Sensors
- Sentinel-1 SAR C-band (ESA): 10 m ground resolution, 6-day repeat at mid-latitudes (12-day for a single satellite), all-weather imaging. C-band backscatter drops sharply over open water, making it the primary tool for flood duration mapping across the growing season. Double-bounce from rice stems complicates open-water discrimination at full canopy, which is why early-season and inter-season acquisitions are most diagnostic.
- ALOS-2 PALSAR-2 L-band (JAXA): 3 m to 100 m resolution depending on acquisition mode; 14-day repeat. L-band (1.27 GHz) penetrates the rice canopy more effectively than C-band, allowing flood detection even under a closed canopy. The longer wavelength also distinguishes double-bounce from flooded vegetation from specular reflection off open water, improving inundation mapping at mid-season when C-band performance degrades.
- MODIS Terra/Aqua (NASA): 250 m to 500 m resolution, near-daily revisit. Useful for establishing the seasonal hydrological context across large river basins: monsoon onset timing, recession dates and inter-annual flood anomalies. Too coarse for field-level risk mapping but valuable for stratifying the study area before applying the finer-resolution SAR analysis.
- Published soil geochemical databases: National and regional soil surveys (including the USGS National Geochemical Survey, the British Geological Survey World Soil Geochemical Atlas data, and country-level surveys from Bangladesh, India and Vietnam) provide arsenic, iron and organic-carbon concentrations in surface soils. These are static layers, not satellite data, but they are the geochemical prior that gives the flood-duration signal its risk meaning.
Why waterlogging is the trigger, not just the context
Arsenic in paddy soil is not uniformly available. In aerobic conditions, arsenic binds tightly to iron oxyhydroxides, the rust-coloured minerals that coat soil particles. Flood the field, exclude oxygen, and soil bacteria begin reducing ferric iron to ferrous iron as they exhaust other electron acceptors. The oxyhydroxide lattice dissolves. Arsenic, adsorbed to that lattice, is released into pore water. From there it enters the rice plant through aquaporin channels that cannot readily distinguish arsenite from silicic acid.
The critical variable is how long the anaerobic condition persists. Short floods allow partial re-oxidation. Prolonged or permanent flooding, common in transplanted rice systems that maintain a standing water layer for weed suppression, sustains the reducing conditions that keep arsenic in solution long enough for root uptake and eventual grain loading. Duration is therefore the quantity that matters most, and duration is something SAR time series can approximate with reasonable fidelity.
What a SAR time series actually measures, and what it does not
Sentinel-1 C-band detects open water through low backscatter: smooth water surfaces return the radar pulse away from the sensor, producing a dark signal. A pixel that is dark across many consecutive acquisitions has been inundated for most of that period. Counting those acquisitions and multiplying by the revisit interval gives a flood-duration proxy in days. This is a proxy, not a measurement. It conflates shallow standing water with deep inundation, misses water hidden under a closed canopy, and can be confused by smooth bare soil or calm wind on large water bodies.
ALOS-2 PALSAR-2 partially addresses the canopy problem. L-band double-bounce, where the signal reflects off the water surface and then off vertical stems before returning to the sensor, produces a bright return that is diagnostic of flooded vegetation. This signature is detectable even at peak biomass, extending the flood-duration estimate into the mid-season period when C-band becomes unreliable. Even so, neither sensor measures dissolved arsenic concentrations, pore-water redox potential, or iron reduction rates. The satellite output is a spatial index of flood permanence, not a geochemical assay.
Combining flood duration with soil geochemistry to build a risk layer
Flood duration alone does not identify risk. A permanently flooded field on low-arsenic alluvial sand poses a different problem from a moderately flooded field on arsenic-enriched Holocene deltaic sediments. The risk proxy requires both inputs: the SAR-derived flood permanence index and a soil arsenic or iron concentration layer from a published geochemical survey.
The method, described in peer-reviewed literature for the Bengal Basin and the Red River Delta, bins fields by flood-duration quartile and cross-tabulates them against soil arsenic concentration classes. Fields in the highest flood-duration quartile and the highest soil arsenic class form the elevated-risk stratum. This is not a prediction of grain arsenic content; it is a spatial prioritisation that tells field programmes where to concentrate soil pore-water sampling and grain testing. The honest limitation is that geochemical survey data are often sparse, dated, or collected at resolutions too coarse to distinguish individual fields. Where survey data are absent, the risk layer has a large uncertainty band that should be stated explicitly in any output.
Resolution floors, cloud and other practical constraints
SAR is cloud-independent, which is its main advantage over optical sensors in monsoon-affected rice systems. But 10 m Sentinel-1 pixels are not always sufficient to resolve the narrow bunds and small field sizes common in smallholder landscapes across Bangladesh, Cambodia or the Philippines. Fields smaller than roughly two or three pixels in width will have mixed returns that understate inundation extent. PALSAR-2 Spotlight mode can reach 3 m, but acquisition is tasked and not freely available at the same cadence as Sentinel-1.
