Microsoft Releases PBC, an Open-Source Weather Bias Corrector

⚡

Key Takeaways

1Microsoft releases the Probabilistic Bias Correction as open-source, an AI method to correct biases in existing weather models

2Researchers report improvements on physical and AI models from ECMWF, with quantified gains

3The tool aims to make medium-term forecasts more accessible, especially for resource-limited agencies and in developing countries

4A collaboration in Kenya, funded by the Gates Foundation, explores its usefulness for early warnings

💡Why it matters — Making weather forecasts more accurate and affordable could enhance planning for farmers, emergency services, and countries with fewer resources.

Microsoft has made available a machine learning method aimed at correcting biases in weather models. Researchers report notable gains on ECMWF systems, hoping to make medium-term forecasts more accessible to resource-limited stakeholders.

Announced Gains on ECMWF Models

Researchers indicate that the Probabilistic Bias Correction improves the accuracy of an ECMWF physical model for 98% of precipitation targets and 92% of temperature targets. They also claim that the framework has doubled the accuracy of the new AI forecasting system from ECMWF in the conducted analysis. These results come in a context where forecasting beyond two weeks remains challenging for both physical approaches and AI, and where approximations can introduce biases that penalize longer-term horizons. Microsoft researchers assert that their method is a cost-effective solution to improve the accuracy of forecasts between two and six weeks in advance.

Accessibility and Target Uses in Resource-Limited Countries

The authors of the study believe that PBC could transform the production of weather forecasts, particularly in developing countries where access to advanced technologies is often limited. Genevieve Flaspohler, co-author, emphasizes the importance of lightweight and easy-to-train models to enable scientists with limited resources to improve them. Researchers hope that this approach will be adopted by meteorological agencies with tight budgets. According to them, making it available as open-source could also reduce the cost of AI-enhanced forecasts, with potential applications for farmers and emergency services.

⚡Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisis ton rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

PBC Corrects Model Biases in Light of Data Limitations

Probabilistic Bias Correction relies on machine learning to identify and correct systematic biases in the outputs of weather models. It can incorporate the effects of climate change into the adjustments. For example, as greenhouse gas emissions warm the planet, models based on historical data may remain biased towards lower temperatures than those observed today. Physical models, like those based on AI, rely on approximations to compensate for data shortages in certain regions or the complexity of poorly understood phenomena, which can introduce systematic biases.

Publication, Openness, and Initial Collaborations

The research results have been published in the journal Nature, and the tool is available as open-source. Microsoft joins Google and Nvidia, which are also developing AI tools for weather forecasting, but here chooses to correct existing models rather than create a new one. Genevieve Flaspohler, who worked on PBC as an intern at Microsoft and now leads Rhiza Research, is collaborating with the Kenya Meteorological Authority to assess whether AI can improve the accuracy of early warnings. This project, funded by the Gates Foundation, aims to anticipate a more extreme rainy season during a record El Niño.