evidoria

← Back to browse

Good practice Imported

AI for Early Warnings for All — Malawi's Pilot to Integrate the 'Bris' AI Weather Model into National Forecasting

Malawi · Blantyre · See the Malawi profile

Evidence: Observational / pre–post Top 75% 47/100 · Ask Evidence Copilot about this practice

Malawi's meteorological service piloted MET Norway's 'Bris' AI weather model with WMO, cutting forecast generation from hours to ~10 minutes — but researchers caution the system is 'not at the stage' for standalone public warnings, with no accuracy gains yet measured.

245,775 USD (CHF 245,000)
Pilot budget (May-Oct 2025)
1,751 staff-hours
MET Norway technical engagement (6-month pilot)
~10 minutes
5-day forecast generation time (AI model)
<20 minutes
10-day forecast generation time (AI model)
AI for Early Warnings for All — Malawi's Pilot to Integrate the 'Bris' AI Weather Model into National Forecasting

Details

Maturity
Pilot
Promoter
Department of Climate Change and Meteorological Services (DCCMS), Malawi, with the World Meteorological Organization (WMO) and the Norwegian Meteorological Institute (MET Norway)
Period
May–October 2025 pilot (CREWS-funded; evaluation coverage into 2026)
Keywords
weather forecasting, artificial intelligence, meteorology, early warning systems, capacity building

Context

Between May and October 2025, Malawi's Department of Climate Change and Meteorological Services (DCCMS) ran a CREWS-funded pilot, executed by the World Meteorological Organization (WMO), to test integrating "Bris", an AI weather model built by the Norwegian Meteorological Institute (MET Norway) and delivered via ECMWF's "forecast-in-a-box" framework, into Malawi's operational forecasting workflow on modest local hardware.

Results

The roughly USD 245,775 (CHF 245,000) budget covered about 1,751 staff-hours of MET Norway technical engagement plus DCCMS staff time, travel and equipment over the six-month, four-phase pilot. Independent reporting found the AI model generated a 5-day forecast in about 10 minutes and a 10-day forecast in under 20 minutes on standard hardware, versus several hours for a comparable traditional numerical weather prediction run.

Conclusions

MET Norway researcher Lene Østvand said "we are not at the stage where this could be used on its own for public warnings" and cautioned that AI models can be wrong outside the conditions they were trained for; the pilot's chief output was a roadmap for regional scaling and staff training rather than a measured improvement in forecast accuracy or warning lead time, and the initiative remains dependent on continued external technical support from Norway.

Implementation
Indicative cost
Low (< €50k) — Approximately USD 245,775 (CHF 245,000) CREWS-funded budget covering technical engagement, staff time, travel and equipment over six months.
Time to results
Short (< 1 year) — Pilot ran May-October 2025 in four phases, timed to report to WMO's Extraordinary Congress in October 2025; evaluation coverage continuing into 2026.
Staffing & skills
Department of Climate Change and Meteorological Services (DCCMS), Malawi, implements the pilot, World Meteorological Organization (WMO) is the executing agency, MET Norway (Norwegian Meteorological Institute) is the technical partner providing the Bris model and staff engagement

Conditions for success

  • CREWS funding provided a clear, itemised budget for technical engagement, staff time, travel and equipment
  • ECMWF's 'forecast-in-a-box' framework allowed the model to run on modest local hardware
  • Formal reporting structure to WMO's Extraordinary Congress gave the pilot institutional visibility

Common failure modes

  • No forecast-accuracy or warning-lead-time gains have yet been measured, only compute-speed benchmarks
  • MET Norway researchers caution the model is 'not at the stage' for standalone public warnings and can be wrong outside its training conditions
  • Continued dependence on external technical support from Norway is an acknowledged sustainability concern for local ownership

Where it fits

Governance type
national meteorological agency with international technical partners (WMO, MET Norway)
Scale
national pilot
Income level
low income

Commonly funded by

National / regional programmes

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful