Skip to content
No. 049On holdHeat-health warningsUpdated Nov 20, 2026

Which heat warnings best flag deadly days in Brazil?

Checking warnings against deaths could show which rules best flag dangerous heat across Brazil's varied climates.

The question the lab is investigating

Brazil's national weather service (INMET) issues heat-wave warnings across the whole country with one rule, daily maximum temperature at least 5 °C above the monthly mean for at least five days; how well does that rule pick out the days when deaths actually rise in Brazil's very different climates, and what simple local rule would do better?

What's going on

  • Brazil's weather warnings use duration tiers, not the single five-day rule first assumed.
  • The health ministry moved to a new heat-health warning panel in June 2026.

Why it's stuck

  • Checks against deaths have mostly focused on individual cities.
  • Work on new warnings is already active; another local rule could arrive too late.

What we'd try

  • The lab would compare weather warnings, the health panel and local rules against deaths.
  • It would test rules on later years and cities not used to fit them.

Where we stand

  • On hold: the original question is out of date, though a narrower comparison gap remains.
  • The lab has no confirmed user for that comparison; the work may duplicate existing efforts.
  • One reason to reopen: a weather or health agency says the comparison would inform warning updates.

Written by the lab's AI (openai/gpt-6.1-sol) and checked against the lab's record by a model from another company (google/gemini-4-argon).

Good it would do
3 / 5
Can AI help
3 / 5
How neglected
2 / 5
Can we check it
4 / 5
Read the lab's full wording

Why it's on hold

Park. INMET's warnings are tiered (2-3, 3-5, more than 5 days), not one five-day rule, and the Ministry of Health moved its national heat-health alerts to an Excess Heat Factor panel in June 2026, so the question as written is out of date; the narrower open gap (scoring INMET's tiers, the EHF panel and a local rule against deaths on held-out years) is real but sits in an active field with no confirmed user.

Why it's stuck

The national rule has stayed in use while the checking has been done one city at a time. A 2026 SciELO preprint on Belo Horizonte (Lucas, Caiaffa and others) states that 'o critério operacional do Instituto Nacional de Meteorologia (INMET) adota temperatura máxima diária ≥5 °C acima da média mensal por pelo menos cinco dias, definição que pode não contemplar a diversidade geoclimática do território' (the rule may not suit the country's varied climates).

What we'd try

Brazil's death records (SIM, on DATASUS) are public by date and municipality, and INMET publishes station data; gridded temperature such as ERA5 covers places without stations. For every municipality or group of municipalities large enough, replay the INMET rule over 2000 to the latest year, and measure how often warning days coincide with excess deaths (hits, misses and false alarms), compared with simple alternatives: local percentile thresholds, shorter persistence, minimum temperature, or humidity-adjusted heat.

We'd look again if

INMET or the Ministry of Health says a held-out comparison of INMET's tiered warnings, the EHF panel and simple local rules would feed into the new extreme-heat warnings or the panel's next revision; or a short check finds the EHF panel's thresholds published with no validation against deaths and nobody doing it; or INMET's alert archive proves complete from mid-2021 with enough clean years for a pooled test in the largest cities.

Source: https://doi.org/10.1590/SciELOPreprints.15303

Think we got it wrong?

Ruled-out problems can come back. Tell us what we missed.

Members can read the full record for this problem, check its sources and comment. Open the member view → Not a member? Here is how to join.