ANR-21-CE04-0014-03

Objectives and research hypothesis

Providing accurate weather predictions is an important or even critical issue in many domains with potential high impact on the society (agriculture, transports, energy, health, insurance, tourism, …). Weather forecast is particularly challenging for an island with steep orography. Our main objective is to devise and to validate a set of tools and methods that provide high-precision forecasts of localized weather events in Corsica region with the specific goals of (i) predicting accurately (in time, in space and intensity) severe events in order to improve alert and monitoring systems for natural hazard prevention and (ii) anticipating and mitigating the production of solar power plants and wind turbine farms for the management of renewable energy facilities. The main research hypothesis supporting this project is that assisting a computational demanding Numerical Weather Prediction (NWP) model by a flux of local and regional data through a Machine Learning (ML) approach shows high potential for important benefits, as already suggested by many studies. Our “Sensor Augmented weather Prediction at HIgh Resolution” (SAPHIR) takes place within the fast growing and very promising context of data-driven weather forecasting methods.

SAPHIR project consists in exploiting the outputs of a high-resolution NWP system and a variety of (real-time or post-processed) data gathered from meteorological safety sensor networks, satellite observations and also from ad hoc local sensor networks through a “Deep Learning” architecture. We will focus more precisely on two different outcomes at some specific locations in Corsica, namely the prediction of episodes of intense thunderstorms with significant rainfalls and lightning activity and wind/solar power forecasting (see Fig. 1 below). Indeed, severe weather conditions and notably thunderstorms and flooding are among one of the most dangerous weather risks in the Mediterranean region. Even if warning systems based on operational NWP forecasts provide reliable recommendations, accurate predictions at high resolution of the occurrence, location, timing and intensity of extreme events remain very challenging especially in a complex mountainous and maritime region such as the island of Corsica. This issue is especially critical since the global warming notably increases the occurrence of extreme weather events. In the field of renewable energy (RE), the power delivery of wind farms and solar power plants is strongly dependent on the volatile energy resource and its intermittency. As a consequence, it is characterized by a large amount of uncertainty that may be costly and difficult to manage for energy suppliers. However, the share of electricity produced by photovoltaics and wind energy systems increases and according to France Multi Annual Energy Plan, the installed capacity for renewable electricity in 2028 will be doubled as compared to 2017. Then, accurate wind speed and solar irradiance predictions and thereby forecast of power delivery are more and more important for the energy managers in order to make these intermittent energy sources amenable to integration into a power grid. Our ambition in that respect will be (i) to study the possibility of improving current predictions at horizons from few tens of minutes to 1 day, (ii) to consider microgrid flow control issues associated with such predictions and (iii) to study the possibility to predict extreme wind events in order to mitigate their impact on energy production.

Figure 1 - SAPHIR objectives and platform architecture. Top down view of data and computation goals.