Inland water-body screening for floating solar deployment

Satellite imagery can rank hundreds of reservoirs, irrigation ponds, and mine-water lakes for floating PV suitability in weeks. The analysis combines surface-area mapping, seasonal water-level change, and topographic shading to filter out sites that will not pencil out economically.

Sensors

Sentinel-2 MSI: 10 m resolution in visible and near-infrared bands; 5-day revisit at the equator with both satellites active. The NDWI (Green/NIR ratio) reliably delineates open water down to roughly one hectare. Multitemporal stacks reveal seasonal drawdown by tracking shoreline retreat across months.
Landsat 8/9 OLI: 30 m resolution, 16-day revisit per satellite (8-day combined). Archive extends to 1972 for Landsat 1, giving multi-decadal water-body history. Useful for detecting long-term trends in surface area that a shorter Sentinel-2 record would miss, particularly for reservoirs affected by interannual drought cycles.
SRTM DEM: Shuttle Radar Topography Mission digital elevation model at 30 m horizontal posting, vertical accuracy approximately 10 m (90th percentile globally). Slope derivatives identify steep surrounding terrain that casts shading on a water surface; horizon-angle calculations from the DEM flag bodies where topographic shading would materially reduce annual energy yield.
ICESat-2 ATL13: Inland water surface height product derived from the ATLAS photon-counting lidar, with along-track spacing of roughly 0.7 m and vertical precision of a few centimetres on calm water. Provides absolute water-surface elevation at repeat passes, allowing seasonal amplitude of water-level change to be estimated where the satellite ground track intersects a target body.

Why floating PV siting is harder than it looks from a map

A national reservoir inventory might list ten thousand water bodies. Most are disqualified before any field visit. Some are too small to support a commercially viable array. Others lose half their surface area between wet and dry seasons, creating mechanical stress on anchoring systems and reducing bankable generation hours. A few sit in valley bottoms so deep that surrounding ridges shade the water for three or four hours each winter morning. Satellite analysis can eliminate the obvious failures in bulk, cheaply, before a single site visit is commissioned.

The key variables are surface area (stable and seasonal minimum), water-level fluctuation amplitude, bathymetric adequacy for anchor systems, and topographic horizon angles. Of these, the first three are directly accessible from open satellite data. The fourth requires a DEM and some trigonometry. None requires commercial tasking; Sentinel-2 and Landsat cover the globe on open access, and SRTM is a freely distributed product from NASA and USGS.

What a floating roof gives away in multispectral imagery

Open water absorbs strongly in the near-infrared. Land and vegetation reflect it. The Normalised Difference Water Index, computed as (Green minus NIR) divided by (Green plus NIR), separates water from non-water with high reliability at Sentinel-2's 10 m resolution. Applied to a time series of cloud-free scenes across a full annual cycle, the same index maps the seasonal minimum extent of each water body. That minimum extent is the operative constraint for floating PV: the array must fit within the dry-season surface, and the mooring system must tolerate the full amplitude of level change between seasons.

The resolution floor is real and worth stating plainly. A water body of one hectare occupies roughly 100 pixels at 10 m resolution. Shoreline mixed pixels, where a single pixel contains both water and bank, degrade the area estimate at the margins. Bodies below about one hectare become unreliable at Sentinel-2 scale. Landsat, at 30 m, pushes that floor to roughly nine hectares. For screening purposes this is generally acceptable: floating PV projects below one hectare of installed area are rarely bankable at grid scale, though community or agricultural applications may justify smaller installations assessed by other means.

Reading seasonal drawdown from a multitemporal stack

A single scene tells you the water-body extent on one date. A stack of twenty or thirty cloud-free scenes across two or three years tells you the distribution of extents: the median, the dry-season minimum, and the interannual variability. Reservoirs operated for irrigation typically show a predictable annual drawdown pattern. Mine-water lakes, by contrast, tend to be more stable because they are not actively managed for abstraction, though they carry their own due-diligence questions around water chemistry and structural safety of impoundment walls.

ICESat-2's ATL13 product adds a dimension that optical imagery alone cannot provide: absolute water-surface elevation with centimetre-scale vertical precision. Where an ICESat-2 ground track crosses a target reservoir, repeat passes separated by 91 days give a direct measurement of level change. Combined with a bathymetric estimate (itself uncertain without in-situ sonar, but approximable from DEM analysis of the surrounding terrain and known spillway elevations), this constrains the depth available for anchor systems at low-water conditions. Shallow seasonal drawdown to less than two metres of depth is a common disqualifier for conventional anchor designs.

Topographic shading: the variable most often ignored in early screening

A reservoir in a steep-sided valley can lose several hundred kilowatt-hours per kilowatt-peak per year to horizon shading, relative to an open-country site at the same latitude. The SRTM DEM at 30 m posting is sufficient to compute horizon elevation angles around each candidate body at the eight cardinal and intercardinal directions. Solar geometry then converts those angles into annual shading loss estimates. The calculation is well-established in photovoltaic yield modelling and requires no proprietary data.

