ClimaSafe · early warning · ML

Heat and cold risk warning
before the day arrives.

An early-warning system against heat and cold. Pick any point in Spain, a day and an hour, and you'll know your cardiovascular risk level: safe, caution or danger. It combines official weather data with real mortality, a calibrated model ensemble and your personal factors — age, health, medication, activity — to tell you the risk you run, not the average. Live forecast from Open-Meteo.

00
impact

The human cost

3,832 deaths Summer 2025 · MoMo · May 16 – Sep 30

deaths attributable to heat in Spain. 57% occurred in August and 84% were people over 74. The second deadliest summer on record — and the hottest since 1961 (AEMET).

4,581 deaths Summer 2026 · MoMo · as of Aug 19

deaths by August 19 alone: already exceeds all of summer 2025. July (2,025) was the second deadliest month of the series. With September ahead, 2022's record (4,789) is under threat.

Since 2015, high temperatures have claimed more than 27,500 lives in Spain. Not headlines: excess mortality estimated over civil registries by the MoMo system (Instituto de Salud Carlos III).
Heat doesn't warn twice — and with climate change, every summer will be hotter and deadlier than the last.

01
problem

What it solves

Heat and cold don't kill only on the day of the blow:
they wear the body down over days, and the damage shows up late.

The system crosses each province's weather with real registered mortality and computes every day's risk: safe, caution or danger.
It's an early warning: it prefers to warn you — even if sometimes wrong — over staying silent.

It doesn't stop at the population average: it adjusts the risk to your age, health, medication and activity, because a 72-year-old is not a 25-year-old runner.

And climate change isn't slowing down: every summer is hotter than the last and, with it, risky days and their human cost keep rising.
Models
XGBoost (heat) · RandomForest (cold) · LSTM province_hybrid · thermal indices
Data
ERA5 (train) · Open-Meteo (live) · AEMET · OpenUV · MoMo mortality
Scope
Personalized risk per person, province, day and hour · heat and cold
02
uses

One system, many scenarios

03
how it works

How it works

Want the details? Everything is in the technical docs.

04
results

Real numbers

Rec_riesgo (calibrated thresholds) XGBoost 0.668 (heat) · RF 0.612 (cold) · LSTM 0.737 heat / 0.708 cold.
Weekly forecast with confidence bands km² risk maps Daily Telegram alerts AI assistants (MCP) RAG over the docs

All built for the ANFAIA Summer Grants 2026, with support from CESGA, and generated from the dskit template. Architecture and evaluation detail lives in the technical docs.

Proyectos patrocinados por ANFAIA CESGA · SkillNet MalariaSentinel Agentic-Smart-Health