Surface-water permanence and seasonal wetland mapping for biodiversity

Multi-temporal Landsat and Sentinel imagery can resolve how long water sits on any patch of ground, year by year. That hydroperiod signal is the foundation of waterbird, amphibian and freshwater-invertebrate habitat assessment.

Sensors

Landsat 8/9 OLI (JRC Global Surface Water): 30-metre pixels, 16-day revisit per satellite (8-day combined). The JRC Global Surface Water dataset uses the full Landsat archive from 1984 onward to compute monthly water occurrence, seasonality, recurrence and transitions at global scale. Cannot reliably detect water bodies narrower than roughly 30 metres.
Sentinel-1 SAR (C-band, 5.6 cm wavelength): Interferometric Wide Swath mode delivers 10-metre pixels at 6-day revisit over most land areas (12-day at equator with a single satellite). C-band backscatter drops sharply over open water, enabling cloud-independent inundation mapping during monsoon and flood periods when optical sensors are blind. Dense vegetation attenuates the signal, causing underestimation of inundation beneath forest canopy.
Sentinel-2 MSI: 10-metre visible and near-infrared bands, 20-metre shortwave-infrared bands, 5-day revisit at mid-latitudes with both satellites. NDWI (Green–NIR) and MNDWI (Green–SWIR) indices separate open water from vegetation and soil with high spatial fidelity, but cloud cover in wet seasons can remove 60–80 percent of acquisitions over tropical wetlands.
MODIS MOD44W: 250-metre annual global water-mask product derived from MODIS Terra and Aqua. Near-daily revisit allows coarse-scale seasonal tracking and gap-filling over large basins, but the spatial resolution is too coarse for small wetland patches and narrow river channels. Useful for continental-scale phenology and as a consistency check.

What a hydroperiod actually tells a biologist

Inundation frequency is not just a hydrological curiosity. The number of months per year that a wetland holds water, and how reliably it does so across years, determines which species can complete a breeding or feeding cycle there. Waterbirds require predictable shallow water for foraging; many amphibians need inundation to last long enough for tadpoles to metamorphose, typically six to twelve weeks depending on species. Freshwater invertebrate assemblages tracked by biodiversity indices such as the Freshwater Living Planet Index respond strongly to hydroperiod stability.

Satellite-derived hydroperiod maps convert this ecological requirement into a spatial layer. A pixel classified as 'permanent water' in the JRC Global Surface Water dataset has been observed as water in more than 75 percent of valid Landsat observations over the full archive. 'Seasonal water' sits between roughly 25 and 75 percent occurrence. Those thresholds are operationally defined and published, not inferred, which matters when a habitat assessment has to survive regulatory scrutiny.

The JRC Global Surface Water dataset: what it delivers and where it breaks

The Joint Research Centre's Global Surface Water dataset, produced from roughly 4.4 million Landsat scenes spanning 1984 to the present, is the most widely used public record of surface-water change at global scale. It publishes six derived layers: occurrence, occurrence change intensity, seasonality, recurrence, transitions and maximum extent. Recurrence is particularly useful for biodiversity work: it shows whether seasonal water returned in the same months across multiple years, distinguishing a reliable seasonal wetland from an anomalous flood.

The 30-metre pixel is the dataset's most consequential constraint. A drainage ditch, oxbow lake or seasonal pool narrower than one pixel width is invisible or intermittently detected depending on adjacency effects. Studies comparing JRC outputs with high-resolution aerial surveys in temperate agricultural landscapes have found systematic underestimation of small-wetland area, sometimes by 30 to 50 percent, though published figures vary by landscape type. The dataset also struggles in arid regions where highly turbid or saline water has spectral signatures that partially overlap with dry soil. Users should treat the maximum-extent layer as a lower bound, not a census.

Cloud contamination is the other honest limit. In persistently cloudy climates, the number of valid Landsat observations per month can fall to two or three, which is insufficient to characterise intra-seasonal variability. The JRC methodology applies a cloud-mask filter before classification, so pixels with few valid observations carry higher uncertainty and should be flagged in any downstream habitat model.

Sentinel-1 fills the gap that clouds create

Synthetic Aperture Radar at C-band penetrates cloud and light rain. Over open water, the smooth surface returns almost no energy toward the sensor, producing a characteristically dark backscatter signature typically below minus 15 dB in VV polarisation. This makes flood and inundation mapping during monsoon seasons feasible when months of optical data are simply unavailable.

The practical workflow combines a pre-event or dry-season baseline backscatter image with a flood-period acquisition. Pixels that drop significantly below the baseline, after accounting for wind roughening and incidence-angle effects, are classified as newly inundated. ESA's Sentinel-1 archive now extends back to 2014, long enough to compute seasonal inundation statistics independently of Landsat and to cross-validate JRC occurrence layers.

Two limits deserve explicit mention. First, C-band does not penetrate dense forest canopy well enough to detect sub-canopy flooding reliably. Flooded forests, which are critical habitat for several threatened fish and bird species in Amazonia and Southeast Asia, are systematically underdetected. L-band SAR (ALOS PALSAR-2, or the forthcoming NISAR mission) performs better in that context, though it is not yet available at Sentinel-1 revisit rates. Second, wind speeds above roughly 3 to 4 metres per second roughen open water enough to raise backscatter and cause false negatives. Acquisition metadata should always be checked.

