Flood pulse extent and duration mapping for floodplain fisheries productivity assessment

Lateral inundation of tropical floodplains drives fish recruitment and nutrient cycling, yet cloud cover blinds optical sensors precisely when the flood peaks. SAR and MODIS time series together map what matters: when the water arrives, how far it spreads, and how long it stays.

Sensors

Sentinel-1 SAR (IW mode, C-band): 10 m ground range resolution in Interferometric Wide Swath mode, 250 km swath, 6-day repeat at the equator with both satellites. C-band backscatter drops sharply over open water, enabling binary flood mapping, but double-bounce return from flooded vegetation can mask inundation beneath dense forest canopy.
ALOS-2 PALSAR-2 (L-band SAR, JAXA): 25 m resolution in ScanSAR mode, L-band (1.27 GHz) penetrates forest canopy more effectively than C-band and detects double-bounce from flooded stems beneath closed-canopy forest, reducing underestimation in the Mekong and Amazon floodplain forest zones. Revisit is 14 days.
MODIS Terra/Aqua surface reflectance (MOD09/MYD09): 250 m to 500 m spatial resolution, combined Terra and Aqua give up to twice-daily overpasses. NDWI and MNDWI indices derived from bands 4 and 6 detect open-water extent on cloud-free days; the dense time series allows flood duration to be estimated statistically even with 30-50% cloud contamination during the wet season.
Landsat-9 OLI-2: 30 m multispectral resolution, 16-day repeat (8 days combined with Landsat-8). Provides higher-resolution flood extent snapshots for calibrating coarser MODIS composites and for mapping flood recession on clear days. Cloud cover during monsoon seasons limits single-pass utility; temporal compositing is required.
Sentinel-2 MSI: 10 m visible and NIR bands, 5-day revisit with both satellites. Useful for recession-phase mapping and for validating SAR-derived inundation boundaries where cloud permits. SWIR band 11 (1610 nm) sharpens water-land discrimination in turbid floodplain conditions.

Why the flood pulse is the productivity signal

The flood pulse concept, formalised by Junk, Bayley and Sparks in 1989 and since validated across the Mekong, Amazon and Congo basins, holds that the lateral exchange between a river channel and its floodplain is the dominant driver of fish biomass production in large tropical systems. When water spills onto the floodplain it dissolves terrestrial organic matter, stimulates aquatic primary production, and opens vast shallow nursery habitat that juvenile fish exploit with little competition from larger predators. The timing, extent and duration of that inundation determine how productive any given year will be.

Catch-per-unit-effort records from the Mekong River Commission and Brazilian fisheries agencies show interannual variation in floodplain fish yields that tracks flood pulse metrics more closely than channel discharge alone. A flood that arrives two weeks late, or recedes two weeks early, can strand juveniles before they reach the channel. That sensitivity is what makes satellite-derived flood metrics genuinely useful for fisheries managers, rather than merely interesting.

What SAR sees that optical sensors cannot

The wet season over the Mekong delta or the Llanos of Venezuela is not a clear-sky event. Optical sensors, including MODIS and Landsat, are frequently obscured by convective cloud for days or weeks at the precise moment flood extent is at its maximum. Sentinel-1 C-band SAR operates at 5.4 GHz and is unaffected by cloud or rain. Over open water, the smooth surface scatters radar energy away from the sensor, producing a characteristically low backscatter return, typically below minus 15 dB in VV polarisation, that is easy to threshold against surrounding land.

The complication is forested floodplain. Beneath a closed canopy, C-band energy is partly absorbed and partly scattered by the canopy itself, so the low-backscatter signature of standing water is suppressed. Studies on the Amazon have shown that C-band SAR can underestimate flooded forest extent by 20 to 40 percent compared with L-band. PALSAR-2 at 1.27 GHz penetrates canopy more effectively and produces a strong double-bounce return from water-trunk interactions, making it the better choice for floodplain forest zones. The practical workflow combines Sentinel-1 for open-water and herbaceous floodplain mapping with PALSAR-2 for forested inundation, accepting that PALSAR-2's 14-day revisit introduces temporal uncertainty near flood peak.

Building a flood duration layer from a noisy time series

A single SAR scene gives extent on one date. Fisheries productivity depends on duration: how many days the floodplain remained inundated above a threshold depth. Deriving duration requires a dense time series, and that means combining sensors. The standard approach uses MODIS NDWI or MNDWI at 250 to 500 m as the backbone, fitting a logistic or piecewise-linear model to the seasonal inundation curve to estimate flood onset date, peak extent, and recession rate. Sentinel-1, acquired every 6 days, anchors the spatial accuracy of that curve at 10 m resolution for open-water areas.

Cloud contamination in MODIS is handled by temporal gap-filling using climatological priors or harmonic regression on multi-year archives. The MODIS archive extends to 2000, giving more than two decades of annual flood pulse metrics for trend analysis. That depth matters: a single anomalous year is noise; a trend in recession timing across 20 years is a signal worth acting on.

Flood duration maps are then intersected with habitat classification layers (open water, herbaceous floodplain, flooded forest, agricultural floodplain) to produce habitat-weighted inundation duration indices. These indices are the environmental covariates that correlate most directly with published catch records.

