Wetland inundation dynamics and long-term change detection
Mapping wetland inundation requires sensors that see through vegetation, not just over it. This page explains how multi-temporal SAR and the Landsat-derived Global Surface Water dataset together give the most complete picture of seasonal flooding and long-term wetland loss.
Sensors
- Sentinel-1 C-band SAR (ESA): 10 m ground range resolution in IW mode, 12-day single-satellite revisit (6-day with both satellites active). C-band penetrates sparse herbaceous vegetation but is largely reflected by dense reed canopy; open water appears as very low backscatter, making it detectable even under partial cloud cover.
- ALOS-2 PALSAR-2 L-band SAR (JAXA): L-band (1.2 GHz) penetrates emergent macrophytes, mangroves and flooded forest canopy to interact with the double-bounce mechanism between water surface and vertical stems. Spatial resolution 3–10 m depending on mode; revisit 14 days. The physics of double-bounce makes flooded vegetation appear brighter than dry vegetation, the opposite of open-water detection in C-band.
- Landsat 5/7/8/9 archive (USGS/NASA): 30 m multispectral, 16-day revisit per satellite. The Global Surface Water dataset, produced by the European Commission Joint Research Centre from the full 1984-to-present Landsat archive, is the authoritative public baseline for long-term inundation change. SWIR bands (bands 5 and 6 on Landsat 8/9) help distinguish turbid shallow water from bare soil.
- Sentinel-2 MSI (ESA): 10–20 m multispectral, 5-day revisit at mid-latitudes with both satellites. NDWI (Green/NIR) and MNDWI (Green/SWIR) are the standard optical water indices; the 20 m SWIR bands (B11, B12) are critical for suppressing false positives over turbid water and wet bare soil. Blocked entirely by cloud, which is a genuine operational constraint in monsoon and tropical wetland regions.
What a floating roof gives away
A reed bed looks, from optical orbit, like a continuous green or brown surface. It gives almost nothing away about what is happening at ground level. This is the central problem in wetland remote sensing: the thing you want to map (standing water) is hidden by the thing that defines the habitat (emergent vegetation). Optical indices such as NDWI and MNDWI work well over open water bodies, but they fail wherever a plant canopy intervenes. For seasonal floodplains where papyrus, Phragmites or mangrove roots are inundated for weeks at a time, optical sensors can underestimate inundated area by a wide margin.
L-band SAR changes the geometry of the problem. At 1.2 GHz, the wavelength (roughly 24 cm) is long enough to pass through herbaceous and even woody emergent vegetation and reflect off the water surface below. When the signal bounces between a vertical stem and a flat water surface, it returns to the sensor much stronger than from dry vegetation. This double-bounce mechanism is well documented in the literature and is the physical basis for ALOS-2 PALSAR-2's value in flooded forest and mangrove mapping. The effect is detectable at the scale of individual tree trunks. C-band Sentinel-1 sees it too, but only through sparser or shorter canopies; dense reed beds at C-band tend to attenuate the signal rather than produce a clean double-bounce return.
Forty years of surface water, pixel by pixel
The Global Surface Water (GSW) dataset, produced by the Joint Research Centre of the European Commission using the full Landsat archive from 1984 onwards, is the most important public baseline for long-term wetland change. Every 30 m pixel in the dataset carries a monthly water history: how often it was classified as water, whether that classification is seasonal or permanent, and how the pattern has changed decade by decade. The dataset is freely available through Google Earth Engine and the Copernicus Data Space.
For a wetland manager or a government planning authority, the GSW gives immediate answers to questions that previously required decades of fieldwork. Which areas have converted from seasonal to permanent inundation? Where has permanent water disappeared entirely? The dataset has known limits: 30 m pixels miss narrow channels and small ponds; cloud contamination in tropical regions can leave entire wet seasons with sparse valid observations; and the 16-day Landsat revisit is too slow to capture the rising limb of a flash flood pulse. Those limits are real, and they are why SAR time series are added rather than substituted.
Sentinel-1 time series: catching the pulse
Seasonal inundation in wetlands is not a binary state. Water arrives, spreads, retreats and leaves behind a mosaic of saturated soil, shallow pools and drying mud that changes week by week. Sentinel-1's 6-day revisit (when both satellites are operational) at 10 m resolution makes it possible to track that pulse in near real time and to build a multi-year archive of inundation frequency at sub-monthly resolution.
The standard processing chain applies a threshold to the VV or VH backscatter intensity: open water returns very low backscatter (typically below about -15 dB in VV), while dry land and vegetation return higher values. The threshold is not universal. Rough water surfaces (wind-driven waves, heavy rain) can raise backscatter above the threshold and produce false negatives. Wet soil and saturated peat can lower backscatter below the threshold and produce false positives. Operational systems therefore use change detection against a dry-season reference image rather than a fixed threshold, and some apply machine-learning classifiers trained on hand-labelled scenes. The 10-day to 6-day revisit is good enough to resolve the hydroperiod (the number of days per year a pixel is inundated) to within a few days for most wetland systems.
