Ignitia's deep dive into AI
Updated: Nov 26, 2019
Our future growth relies on innovation, it is what our company was built on. We continue to explore different frontiers of science and harness the power of technology to improve our renowned weather forecasts.
At Ignitia we are investigating the possibility of utilizing machine learning techniques to further improve the weather forecasts. Currently we are developing a model which will be used to nowcast thunderstorms over a period of three hours. The model is fed with historic data about thunderstorms and learns to predict a sequence of future positions for thunderstorm cells.
A huge amount of data is generated by sensors and satellites monitoring the weather everyday. This forms a perfect setting to explore how machine learning methods can be used to improve current weather forecasts. One area where the numerical methods usually have problems producing accurate forecasts has been that of thunderstorms. Thunderstorms typically have a short lifespan and can appear and disappear within minutes, which makes the forecasting of details almost impossible.
Recently, Ignitia’s R&D team has started developing a machine learning model to tackle this problem of nowcasting thunderstorms over a period of three hours (to begin with). Based on almost a decade of down-to-the-second timing and position of individual thunderstorms in the West Africa region, patterns are emerging that machine learning techniques are picking up and that complements our probabilistic ensemble forecasting system. Our goal is to create a model that will be able to learn/mirror the dynamics of the storm for a number of time steps into the future.
By combining scientific expertise from our meteorologists with modern machine learning methods we are starting to see promising results. Using only time and position of the thunderstorms, the model has been able to extract data on storms movement and change in shape after just a short period of assimilation. We are now just scratching the surface of possibilities and are further investigating additional features which could help explain the behaviour of the storms in terms of path, growth/decay and intensification.