Most of the water that causes trouble on the north Indian plain does not fall on the plain. It falls upstream, on ground that is steeper, higher and usually in another jurisdiction, and then it travels.
That travel time is the only reason early warning is possible at all. A catchment converts rainfall into river level with a delay set by its slope, its soil and how wet it already was. On a steep Himalayan headwater that delay is hours. By the time the same water is moving across a delta it can be a day or more. Everything a warning system does happens inside that window.
The same physics, three different problems
It is worth being precise about which basin does what, because they are often spoken about as one thing:
- The Ganga and Brahmaputra systems collect from Nepal, Tibet and the eastern Himalaya, and deliver into Bihar, Bengal and Assam. Their headwaters are largely outside India.
- Odisha’s rivers, the Mahanadi, the Brahmani and the Baitarani, rise inside India, mostly in Chhattisgarh, Jharkhand and Odisha itself. Their upstream problem is domestic, not international.
- Both share the same failure mode: the place the water starts is not the place that has to decide, and the two are often not talking on the same clock.
Upstream rain is information. It only becomes a warning if it crosses an institutional boundary faster than the water does.
What this means for a decision engine
It means rainfall alone is a weak signal, and we have the measurements to say so rather than the intuition. Scored on heavy monsoon days in north Odisha, days with a recorded flood had a median of 1.94 times the local wet-day p95. Days with no recorded flood had 2.03. The days without floods had more rain.
Rain tells you what the sky did. River level tells you what the catchment did with it, which is the part that decides whether a road goes under. That is why river telemetry sits on the risk path, and why a station only counts toward a score when it is genuinely live rather than merely present.
It also means the honest unit of a warning is not a probability. It is a length of usable road. Forty-eight hours of lead is roughly two days in which a household can move grain above waist height and an officer can pre-position at a bridge. That is the thing being bought, and it is the thing worth measuring.
Explainer. The rainfall medians and the gating rule appear on the research page with their labels. This post makes no claim about any specific recent flood.