The objective of WP2 is to develop weather predictions with high resolution and lower errors than the state of the art. To get more insights on the work carried out, the WP Leader, Météo-France, as well as its contributing partners, Whiffle and DLR, have been interviewed.

Q1. How do you proceed to decrease errors associated with Numerical Weather Predictions (NWP) and validate their performance in RES-related applications?

Multiple approaches can be used to improve NWP. In Smart4RES, our main strategy is to increase the spatial resolution of the model, typically of a factor of 2, with respect to the current operational configuration. This allows to better represent small-scale physical processes such as convection, cloud formation and turbulence.
We also increase the temporal resolution of the outputs, to capture the high frequency variations of the atmosphere, which is crucial for energy applications that rely on highly fluctuating variables such as wind and solar radiation.
In addition to these resolution refinements, the number of members used in the ensemble simulations is nearly doubled. This allows to better explore the range of possible atmospheric states and to better quantify the uncertainty associated with a forecast.
In parallel, we work on the improvement of the physical parametrisations used in the NWP model, for instance through a better representation of the interactions between clouds and solar radiation.
A final strategy is to assimilate new observations in the model, including energy production, as will be mostly explored for very high resolution simulations.
As for the validation of the predictions, standard scores exist to rate models on the basis of standard meteorological measurements.
Within Smart4RES, additional evaluations will be performed by comparing energy diagnostics from the model to actual power production from wind and solar power plants.

Q2. Can you summarize the innovative contributions of your high-resolution predictions, which address multiple scales from a single power plant to several European countries?

The major innovation will be to extract consistent, useful and relevant information for the final users from the huge amount of information contained in these unique high-resolution simulations.
This implies building seamless forecasts by combining distinct models that have different spatial resolutions and time horizons.
This also means reconstructing pseudo-deterministic simulations by combining several members of an ensemble simulation, in order to pass tractable forecasts in a standard format to our partners.
For very local applications, dedicated large-eddy simulations (LES) will also be run at very high spatial resolution (~100 m), which is a very novel approach in NWP. LES explicitly resolve clouds, as well as the most energetic scales of the turbulence, providing access to details absent from standard NWP. Such simulations can for instance account for the impact of wind turbines on the wind field. They can also simulate the impact of individual cloud shadows on the available solar energy at the surface.
Another innovation that will be investigated in Smart4RES is to assimilate local observations in the LES model. While data assimilation is a standard practice in traditional numerical weather prediction models, it is challenging to do this in a turbulence resolving model like LES.

Q3. Apart from numerical simulations, how can observations be used to improve the forecasts?

Observations are first used to define the initial state of the atmosphere, the evolution of which is then predicted by an NWP model. But they can also be used standalone, in particular for short-term forecasts, when standard NWP are generally less reliable.
For solar energy applications for instance, both observations from the ground and from satellite are used, mostly to predict the displacement of clouds and their impact on solar radiation.
Within Smart4RES, very innovative tools are being developed to estimate high resolution spatial fields (on meter resolution) of solar irradiance from combined observations of full-sky imagers.
These products provide dynamical maps of irradiance that can be used to predict at very high spatial resolution infra-hourly energy production. Even more innovative is the combination of these high-resolution irradiance maps with LES, satellite and NWP models. This new approach to data assimilation combines the best of the observations with the best of the physical models to produce optimized temporal and spatial irradiance maps and forecasts. Such detail can be critical for the optimization and grid management of a solar power plant.