Lake and reservoir water-surface temperature and thermal stratification

Thermal infrared radiometry from Landsat TIRS, MODIS and ECOSTRESS retrieves skin-layer water temperature to within roughly 0.5 K under clear skies, revealing stratification onset, upwelling and cold dam releases. Cloud cover and a 100 m resolution floor are the binding constraints.

Sensors

Landsat 8/9 TIRS: Two thermal bands centred at 10.9 µm and 12.0 µm; 100 m native resolution resampled to 30 m in the data product (the 100 m figure is the honest spatial floor). 16-day single-satellite revisit, 8-day with both satellites combined. Noise-equivalent temperature difference below 0.4 K. Best suited to lakes wider than roughly 300 m.
MODIS Terra and Aqua: Bands 31 and 32 (11 and 12 µm) at 1 km resolution; combined Terra/Aqua gives up to four overpasses per day at mid-latitudes. Coarse resolution excludes reservoirs smaller than a few square kilometres but the daily revisit captures diurnal heating cycles and rapid upwelling events that Landsat misses entirely.
ECOSTRESS (ISS-mounted): Five thermal bands from 8.3 to 12.5 µm at 70 m resolution. Non-sun-synchronous orbit from the International Space Station produces variable overpass times, which is genuinely useful for diurnal temperature studies. Revisit is irregular, typically 3 to 5 days at mid-latitudes. Land-surface temperature accuracy documented at better than 1.5 K in validation studies.
Sentinel-3 SLSTR: Dual-view thermal channels at 11 and 12 µm, 1 km resolution, with a dual-satellite revisit of roughly 1 to 2 days globally. Designed for sea-surface temperature but validated for large inland water bodies. The dual-view geometry allows aerosol correction that single-view sensors cannot match.

What the skin layer actually tells you

Thermal infrared sensors measure radiance from the top 10 to 100 micrometres of the water column, the so-called skin layer. That is not the bulk mixed-layer temperature a field thermistor records a metre below the surface, and the two can differ by 0.2 to 0.5 K even under calm conditions. The skin is cooler than the bulk at night due to evaporative and radiative loss, and warmer in the early afternoon when solar heating outruns mixing. Buyers who need bulk temperature for ecological or engineering models must apply a skin-to-bulk correction, which is well documented in the literature but adds uncertainty.

Despite that caveat, the skin signal is far from trivial. Stratification onset is visible as a rapid warm anomaly in the epilimnion relative to deeper-water inflows. Cold upwelling events, driven by wind stress reversals or drawdown operations, produce temperature contrasts of 2 to 8 K against the surrounding surface, well above any sensor's noise floor. Dam releases from deep hypolimnetic outlets show as persistent cold plumes extending downstream, detectable in Landsat TIRS imagery as long as the plume is wider than roughly 200 to 300 m.

The cloud problem is worse than most buyers expect

Landsat's 8-day revisit (both satellites combined) sounds adequate until you apply a cloud mask. In humid tropical and temperate climates, cloud-free acquisitions over a given lake can drop to four or five per year. MODIS partially compensates with its near-daily revisit, but at 1 km resolution it cannot resolve the spatial structure of stratification in reservoirs smaller than roughly 10 km across.

ECOSTRESS partially bridges the gap: 70 m pixels and variable overpass times give it a different cloud-collision probability than sun-synchronous sensors, so a composite of Landsat, MODIS and ECOSTRESS scenes often recovers more usable observations than any single source. Even so, a month-long cloud event over a monsoon-season reservoir will produce a genuine data gap. Honest gap-filling requires either interpolation with physical priors (degree-day models, heat-budget equations) or assimilation of reanalysis air-temperature fields, each of which introduces its own uncertainty. There is no satellite-only solution that eliminates cloud contamination.

Resolution floors and the narrow-reservoir problem

Landsat TIRS has a point-spread function that spreads energy across roughly 100 m on the ground, even though the archived product is gridded at 30 m. A pixel that straddles the water-land boundary is a mixture of both, and mixed pixels read warmer than open water in daytime and cooler at night. For reservoirs narrower than roughly 200 to 300 m, the majority of pixels are contaminated. MODIS is worse: a 1 km pixel over a narrow reservoir is mostly land.

ECOSTRESS at 70 m is the best currently operational option for narrow water bodies, but its irregular revisit makes time-series analysis difficult. Airborne thermal sensors can reach sub-metre resolution and are the correct tool for narrow irrigation reservoirs or canal reaches, but they are not a satellite product. Buyers with narrow-reservoir monitoring needs should be told this clearly at the outset.

Detecting stratification onset and turnover

Seasonal stratification in a temperate lake typically begins when the surface warms past 4 °C and wind energy is insufficient to mix the water column. In Landsat time series, this appears as a progressive warming of the surface relative to inflow temperatures, often accompanied by spatial gradients: shallow bays warm faster than the main basin, and wind-sheltered areas warm before exposed ones. The spatial pattern is itself diagnostic, distinguishing wind-driven from thermally driven stratification.

