On each river, we link each station to the one above it. We measure in the history with how many days of delay the 5-day change at one shows up at the other and how much of the change passes (correlation and coefficient). We only use it when the correlation is 0.6 or more and the upstream change is not small.

On big, slow rivers, like the Solimões, water from upstream takes days to arrive. On others, the delay is under a day and the link is only context.

Fixed rules: a station at alert or above the flood level; a 7-day change beyond one and a half times the station's typical change; a rising river whose range reaches the high-water mark within 15 days; a river in low water and still falling; upstream water on its way. We show at most four. No text is written by artificial intelligence.

Each station's ruler is its own zero, not sea level, and official levels use the same ruler. Stations near the sea feel the tide. Stations can fail for days, and ANA corrects readings later.

The history is about 35 years. Climate change and changes to riverbank use make the past represent the future only in part.

Rivers regulated by dams, like the Xingu at Altamira or the São Francisco at Pirapora, can sit outside normal because of how the plants are operated rather than because of rain. At stations whose level shifted to a new range (a dam, works or a gauge), the reference uses only the years since the change.