Method & accuracy

How Doobega decides — and how often it’s right.

Doobega never guesses a score. It collects dated, sourced evidence for the exact spot you search, then applies a few plain rules. This page explains those rules, and shows an honest test against real floods.

Four kinds of evidence

Bengaluru is divided into about 235,000 hexagonal city blocks (Uber’s H3 grid, resolution 10 — each roughly the size of a large apartment complex). For each block, and for the exact point you search, Doobega checks:

  1. Floods seen from space. Every pass of the Sentinel-1 radar satellite over Bengaluru since 2015 (via the Copernicus Global Flood Monitoring archive) — about 1,000 dates, 325 with usable views of the city.
  2. Water that used to be. The EU Joint Research Centre’s Global Surface Water record (Landsat, 1984–2021, 30 m), plus lakes documented as built over in government records. We cross-check lost water against ESA WorldCover 2021 to tell land that is now built over apart from lake beds that have simply dried up — most lost water around Bengaluru is the second kind.
  3. Drains and low ground. Lakes, tanks and storm-water drains from OpenStreetMap, and the GEDTM30 bare-earth terrain model (buildings removed) to measure how far above the nearest water a spot sits.
  4. Official lists and reported floods. The city’s flood-prone and low-lying lists and the Karnataka flood-vulnerability survey (391 locations), plus major floods reported by named news sources since 2017, each placed on the map.

How a verdict is chosen

The report shows the strongest level any piece of evidence reaches. Distances are measured from the exact point you searched.

VerdictWhen
Flooding recorded hereA flood was reported at this spot (within 250 m of a specific complex or junction, or 300 m of a reported layout); the city lists it as flood-prone (within 200 m); or the radar satellite saw standing water in this block or the next one, right after heavy rain.
Flooding reported nearbyThe same kinds of evidence, further away: reported floods within 550 m, official flood-prone spots within 350 m, or satellite-seen water within about 450 m.
Warning signsNo recorded flood, but conditions that usually precede one: built on former water (1984–2021 record) or a documented lost lake; inside or within 30 m of a lake; within 40 m of a major drain or stream; ground less than 2 m above nearby water; or an official low-lying area.
No signs foundNone of the above. This is good news, not a guarantee.

Confidence is high when two independent kinds of evidence agree (for example a news report and the satellite), medium for a single strong source, and low for a single warning sign.

Cleaning the satellite signal

Radar sometimes mistakes wet fields, sewage ponds or shadows for flood water. So we keep a detection only if it was seen on a storm pass — a pass right after heavy rain, or one where unusually much of the city looked wet — and only if the same spot is not repeatedly “wet” on dry-weather passes. Lakes that exist today don’t count as floods.

The cleaned signal lines up with the city’s flood history on its own: the passes with the most flood water are 5 September 2022 (the Outer Ring Road floods), 13 October 2017, 21 November 2021 (Kendriya Vihar) and 24 October 2024 (Yelahanka). The satellite cannot see well between tall buildings, so dense neighbourhoods rely more on reports and official lists.

Accuracy test

Loading results…

What it can’t do

  • Tell you whether a particular building’s basement floods — plinth height and drainage vary building by building.
  • Know about floods nobody reported, or water the satellite missed.
  • Reflect drain works or lake restoration done after the data was collected. Reports show the date of the data.
  • Replace a site visit during the monsoon, or a conversation with neighbours.

Found something wrong, or know of a flood we missed? Tell us — corrections are reviewed and added to the next update.