GridRaven is proud to announce that the Estonian Environment Agency, which is responsible for weather forecasting in Estonia, have signed a cooperation agreement with us aimed at exploring the potential of improving wind prediction with the help of machine learning.
Machine learning for downscaling Numerical Weather Predictions
This year-long cooperation is centered around sharing ideas and knowledge. The Environment Agency provides expertize about meteorology and weather predictions while GridRaven will contribute its machine learning know-how and develop new prediction models.
Traditionally the most accurate weather forecasts are done by crunching physical equations in supercomputers. Usually these models operate at a spatial resolution of between 2 to 10 kilometers, whereby many simplifications have to be made. These simplifications enable the models to be computed within sufficient time and quality to be useful in operational forecasts.
The promise of machine learning for weather prediction is the ability to take existing forecasts as an input and then improve on them by “downscaling” the prediction for a specific location. The aim is to predict wind on a specific street, in a valley or behind trees. To achieve this, the machine learning model can be trained onwind measurements together with LiDAR and satellite imagery, so that the prediction model learns the physics of wind.
The expectation in the project is to significantly improve the accuracy of wind speed and direction forecasts in complex terrain. After the project, the resulting models can be applied in Estonia’s weather forecasts in general or in specific applications, such as road transport safety.
Year-long project to learn from each other
We at GridRaven are very happy about the opportunity to learn from our project partners at the Estonian Environment Agency, who are passionate and extremely knowledgeable about weather forecasting. We are looking forward to finding out what we can build based on our combined knowledge in machine learning and meteorology and are proud to investigate improving the quality of wind predictions together with them.