The turbid-water problem in optical indices
NDWI (McFeeters 1996) uses the green and NIR bands. It works well over clear, deep water. It struggles badly over shallow turbid water, where suspended sediment raises NIR reflectance enough to push the index below zero, making the pixel look like dry land. Seasonal wetlands are almost always shallow and often carry a heavy sediment load after rainfall. MNDWI (Xu 2006) substitutes SWIR for NIR and performs better because water absorbs strongly in the SWIR regardless of turbidity. Even so, wet bare soil and damp peat can produce MNDWI values that overlap with shallow water, particularly at 30 m resolution where mixed pixels are common at the water's edge.
The practical answer is to use MNDWI as a first pass and then apply a SWIR-based soil index (such as the Automated Water Extraction Index, AWEI) to suppress false positives over bare soil. Neither approach is perfect at the margin. When the distinction between very shallow turbid water and moist bare soil is operationally important, the correct tool is SAR, not optical. Combining both in a decision-tree or random-forest classifier consistently outperforms either alone in published validation studies.
What the numbers can and cannot tell you
Spatial resolution sets a hard floor on what is detectable. At 30 m (Landsat/GSW), channels narrower than about 60–90 m are systematically underestimated. At 10 m (Sentinel-1, Sentinel-2), the floor drops to roughly 20–30 m, which captures most major wetland channels but still misses drainage ditches and narrow tidal creeks. PALSAR-2 in spotlight mode reaches 3 m, but at the cost of narrow swath width and infrequent revisit, making it better suited to targeted validation than basin-wide monitoring.
Cloud cover is the dominant constraint on optical time series in tropical and monsoon wetland regions. In parts of the Congo Basin or the Amazon, months of the wet season may have fewer than two cloud-free Landsat or Sentinel-2 observations. SAR is the operational sensor in those conditions. The trade-off is that SAR cannot distinguish water quality, turbidity or aquatic vegetation type: it tells you where water is, not what kind. A complete wetland monitoring system uses both, with SAR providing the all-weather inundation mask and optical providing the ecological context when skies clear.
Satellize runs multi-sensor inundation analytics on open Sentinel and Landsat archives, with optional PALSAR-2 tasking for vegetated-wetland campaigns. The approach is similar in structure to the crop-mask methodology developed for the Kingdom of Tonga programme, adapted to hydrological rather than agricultural classification targets.
Typical figures
| Spatial resolution (Sentinel-1 IW mode) | 10 m ground range, 10 m azimuth |
| Spatial resolution (PALSAR-2 standard mode) | 6–10 m; spotlight mode to 3 m |
| Spatial resolution (Landsat / Global Surface Water) | 30 m |
| Revisit (Sentinel-1, both satellites) | 6 days at mid-latitudes; 12 days with single satellite |
| Revisit (ALOS-2 PALSAR-2) | 14 days |
| Revisit (Landsat 8 + 9 combined) | 8 days |
| Minimum detectable open-water body (Sentinel-1) | Approximately 0.01 ha at 10 m resolution; narrower channels systematically underestimated |
| Archive depth (Global Surface Water baseline) | 1984 to present (Landsat 5, 7, 8, 9) |
| SAR frequency bands used | C-band (5.4 GHz, Sentinel-1); L-band (1.2 GHz, PALSAR-2) |
| Cloud penetration | SAR: full; Optical (Sentinel-2, Landsat): none |
Analytics Satellize can run
| Monthly inundation frequency map | Multi-temporal Sentinel-1 VV/VH thresholding with change-detection reference; cloud-gap-filled using temporal compositing | GeoTIFF raster stack, monthly cadence, delivered as GIS layer or via API |
| Long-term surface water change report | Joint Research Centre Global Surface Water dataset analysis: permanent water loss, seasonal variability trends, decade-on-decade comparison | PDF report with annotated maps and tabular area statistics per wetland unit |
| Flooded-vegetation extent (under-canopy water) | ALOS-2 PALSAR-2 L-band double-bounce classification for flooded forest and mangrove; validated against coincident Sentinel-1 open-water mask | Polygon shapefile with confidence score per class; updated per PALSAR-2 acquisition cycle |
| Hydroperiod map (days inundated per year) | Pixel-wise summation of binary inundation classifications across Sentinel-1 annual time series | Annual GeoTIFF raster; time series chart per user-defined wetland polygon |
| Optical water index composite (MNDWI/AWEI) | Sentinel-2 SWIR-based indices with soil-suppression filter; cloud-masked seasonal composites | Seasonal GeoTIFF composites with per-pixel validity count; suitable for ecological baseline reporting |
| Wetland loss and degradation alert | Change detection against multi-year GSW and Sentinel-1 baseline; anomaly flagging when inundation frequency drops below historical interquartile range | Email or webhook alert with affected area polygon and magnitude estimate |
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.