Degdeg — Climate–Health Early Warning for Eastern and Southern Africa

A GIS-Enabled, AI-Powered Climate-Health Early Warning System

Climate–health early warning for Eastern and Southern Africa

Satellite Earth observation of current conditions, combined with climate forecasts and turned into anticipated health impacts — before a hazard becomes an emergency. Every advisory cites a protocol rule and is approved by a person before it publishes.

Operational in Ethiopia's Somali Region. Pilot districts live in Kenya, Uganda and South Africa.

Designed in alignment with the UN Early Warnings for All initiative

View Live Demo Read Our Methodology

View live dashboard →

4 countries

106 districts monitored

4 satellite Earth-observation sources

7 interactive map layers

Assessments updated 39h ago

8 WHO & national protocol sources

~2,000 indexed protocol passages

Weekly Climate-Health Outlook bulletin

Every advisory human-approved

What We're Grounded In

Degdeg isn't driven by a single forecast — it combines satellite-derived Earth observation of current conditions with forecast data, using true polygon zonal statistics computed across each district's full boundary, not a single sampled point.

CHIRPS

Observed

Observed rainfall and its anomaly

Every 5 days (pentad)

ICPAC Combined Drought Indicator

Observed

Combined Drought Indicator from a WMO-accredited regional climate centre

Every 10 days (3–13 week publication lag)

FEWS NET eVIIRS NDVI

Observed

Vegetation condition

Every 10 days

GloFAS

Forecast

River discharge forecast

Daily

HydroRIVERS

Static geography

Riverine classification of every woreda

Fixed (doesn't change)

Open-Meteo

Forecast

14-day rainfall and temperature forecast

Daily

The Problem

Early warning science is strong — but three gaps keep it from reaching the people who need to act:

The last-mile gap

Forecasts rarely reach district health offices or health workers in a form they can act on — in Ethiopia, that means plain Somali.

The climate → health gap

A rainfall anomaly map doesn't tell a health officer to expect a cholera risk window in flood-hit kebeles.

The warning → action gap

Even when warnings arrive, nothing converts them into a protocol-based preparedness checklist.

How It Works

Four stages, each with an explicit human-auditability guarantee — rule citations and human approval are never skipped.

1. Interpret

The AI reasons over multiple observed Earth-observation signals and forecast data together — observed rainfall, vegetation condition, a drought indicator, and river discharge — and produces a graded hazard assessment: level, confidence, and its stated reasoning.EW4All Pillar 2 — Forecast Data

2. Anticipate

The assessment is checked against 15 vetted climate → health impact rules. Every anticipated risk must cite a matching rule — uncited output is rejected, not published.EW4All Pillar 1 — Risk Knowledge

3. Advise

The AI drafts advisory text — a WhatsApp message, radio script, and official memo — grounded only in the cited risks above, in Somali and English.EW4All Pillar 3 — Warning Dissemination

4. Approve

Nothing publishes automatically. A health officer or admin reviews and approves every advisory before it reaches a district.EW4All Pillar 4 — Preparedness & Response

Why You Can Trust It

"AI-native" doesn't mean unaccountable. Four non-negotiable rules keep the pipeline auditable end to end:

Grounding, not freestyling

Every anticipated health risk must cite a vetted impact rule, and its timing must fall inside that rule's range — unsupported output is rejected outright.

Structured validation

Every AI output is validated against a strict schema. Invalid output is retried once, then escalated to human review — never silently published.

Confidence + human-in-the-loop

Low-confidence or emergency-level assessments are automatically held back from public display until a human confirms them.

Full audit logging

Every AI call — prompt, model, response, validation outcome — is logged. That log is the methodology annex for due diligence.

Coverage

Earth-observation data layers are ingested and displayed for every district below. Hazard levels and advisories are issued only where national hazard rules have been agreed — today, that is Ethiopia alone.

Ethiopia

Admin unit

Woreda

100 in total

Operational

Kenya

Admin unit

County

Garissa, Tana River

Data live · rules in co-design

Uganda

Admin unit

District

Manafwa, Mbale

Data live · rules in co-design

South Africa

Admin unit

District Municipality

Vhembe, West Coast

Data live · rules in co-design

Built for Funders and Partners

Degdeg is built for a conversation with funders like UNICEF and Wellcome, and with national health authorities — a system that is accountable, auditable, and verifiable, not just another AI demo.

Read Our Methodology

Developed By

Degdeg is developed by the AI Innovation Lab at the Armauer Hansen Research Institute (AHRI), Ethiopia's national biomedical research institute. AHRI leads CHOICE — the Coalition for Health AI Innovation and Ethics, an Africa CDC initiative — and Lab leadership serves as an expert member of that coalition and sits on Ethiopia's National AI in Healthcare Programs Technical Working Group. Degdeg joins an existing portfolio that includes AI-PRESCRIBER, SUSTAIN-AI ECG, AI-HEALS AMR, and AI4GMP, alongside international collaborations with MI4People.

Guided by International Frameworks

EW4All WHO Climate Resilience Framework WMO Impact-based Forecasting WHO Training Module 5 INFORM Climate Change Sendai Framework National AI-Health TWG

See all frameworks →