Speaker
Description
High-resolution simulation models (of order kms or less) can deliver highly accurate low-level winds and their climatologies. The problem is that one cannot afford to run simulations at very high resolution over extended spatial domains for long periods because the computational power needed is prohibitive.
Instead, we propose using neural networks to downscale low-resolution wind-field simulations (input) to high-resolution fields (targets) to try to match a high-resolution simulation. Based on short-range wind-field forecasts (at the 100m level) from the ECMWF ERA5 reanalysis, at 31km resolution, and HRES (deterministic) short-range forecasts at 9km resolution, we conduct an initial 'proof-of-concept' study using two complementary modelling approaches.
In a first step, we evaluate the ability of U-Net-type convolutional neural networks to learn a one-to-one mapping of low-resolution input data to high-resolution simulation results. By creating a compressed feature-space representation of the data, networks of this kind manage to encode important flow characteristics of the input fields and assimilate information from additional data sources. Next to wind vector fields, we use topographic information to inform the network, and include additional parameters that strongly influence wind-field prediction in simulations, such as boundary layer height (as a metric of vertical stability) and the land-sea mask. The network considers also the high-resolution topography to infer situation- and location-dependent wind structures that could not be retrieved otherwise.
In typical situations, however, it may be inappropriate to consider only a single estimate for the high-resolution wind field. Especially in regions where strong variations in orography foster the emergence of complex wind patterns, a variety of different high-resolution estimates may be equally compatible with the low-resolution input and physical reasoning. In a second step, we therefore extend the learning task from optimizing deterministic one-to-one mappings to modelling the distribution of physically reasonable high-resolution wind vector fields, conditioned on the given low-resolution input. Using the framework of conditional variational autoencoders, we realize a generative convolutional neural network model, which is able to learn the conditional distributions from data. Sampling multiple estimates of the high-resolution wind vector fields from the model enables us to explore multimodalities and uncertainties of the model output.
In a future customer-oriented extension of this proof-of-concept work, we envisage using a target resolution down to 1-4km to deliver much better representivity for users.
