Thirty-Hour Flood Extent for a Breached Embankment District
A district administration needs a flood-extent map within 30 hours for a 2,400 km2 area hit by a monsoon depression. Field teams report embankment breaches near three towns, but roads are cut and only four ground observers are available. Available imagery: Sentinel-1 SAR (10 m, acquired 14 hours after peak flow, ascending pass), Sentinel-2 (10 m, but 85% cloud over the district), and tasked LISS-IV (5.8 m, earliest usable acquisition in four days, cloudy season). The irrigation department holds daily gauge readings at six stations and a 2019 flood-line map. The district also asks whether the map can be used to declare agricultural losses at village level, which would require resolving field boundaries averaging about 1 hectare.
1. Which dataset should the first 30-hour flood map be built on, and how must its limitations be handled before release?
Read the worked answer
Choice: Sentinel-1 SAR. The 30-hour deadline eliminates LISS-IV (four days), and Sentinel-2 at 85% cloud cannot give continuous extent, however attractive its 10 m optical detail. SAR is all-weather and the scene is already 14 hours post-peak.
Reasoning about fitness: 10 m SAR is adequate for block- and village-level extent but marginal for 1 ha fields, so the product must be labelled as an extent map, not a parcel map.
Limitations to handle explicitly:
- Speckle: multilook and filter, but verify that filter size is not erasing narrow channels and field bunds.
- Urban and embankment areas: layover, shadow and strong double-bounce returns cause urban flood to be missed, not exaggerated. Flag built-up zones as low-confidence rather than dry.
- Wet soil, shallow water and dense canopy all dampen or mimic water; a single global threshold will over- or under-map. Use a locally calibrated threshold with visual QC against known permanent water.
- State the detection floor (depth and vegetation condition) in the metadata.
Release with confidence tiers and request a second pass to confirm persistence.
2. How should the six gauge stations and the 2019 flood-line map be used, and what are the specific pitfalls?
Read the worked answer
Gauges - use for timing and plausibility, not for mapping. Six points cannot define a 2,400 km2 footprint, but the readings tell you the hydrograph: when the peak passed each station, and how fast recession is proceeding. That matters because the SAR image is 14 hours after peak; if recession is fast, the mapped extent systematically understates the maximum flood and the loss assessment built on it will undercount. Report the recession rate alongside the map so users know the direction of the bias.
2019 flood-line map - use as a prior, not as truth. It helps separate permanent water bodies and long-standing tanks from new inundation, and it can highlight channels the current threshold missed. Pitfalls:
- The 2019 footprint reflects 2019 breaches and possibly different embankment conditions; do not transfer it forward without checking current breach locations.
- Its own errors will be inherited if you simply union or intersect it with the new extent.
- Check whether its water class included wet sand, fallow paddy or riverbed sand - a common inflation source.
Use it for QC and for annotating permanent water, and say so in the accuracy statement.
3. Can this map support village-level agricultural loss declaration? Justify your position and propose what it can support.
Read the worked answer
No. Three independent reasons:
- Support mismatch. Field boundaries average 1 ha; at 10 m a field is roughly 10 x 10 pixels before filtering. Speckle, bunds and mixed pixels make parcel boundaries unreliable, and boundary error translates directly into false claims and disputes.
- Loss is a duration and stage question, not an extent question. Crop loss depends on how long water stood and at what growth stage. A single post-peak image cannot give duration; at best it hints at maximum depth.
- Consequence severity. Loss declaration triggers compensation with legal standing. A single-pass SAR extent is not defensible evidence at that standard.
What it can support: a ranked list of villages for field verification, with confidence tiers, and an indicative area range rather than a point figure. Pair it with gauge-based duration estimates and a rapid field sample of, say, 40-60 fields to estimate omission and commission rates; then report loss ranges with stated uncertainty and let field verification, not the image, make the final call.