surface-area mapping resolution | 10 m (Sentinel-2 MNDWI); minimum reliable delineation ~0.5–1 ha |
| Altimeter repeat cycle | 10 days (Sentinel-6); 21 days (SWOT); 91 days (ICESat-2) |
| Optical revisit (cloud-free) | 5 days (Sentinel-2A+2B combined); cloud cover may extend effective gap to weeks in tropics |
| SWOT swath width | 120 km total, with 20 km nadir gap |
| Minimum reservoir size for SWOT height retrieval | ~1 km² (science-phase threshold; smaller bodies have degraded accuracy) |
| Archive depth (altimetry) | ~1992 to present (TOPEX/Poseidon lineage); SWOT from late 2022 |
| Volume uncertainty (no operator bathymetry) | Typically 10–25% depending on hypsometric reconstruction method |
| Delivery formats | Time-series CSV, GeoTIFF (area maps), GIS polygon layers, anomaly alert feed |
Analytics Satellize can run
| Reservoir water-level time series | Radar altimeter waveform retracking at virtual stations; ICESat-2 cross-calibration where tracks coincide | Monthly CSV with level anomaly relative to 30-year climatological percentile |
| Inundated surface-area maps | MNDWI thresholding on Sentinel-2 MSI; Sentinel-1 SAR backscatter fallback under persistent cloud | GeoTIFF water-mask per acquisition; area time series in tabular format |
| Storage-volume estimates | Hypsometric curve integration combining altimeter elevation and Sentinel-2 area; curve reconstructed from multi-year imagery archive where operator bathymetry is absent | Weekly volume estimate with explicit uncertainty bounds; anomaly flag when drawdown rate exceeds seasonal norm |
| Drought early-warning alert | Storage anomaly scoring against historical percentile distribution; threshold-triggered alert when volume falls below operator-defined percentile | Automated alert with current storage estimate, uncertainty range, and days-to-minimum-operating-level projection |
| Multi-reservoir network dashboard | Batch processing across defined reservoir portfolio; aggregated storage index weighted by installed capacity | GIS layer and tabular dashboard updated on each new satellite acquisition |
| Hypsometric curve reconstruction | Shoreline extraction at multiple known water levels from historical Sentinel-2 archive; area-elevation regression fitted to observed pairs | Reservoir-specific hypsometric table and uncertainty characterisation report |
Who does the work
We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.