We developed a new version of our unique mLM, which brings further forecast accuracy improvements for temperature, wind speed and dewpoint temperature.
The meteoblue Learning MultiModel (mLM) is our unique post-processing technique, which trains the output from numerical weather prediction with measurement data. It uses AI (Artificial Intelligence) - since 2018, long before the entire world (and Basel Fasnacht) started to talk about AI.
mLM uses current weather measurement data and 10-20 different weather models to select and weight the model with the most accurate simulation for the local downscaled forecast.
The new mLM version was developed over several months by adding computation power and mathematics. This version now delivers even better and more complete probability distributions for each forecast hour and station. This data is then used to calculate temperature, dewpoint temperature and wind speed predictions. It also enables us to continue adding many more weather stations, as we did recently for France. In addition, the number of outliers was significantly reduced. During development, we have carefully validated the new mLM version, resulting in an accuracy improvement of more than 5% compared to the previous forecast data, making this the weather forecast with the highest publicly documented accuracy worldwide. We will soon publish a detailed validation study on our verification pages.
The mLM uses measurement data as input for model training. Consequently, we also improved the quality of the input measurement data by implementing various new quality control algorithms, ensuring a high quality of the measurement data.
We hope that you notice the forecast improvement in your daily operations or systems, and will continue to work on improvements based on your feedback.
Share
Comments
Will we be able to benefit from this AI model? If for example we own a personal weather station, could the precise data from our exact location factor in the model so that we can get more accurate prediction for this specific area?
If we have a weather station, could the precise data from our exact location factor in the model so that we can get more accurate prediction ?
Yes, the updated model data is available on our website and app. There is also the option to register your own weather station here: http://wxqa.com/SIGN-UP.html
The data from this site are also used in our forecast processing chain.
Interesting! Now, as you've just mentioned your system was already predicting and analyzing the data with pretty darn accuracy in my eyes and for my usage.. Towards what effects will AI bring to the table to show it'll be even more precise against what your system actually is doing?? Will I or the end users be able to see any difference between the AI and what you currently have?? Last, will this have an increase or decrease in your personnel staffing?
We are currently working on an article that explains how AI can help in weather forecasting and where the limitations are. Since we are continuously working on new innovations, we need staff to do this.
Write comment
You need a meteoblue account to comment on articles
