Search

Deep learning techniques (in particular convolutional neural networks, CNNs) have recently emerged as a promising approach for statistical downscaling due to their ability to learn spatial features from huge spatiotemporal datasets. However, existing studies are based on complex models, applied to particular case studies and using simple validation frameworks, which makes a proper assessment of the (possible) added value offered by these techniques difficult. As a result, these models are usually seen as black boxes, generating distrust among the climate community, particularly in climate change applications. In this paper we undertake a comprehensive assessment of deep learning techniques for continental-scale statistical downscaling, building on the VALUE validation framework. In particular, different CNN models of increasing complexity are applied to downscale temperature and precipitation over Europe, comparing them with a few standard benchmark methods from VALUE (linear and generalized linear models) which have been traditionally used for this purpose. Besides analyzing the adequacy of different components and topologies, we also focus on their extrapolation capability, a critical point for their potential application in climate change studies. To do this, we use a warm test period as a surrogate for possible future climate conditions. Our results show that, while the added value of CNNs is mostly limited to the reproduction of extremes for temperature, these techniques do outperform the classic ones in the case of precipitation for most aspects considered. This overall good performance, together with the fact that they can be suitably applied to large regions (e.g., continents) without worrying about the spatial features being considered as predictors, can foster the use of statistical approaches in international initiatives such as Coordinated Regional Climate Downscaling Experiment (CORDEX). [ABSTRACT FROM AUTHOR]

Many climate extremes, including heatwaves and heavy precipitation events, are projected to worsen under climate change, with important impacts for society. Future projections required for adaptation are often based on climate model simulations. Given finite resources, trade-offs must be made concerning model resolution, ensemble size, and level of model complexity. Here we focus on the resolution component. A given resolution can be achieved over a region using either global climate models (GCMs) or at lower cost using regional climate models (RCMs) that dynamically downscale coarser GCMs. Both approaches to increasing resolution may better capture small-scale processes and features (downscaling effect), but increased GCM resolution may also improve the representation of the large-scale atmospheric circulation (upscaling effect). The size of this upscaling effect is therefore important for deciding modelling strategies. Here we evaluate the benefits of increased model resolution for both global and regional climate models for simulating temperature, precipitation, and wind extremes over Europe at resolutions that could currently be realistically used for coordinated sets of climate projections at the pan-European scale. First we examine the benefits of regional downscaling by comparing EURO-CORDEX simulations at 12.5 and 50 km resolution to their coarser CMIP5 driving simulations. Secondly, we compare global-scale HadGEM3-A simulations at three resolutions (130, 60, and 25 km). Finally, we separate out resolution-dependent differences for HadGEM3-A into downscaling and upscaling components using a circulation analogue technique. Results suggest limited benefits of increased resolution for heatwaves, except in reducing hot biases over mountainous regions. Precipitation extremes are sensitive to resolution, particularly over complex orography, with larger totals and heavier tails of the distribution at higher resolution, particularly in the CORDEX vs. CMIP5 analysis. CMIP5 models underestimate precipitation extremes, whilst CORDEX simulations overestimate compared to E-OBS, particularly at 12.5 km, but results are sensitive to the observational dataset used, with the MESAN reanalysis giving higher totals and heavier tails than E-OBS. Wind extremes are somewhat stronger and heavier tailed at higher resolution, except in coastal regions where large coastal grid boxes spread strong ocean winds further over land. The circulation analogue analysis suggests that differences with resolution for the HadGEM3-A GCM are primarily due to downscaling effects. [ABSTRACT FROM AUTHOR]