Autumn turnover, when the surface cools to match hypolimnion temperature and convective mixing homogenises the column, appears as a rapid convergence of surface temperatures across the basin. The timing of both events has ecological consequences: stratification onset controls the depth of the photic zone and nutrient availability; turnover timing affects dissolved oxygen replenishment. Multi-year Landsat archives going back to 1982 (Landsat 4 and 5 TM) allow detection of long-term shifts in stratification phenology, though inter-sensor calibration across the archive requires care.

Operational limits and what the numbers actually mean

Published validation of Landsat TIRS Lake surface temperature retrievals typically reports root-mean-square errors of 0.5 to 1.0 K against in-situ buoys under clear-sky conditions. MODIS LST products over inland water report similar figures in low-aerosol conditions, degrading to 1.5 to 2 K in dusty or humid atmospheres. ECOSTRESS validation over water is less extensive in the published record, with reported accuracies in the 1 to 1.5 K range.

These figures assume accurate atmospheric correction using either a radiative transfer code (MODTRAN-based) or a split-window algorithm applied to dual thermal bands. Single-band sensors cannot apply split-window correction and are more sensitive to water-vapour uncertainty. For operational reservoir management where a 0.5 K error matters, in-situ validation with at least one thermistor buoy is not optional. Satellite thermal data without any ground truth is a monitoring tool, not a measurement standard.

Turning temperature maps into decisions

The most direct operational application is selective withdrawal management at dams. Knowing the depth of the thermocline from surface temperature patterns, combined with a hydrodynamic model, allows operators to choose which outlet level to open in order to meet downstream temperature standards. Several water utilities in Europe and Australia have integrated satellite-derived surface temperature into their reservoir management dashboards alongside in-situ profilers.

For water-quality monitoring, surface temperature is a first-order predictor of cyanobacterial bloom risk: most bloom-forming species prefer temperatures above 20 to 25 °C combined with stable stratification. A thermal anomaly alert, issued within 24 to 48 hours of a cloud-free overpass, gives water treatment operators a lead time that manual sampling cannot match. Satellize runs analytics on open constellations including Landsat and MODIS for exactly this kind of operational alert layer, combining multi-source thermal compositing with atmospheric correction workflows calibrated to client lake conditions. The honest caveat remains: if cloud persists for three weeks, no satellite thermal product will fill that gap without modelled priors.

Typical figures

Best available spatial resolution (thermal)	70 m (ECOSTRESS); 100 m effective (Landsat 8/9 TIRS); 1 km (MODIS, Sentinel-3 SLSTR)
Revisit frequency	Near-daily (MODIS Terra + Aqua combined); 1–2 days (Sentinel-3 dual satellite); irregular 3–5 days (ECOSTRESS); 8 days (Landsat 8 + 9 combined)
Temperature retrieval accuracy (clear sky)	0.5–1.0 K RMSE for Landsat TIRS; 0.5–1.5 K for MODIS; 1.0–1.5 K for ECOSTRESS (published validation ranges)
Spectral bands used	10.9 µm and 12.0 µm (Landsat TIRS); 11 µm and 12 µm (MODIS bands 31–32, Sentinel-3 SLSTR); 8.3–12.5 µm multi-band (ECOSTRESS)
Minimum detectable water-body width	~200–300 m for Landsat TIRS; ~2–3 km for MODIS; ~150–200 m for ECOSTRESS (due to mixed-pixel contamination)
Cloud contamination impact	Can reduce usable Landsat acquisitions to 4–5 per year in humid climates; MODIS daily revisit partially compensates at coarser resolution
Archive depth	Landsat thermal back to 1984 (TM band 6); MODIS from 2000; ECOSTRESS from 2018; Sentinel-3 SLSTR from 2016
Latency (open data products)	MODIS LST products available within ~6 hours of overpass; Landsat Collection 2 Level-2 within ~12–24 hours; ECOSTRESS L2 within ~12 hours
Delivery formats (processed analytics)	GeoTIFF temperature maps, NetCDF time-series stacks, CSV anomaly tables, alert feeds (JSON or email)

Analytics Satellize can run

Cloud-composited lake surface temperature map	Multi-source thermal fusion (Landsat + MODIS + ECOSTRESS), cloud masking, split-window atmospheric correction	GeoTIFF temperature map per overpass, with cloud-gap fraction flagged per pixel
Stratification onset and turnover date series	Threshold detection on surface temperature time series; spatial variance metrics across basin to identify mixing events	Annual phenology report with onset/turnover dates and inter-annual trend chart
Cold plume detection from dam releases	Anomaly mapping relative to ambient surface temperature; plume extent delineated by isotherm contouring	GIS polygon layer per event; temperature contrast and plume area statistics
Upwelling event alert	Near-real-time MODIS LST anomaly detection against rolling 30-day baseline; flagged when contrast exceeds 2 K over defined area	Alert notification within 24 hours of qualifying overpass, with supporting temperature map
Long-term stratification trend analysis	Landsat archive time series (1984 onward) with inter-sensor calibration; Mann-Kendall trend test on onset-date series	Decadal trend report with confidence intervals; suitable for climate-impact assessments
Bloom-risk thermal precursor layer	Surface temperature threshold exceedance (configurable, typically 20–25 °C) combined with stratification stability index derived from temperature spatial variance	Weekly risk-score raster; integrates with separate cyanobacteria optical detection workflow

Sources

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.