Archive depth matters for inter-annual comparison. Sentinel-1 data run from April 2014 (Sentinel-1A launch). PALSAR-2 data are available from 2014 onwards through JAXA's research access agreements. MODIS extends back to 2000, useful for establishing long-term flood climatology. Soil geochemical maps are often decades old and may not reflect changes from groundwater irrigation that has redistributed arsenic vertically through the soil profile. Any risk map should carry a clear vintage statement for each input layer.
From risk proxy to a decision a ministry can act on
The output of this analysis is a tiered risk map: a GIS polygon layer in which each mapped paddy field or field cluster carries a flood-duration score, a soil-arsenic class, a composite risk tier, and an explicit confidence rating based on data density. High-risk, high-confidence polygons are the first candidates for targeted soil and grain sampling campaigns. High-risk, low-confidence polygons flag where the geochemical survey needs updating before any regulatory action is warranted.
Ministries of agriculture and food safety agencies have used risk-stratified sampling designs of this kind to reduce the number of field samples needed to characterise a region, concentrating laboratory costs where the probability of exceedance is highest. The satellite layer does not replace the laboratory; it makes the laboratory programme cheaper and more defensible. Satellize has applied analogous risk-stratification logic in crop-estimation work, including the Kingdom of Tonga programme, and can structure the analytical pipeline and delivery format to match the data systems of a national food safety authority.
A practical programme would run the SAR flood-duration analysis annually across each growing season, update the composite risk layer after harvest, and flag any fields that have shifted tier due to changes in water management. That annual cadence is achievable with freely available Sentinel-1 data and standard change-detection processing.
Typical figures
| Primary SAR spatial resolution | 10 m (Sentinel-1 IW mode); 6 m (PALSAR-2 Fine Beam); 3 m (PALSAR-2 Spotlight, tasked) |
| Revisit interval | 6 days at mid-latitudes (Sentinel-1A+B combined); 14 days (PALSAR-2 single pass) |
| Radar frequency / wavelength | C-band 5.405 GHz / ~5.6 cm (Sentinel-1); L-band 1.27 GHz / ~23.6 cm (PALSAR-2) |
| Flood detection under closed canopy | Unreliable at C-band mid-season; L-band double-bounce detectable under rice canopy up to ~1 t/ha above-ground biomass (published estimates vary) |
| Minimum field size reliably mapped | Approximately 0.1 ha at 10 m resolution in practice; smaller fields require PALSAR-2 fine-beam or Spotlight |
| Soil geochemical layer resolution | Highly variable: national surveys typically 1:250,000 to 1:50,000 scale; point data may be kilometres apart |
| SAR archive depth | Sentinel-1 from April 2014; PALSAR-2 from 2014 (research access); MODIS from 2000 |
| Cloud sensitivity | None for SAR; MODIS optical bands affected by monsoon cloud cover |
| Deliverable format | GeoTIFF risk-tier raster; field-polygon GeoPackage with per-field attributes; PDF summary report with confidence ratings |
Analytics Satellize can run
| Seasonal flood-duration index map | Sentinel-1 IW backscatter time series thresholding and change-point detection; Otsu or histogram-based water/land classification applied per acquisition, then summed across the season | GeoTIFF raster (10 m) with per-pixel inundation-day count for each growing season; updated annually |
| Canopy-corrected inundation layer | PALSAR-2 L-band double-bounce decomposition to detect flooded vegetation; fused with Sentinel-1 open-water mask to extend detection into mid-season | GeoTIFF binary inundation mask per PALSAR-2 acquisition; time-series stack for duration calculation |
| Composite arsenic mobilisation risk tier | Cross-tabulation of flood-duration quartile against published soil arsenic concentration class; risk tier assigned per field polygon; confidence score based on geochemical data density within 10 km radius | Field-polygon GeoPackage with risk tier (1 to 4), confidence class (high/medium/low), and input data vintage for each attribute |
| Prioritised field sampling frame | Stratified random sample design drawn from the risk-tier layer; sample size calculated to achieve specified detection probability at each tier given laboratory budget | Shapefile of proposed sampling points with GPS coordinates, field ID and stratum label; exportable to ODK or KoBoToolbox for field teams |
| Inter-annual flood anomaly report | MODIS Terra/Aqua NDWI seasonal composites compared against the 10-year median flood extent; anomaly expressed as deviation in inundated-area fraction per administrative unit | Annual PDF bulletin with basin-level flood anomaly maps and tabulated statistics per province or district |
| Risk layer update after water-management change | Year-on-year comparison of flood-duration index to detect fields that have shifted from intermittent to permanent flooding or vice versa; tier reassignment flagged for ground-truth | Change-detection GeoPackage highlighting fields with tier change, with before/after flood-duration values and flag for field verification |
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.