Combining sensors: a practical fusion approach

No single sensor solves the problem. The defensible approach fuses JRC occurrence layers with Sentinel-1 seasonal composites and Sentinel-2 index maps, treating each as an independent evidence layer and resolving disagreements by examining acquisition conditions rather than simply averaging.

For a typical wetland habitat assessment, the workflow runs in three stages. First, generate a maximum-extent mask from JRC and Sentinel-1 union, accepting that this will include some false positives at the margins. Second, classify each pixel's hydroperiod category using multi-year monthly composites from whichever sensor had the most valid observations in each month. Third, apply Sentinel-2 MNDWI at 10 metres to refine boundaries within the coarser Landsat grid, particularly around small ponds and narrow channels that matter for amphibian connectivity. The result is a layered confidence map rather than a single hard classification, which is more honest and more useful for habitat modelling.

Satellize applied a comparable multi-sensor fusion approach in its Tonga crop-estimation programme, where cloud cover and small field sizes created analogous sensor-selection trade-offs. The methodological logic transfers directly to wetland contexts.

Honest limits for the habitat modeller

Satellite-derived water maps are observations of surface reflectance or backscatter, not measurements of water depth, water quality or submerged aquatic vegetation. A pixel classified as seasonal water could be 2 centimetres or 2 metres deep. Amphibian habitat models that require depth information must incorporate ground-truth bathymetry or lidar data; satellite alone cannot supply it.

Change detection over decades is complicated by sensor differences between early Landsat missions and current OLI. The JRC team applied cross-calibration corrections, but users comparing pre-2000 and post-2015 layers should be aware that some apparent changes reflect calibration shifts rather than real hydrological change. Published validation studies estimate overall accuracy of the JRC occurrence product at above 90 percent globally, but accuracy in arid and semi-arid regions is lower, and accuracy for narrow water bodies is substantially lower still.

Revisit frequency also constrains what can be said about rapid hydrological events. A flash flood that fills and drains a wetland within 48 hours may be entirely missed by a 6-day SAR revisit, let alone a 16-day Landsat cycle. For event-scale hydrology, ground sensors or commercial very-high-resolution tasking are necessary complements.

Typical figures

Spatial resolution (optical)	30 m (Landsat OLI / JRC GSW); 10 m visible and NIR, 20 m SWIR (Sentinel-2 MSI)
Spatial resolution (SAR)	10 m ground range detected (Sentinel-1 IW mode)
Minimum detectable water body	~30 m width (Landsat); ~10 m width (Sentinel-2 MNDWI); ~10 m (Sentinel-1); sub-pixel water systematically underdetected
Revisit period	8 days combined (Landsat 8+9); 5 days (Sentinel-2 A+B at mid-latitudes); 6 days (Sentinel-1 over most land)
Cloud penetration	None (optical sensors); full (Sentinel-1 C-band SAR, except heavy precipitation)
Archive depth	1984–present (Landsat / JRC GSW); 2014–present (Sentinel-1); 2015–present (Sentinel-2)
Spectral bands used	Green, NIR, SWIR1, SWIR2 for NDWI/MNDWI (optical); VV and VH polarisation C-band (SAR)
Coarse-scale water mask	MODIS MOD44W at 250 m, annual, global
Typical processing latency	JRC GSW updated to near-present via Google Earth Engine; Sentinel-1 NRT products available within 1–3 hours of acquisition via Copernicus
Delivery formats	GeoTIFF, Cloud-Optimised GeoTIFF, GeoPackage, WMS/WMTS; tabular hydroperiod statistics as CSV

Analytics Satellize can run

Hydroperiod classification map	Multi-year monthly water-occurrence compositing from Landsat OLI (JRC GSW methodology), classified into permanent, seasonal and ephemeral categories	GeoTIFF raster with per-pixel hydroperiod class and observation-count confidence layer; PDF summary for habitat assessment appendix
Seasonal inundation extent polygons	Sentinel-1 VV backscatter change detection against dry-season baseline, thresholded and vectorised per acquisition date	Dated polygon GIS layers (GeoPackage) covering each monsoon or flood season in the analysis period
Wetland-boundary time series	Sentinel-2 MNDWI index computed monthly, water/non-water threshold applied at 10-metre resolution, boundaries extracted and stacked across years	Annual maximum and minimum extent shapefiles with area statistics table; trend chart showing inter-annual variability
Hydroperiod-based habitat-suitability input layer	Occurrence frequency and recurrence statistics from JRC GSW, reclassified to ecological suitability bands defined by client species requirements	Reclassified raster ready for input to MaxEnt or similar species-distribution models; metadata sheet documenting classification thresholds
Flood-season gap-fill composite	Sensor fusion of Sentinel-1 SAR inundation mask with available Sentinel-2 and Landsat clear-sky observations, using per-pixel source flagging	Monthly inundation composites with source-sensor provenance band; confidence raster indicating observation density
Long-term wetland change report	JRC GSW transition layer analysis (1984 to present) identifying permanent loss, seasonal-to-permanent conversion and newly emerged seasonal water	Change map and tabular area statistics by transition class; annotated PDF report suitable for protected-area management plans

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.