Connecting inundation metrics to catch records: honest limits

The correlation between satellite-derived flood pulse metrics and catch-per-unit-effort is well-established in the literature for the Mekong and for the central Amazon, but it is not a simple linear relationship and it is not universal. Fish assemblage responses depend on species composition, fishing effort, gear type, and the spatial structure of the floodplain, none of which satellites observe directly. A productivity index built from inundation metrics is an environmental covariate, not a catch forecast.

The practical value is in anomaly detection and interannual comparison. When the 2024 flood pulse duration in a given floodplain reach is 30 percent below the 20-year median, that is a defensible early warning that catch rates may be suppressed, giving managers time to adjust effort or open supplementary areas. It is not a guarantee. Managers who have ground-truth catch records to calibrate against will get more from the index than those who do not.

Spatial resolution is also a genuine constraint. At 10 m, Sentinel-1 resolves individual oxbow lakes and channel connections that matter for fish movement. At 500 m, MODIS misses them entirely. For the main flood extent, MODIS is adequate. For habitat connectivity analysis at the scale of individual fish passages, Sentinel-1 or Sentinel-2 is required, and even then, water depth is not directly observable from backscatter or reflectance without ancillary hydrological modelling.

From annual maps to operational productivity indices

The analytic workflow that Satellize runs for floodplain fisheries clients begins with a 20-year MODIS inundation archive to establish the climatological flood pulse envelope for each river reach of interest, then layers in Sentinel-1 time series from 2014 onward for spatial precision. Annual flood pulse metrics, onset date, peak extent in square kilometres, recession rate in kilometres per day, and total inundation duration in days per pixel, are extracted and compared against the climatological baseline.

The output is a set of GIS layers and a seasonal bulletin, issued at flood peak and again at recession, that summarises the year's flood pulse relative to the historical distribution. Where catch-per-unit-effort records exist, a simple regression model translates the environmental index into a relative productivity score by reach. Satellize's crop-estimation work for the Kingdom of Tonga uses a comparable approach of combining open-constellation time series with ground-truth records to build calibrated environmental indices, and the methodological architecture transfers directly to floodplain fisheries.

The most immediate operational use is not prediction but diagnosis. When a fishing cooperative or a government fisheries agency wants to understand why catch rates in a particular reach dropped in a given year, a satellite-derived flood pulse reconstruction answers the question in days rather than field seasons.

Typical figures

Spatial resolution (SAR open water)	10 m (Sentinel-1 IW); 25 m (PALSAR-2 ScanSAR)
Spatial resolution (optical)	250–500 m (MODIS); 30 m (Landsat-9); 10 m (Sentinel-2)
Revisit period	6 days (Sentinel-1, both satellites); 14 days (PALSAR-2); 1–2 days (MODIS Terra+Aqua combined)
Frequency / spectral bands	C-band 5.4 GHz (Sentinel-1); L-band 1.27 GHz (PALSAR-2); MODIS bands 1–7 (459–2155 nm)
Minimum detectable open-water patch	~0.01 ha at 10 m SAR; ~6 ha at 250 m MODIS (sub-pixel mixing limits detection below this)
Flood duration accuracy	±7–14 days depending on cloud gap-fill method and revisit frequency; better in drier climates
Archive depth	MODIS from 2000; Sentinel-1 from 2014; Landsat from 1972 (lower temporal density)
Cloud penetration	Full (SAR); none (optical; managed by temporal compositing)
Delivery format	GeoTIFF inundation rasters, vector flood extent polygons, CSV time-series tables, PDF seasonal bulletin
Latency (near-real-time SAR)	Sentinel-1 NRT products available within 1–3 hours of acquisition via Copernicus Dataspace

Analytics Satellize can run

Annual flood pulse metrics by river reach	Threshold-based SAR flood detection combined with MODIS NDWI/MNDWI harmonic time-series fitting; logistic curve parameterisation of onset, peak and recession	GeoTIFF rasters of flood onset date, peak extent, recession rate and total inundation duration; one set per hydrological year
20-year climatological inundation baseline	MODIS MOD09 surface reflectance archive processed through MNDWI thresholding with cloud-gap filling by climatological median compositing	Percentile envelope maps (10th, 50th, 90th) of annual peak flood extent and duration per pixel; GeoTIFF and vector format
Flooded forest extent (L-band supplement)	PALSAR-2 double-bounce detection using VH/VV ratio thresholding calibrated against Sentinel-1 open-water mask; canopy-corrected inundation layer	Separate GeoTIFF layer for forested inundation zone, merged with open-water layer into composite flood extent product
Environmental productivity index by reach	Regression of satellite-derived flood duration and extent metrics against published or client-supplied catch-per-unit-effort records; standardised anomaly scoring relative to 20-year baseline	CSV table of annual productivity index scores by reach with confidence intervals; visualised as choropleth GIS layer
Seasonal flood pulse bulletin	Near-real-time Sentinel-1 composite updated at 6-day cadence during flood season; anomaly flagging against climatological envelope	PDF bulletin issued at flood peak and recession, with maps and reach-level summary table; distributed to fisheries agency or cooperative
Habitat connectivity change analysis	Sentinel-1 and Sentinel-2 derived water masks intersected with classified habitat layer (open water, herbaceous floodplain, flooded forest, agricultural floodplain); graph-theoretic connectivity scoring	Annual connectivity matrix by habitat patch; GIS layer of connected floodplain area accessible from main channel on each flood date

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.