The honest caveat is that SRTM's 30 m posting smooths narrow ridges and misses deep valley incisions at sub-30-m scale. A reservoir flanked by a sharp ridge 50 m wide will have that ridge underrepresented in the DEM, producing an optimistic shading estimate. Where the screening model flags a site as borderline on shading, a higher-resolution commercial DEM or airborne lidar survey is the appropriate next step, not further satellite analysis.

Ranking candidates and what the output actually looks like

The output of a regional screening exercise is a ranked candidate list, typically delivered as a GIS layer with per-body attributes: stable annual minimum area in hectares, seasonal area-change ratio, estimated topographic shading loss as a percentage of open-sky irradiance, ICESat-2 level-change amplitude where track coverage exists, and a composite suitability score. Each attribute carries an honest confidence band. A body with ten years of Landsat history and three ICESat-2 overpasses has a tighter confidence interval than one seen in only four cloud-free Sentinel-2 scenes.

Satellize runs this screening workflow on open constellations for clients who need to assess large national or regional portfolios quickly. The methodology is the same class of multitemporal water-body analysis applied in the Tonga crop-estimation programme, adapted from vegetation indices to water indices and extended with DEM-derived shading geometry. The deliverable is a ranked GIS layer and a one-page summary card per shortlisted site, ready to hand to a project developer for field validation.

Where satellite analysis stops and field work begins

Satellite screening is a filter, not a feasibility study. It removes the obviously unsuitable and orders the remainder by proxy indicators of suitability. It does not measure actual water depth, anchor-point sediment type, grid-connection distance, land tenure, water-use rights, or the structural integrity of dam walls. All of those require site visits and specialist surveys. The value of the satellite step is that it concentrates field-survey budgets on the twenty sites most likely to be viable, rather than spreading them across two hundred.

Cloud cover is the operational constraint most likely to affect delivery timelines in tropical and monsoonal regions. A wet-season screening in a persistently cloudy basin may require a longer archive window to accumulate enough cloud-free scenes for a reliable dry-season minimum estimate. Synthetic aperture radar can detect open water through cloud, but at coarser effective resolution for small bodies and without the spectral discrimination that separates turbid shallow water from moist bare soil. For most inland water-body screening, optical imagery remains the primary tool, with cloud management handled through archive depth rather than sensor switching.

Typical figures

Primary optical resolution	10 m (Sentinel-2 MSI visible/NIR); 30 m (Landsat 8/9 OLI)
Revisit frequency	5 days at equator (Sentinel-2 twin satellites); 8 days combined (Landsat 8 + 9)
Minimum reliably mapped water body	Approximately 1 ha at 10 m resolution; approximately 9 ha at 30 m resolution
Water-level measurement precision (ICESat-2 ATL13)	A few centimetres on calm inland water surfaces; 91-day repeat cycle
DEM vertical accuracy (SRTM)	Approximately 10 m at 90th percentile globally; 30 m horizontal posting
Spectral bands used	Green (560 nm), NIR (842 nm) for NDWI; Red-edge and SWIR for land/water disambiguation
Archive depth	Sentinel-2: from 2015; Landsat: from 1972 (mission-dependent); SRTM: single 2000 acquisition
Topographic shading model input	SRTM 30 m DEM; horizon angles computed at 8 azimuths per candidate body
Deliverable format	GeoPackage or GeoJSON ranked candidate layer; per-site PDF summary cards
Typical screening latency	2 to 6 weeks depending on cloud-free archive availability and number of candidate bodies

Analytics Satellize can run

Seasonal minimum surface-area map	Multitemporal NDWI thresholding on Sentinel-2 and Landsat time series; percentile compositing to extract dry-season minimum extent	GIS polygon layer with per-body minimum and median area attributes, confidence band, and scene-count metadata
Seasonal water-level fluctuation estimate	ICESat-2 ATL13 repeat-pass surface elevation differencing; supplemented by shoreline-retreat proxy from optical time series where lidar track coverage is absent	Per-body attribute table with level-change amplitude in metres and data-source flag (lidar or optical proxy)
Topographic horizon shading loss index	SRTM DEM horizon-angle computation at eight azimuths; solar geometry integration to estimate annual shading fraction relative to open-sky conditions	Per-body shading-loss percentage attribute appended to candidate GIS layer, with flagged uncertainty for bodies in narrow valleys
Composite suitability ranking	Weighted scoring across stable minimum area, seasonal area-change ratio, shading loss index, and ICESat-2 level-change amplitude; weights configurable to client anchor-system specifications	Ranked candidate list as GIS layer and CSV, with score breakdown per criterion
Long-term water-body trend assessment	Landsat archive NDWI time series from 1984 onward; Mann-Kendall trend test on annual minimum area to identify bodies in secular decline	Trend classification (stable, declining, recovering) per body, with time-series chart in PDF site summary
Cloud-gap analysis and data-confidence report	Scene-availability audit per candidate body across Sentinel-2 and Landsat archives; cloud-fraction statistics by month to flag regions where dry-season screening is data-limited	Per-region data-confidence table identifying candidate bodies where additional SAR or commercial optical tasking would be needed to reduce uncertainty

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.