HomeAIAI Science & HealthECMWF Starts Forecast-in-a-Box Weather Pilot in Malawi

ECMWF Starts Forecast-in-a-Box Weather Pilot in Malawi

The European Centre for Medium-Range Weather Forecasts (ECMWF) is piloting Forecast-in-a-Box, a portable software package that lets AI weather forecasts run on desktop-like machines, with the Norwegian Meteorological Institute (MET Norway) and Malawi’s Department of Climate Change and Meteorological Services (DCCMS). In a 19 November 2026 news release, ECMWF describes a WMO-endorsed pilot in Malawi that is testing how the system could support timely and accessible forecasts and decision-making in extreme weather. The partners have not published accuracy figures for the pilot, and the project is not being presented as an operational forecast service. A 12 November account on MET Norway’s site, bylined Lucy Stone of ECMWF, had already set out the same collaboration.

What the Malawi Forecast-in-a-Box pilot is testing

The software is designed so that public institutions such as national meteorological services, and industry users, can run AI forecasts on their own infrastructure. ECMWF is careful about what the package replaces and what it leaves in place. High-performance computing, it says, remains central to generating the underlying initial conditions, training AI-based models, and running physics-based numerical weather prediction. The box is a portable option for running pre-trained machine learning models. It is being tested in Malawi and trialled in the European Commission’s Destination Earth initiative.

Through the system, the Malawian team can run a variant of MET Norway’s AI weather model Bris, which builds on ECMWF’s Artificial Intelligence Forecasting System (AIFS) and, ECMWF says, is specifically tuned for Africa with local expertise. The package also includes lightweight models for trying the system on a laptop, models described as operational from ECMWF, MET Norway and the Finnish Meteorological Institute, and experimental models from the research community. Those models have been trained using the open-source Anemoi framework. ECMWF says the project itself is open source, funded by ECMWF and by the European Commission through DestinE and the Africa Regional Centres of Excellence (ArcX) programme, and can be downloaded from ECMWF’s GitHub.

The Malawi pilot builds on cooperation between MET Norway and DCCMS that began in 2019, including training in meteorological analysis systems and a weather app based on data from Yr. The system was deployed in Malawi for the first time last year and was completely redesigned in 2026, leading to what ECMWF calls the upcoming second stable release presented in Malawi.

Droughts, floods and tropical cyclones, ECMWF notes, threaten public health and food security in a country where the agricultural sector employs 80% of the workforce. Lucy Mtilatila, Director of DCCMS, said: “We are proud to be piloting and testing AI in weather forecasting in Malawi, and we look forward to implementing the Forecast-in-a-Box and Bris forecasting model to operationalize and integrate it into our meteorology and climate services. This collaboration, which ensures our involvement throughout the entire process, means that the capacity and knowledge remain with us, and we greatly appreciate that.”

Florian Pappenberger, Director-General of ECMWF, said: “A portable forecasting system such as Forecast-in-a-Box literally puts forecasting into the hands of those who need it. The system puts users in control, strengthens their sovereignty and enables them to generate the products they need.” Lene Østvand, researcher and project manager for MET Norway’s capacity-building project in Malawi, said: “With the latest improvements we think the model is good enough for this pilot.” She added that adjustments would be needed and that a pilot gives feedback to improve Bris.

MET Norway’s account says the goal is to reach implementation within a year’s time. In addition to Malawi, the system will be made available to several African countries through ArcX, and MET Norway is also collaborating with Ethiopia, Mozambique and Tanzania. ECMWF says it will also be deployed on the Destination Earth platform in 2027, where DestinE users will be able to test a collection of AI models. Malawi is among the countries working towards the UN Early Warnings for All goal of universal early-warning coverage by the end of 2027.

The collaboration sits in a stretch of weather-AI work TSN has already covered, including ECMWF’s prototype for making AI-DOP forecasts straight from observations, DeepMind’s release of the PARDIG and SatWeave weather datasets, and an MIT and Microsoft study on machine learning for forecasts two to six weeks out.

A portable box is still a pilot

It is tempting to read the product-like name as a finished appliance that has already been shown to outperform a national forecast office. That is not what the partners have announced. ECMWF’s own wording is about a pilot, a trial and an upcoming second stable release. Østvand’s “good enough for this pilot” is a qualitative judgement, not a published accuracy score, and she is explicit that further adjustments are expected. Mtilatila looks forward to implementing the system and integrating it into DCCMS services; that is a goal, not a declaration that operations have already switched over.

Two other limits should stay in view. First, the box runs pre-trained models. On ECMWF’s account, high-performance computing remains central to creating initial conditions and to training the models in the first place. Portability helps a meteorological service generate products locally once those models exist. It is not a substitute for the upstream infrastructure. Second, the further roll-out — ArcX availability in other African countries, a Destination Earth platform deployment in 2027, and MET Norway’s work with Ethiopia, Mozambique and Tanzania — is described as planned, not as completed.

Roar Skålin, Director-General of MET Norway, put the open question plainly: “The question is when these models and tools will become operational, and what services they will provide.” Until that question is answered with an operational service and with published verification, the Malawi project should be read as a capacity-building pilot in a WMO-endorsed effort, open to download and useful as a test of local AI forecasting, not as a scored upgrade to the national forecast.

Sources

Related stories

Follow @tsnmediahub on X for breaking AI news as it happens.

Follow on X

Share this story

Latest stories

  1. House Members Release an Outline of the SAFE AI Act
    AI Policy & Safety

More in this category

Latest stories

  1. House Members Release an Outline of the SAFE AI Act
    AI Policy & Safety

Explore the TSN Atlas: Every topic we cover, mapped. Click a bubble to dive in.