Configuration and intercomparison of deep learning neural models for statistical downscaling

Abstract

Abstract. 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).

Statistical downscaling of daily temperature and precipitation over China using deep learning neural models: Localization and comparison with other methods

“…In addition, the use of ERI is consistent with previous work from Baño‐Medina et al . (2020), which facilitates the comparison of CNN model skills for the application over different large regions. In particular, ERI raw data at the temporal and spatial resolutions of 6 hr interval and 0.7° are downgraded into a common 2° grid on the daily scale following VALUE framework (Maraun et al ., 2015).…”

“…In this study, the CNN model follows the same architecture as Baño‐Medina et al . (2020). It contains three convolutional layers and no pooling layers with the filter size set to 3 × 3 × N. Large‐scale (36 × 21) daily predictors are treated as input layers to predict the fine‐resolution (10,862 land grids within 232 × 132 grids) daily temperature and precipitation.…”

“…Following Baño‐Medina et al . (2020), a series of CNN sensitivity experiments (i.e., CNN‐LM, CNN1, CNN10, CNN‐PR, and CNNdense) are conducted. In particular, the depths of three convolutional filters for CNN1 (as reference) are 50, 25, and 1 respectively with the use of rectified linear unit (ReLU) as activation function.…”

“…Among CNN models, CNNdense achieves the best skill for the improved performance over the region between two rivers, followed by CNN1. CNN‐PR performs even worse than CNN0, which reveals the fact that the increased neurons in the last hidden layers worsens the precipitation performance due to the over‐parametrization (Baño‐Medina et al ., 2020). For MAE, CNNdense yields lower median and mean lines and smaller IQR than ERI and other models, which again demonstrates its overall good performance.…”

A comparative study of convolutional neural network models for wind field downscaling

“…While the influence of model complexity has been examined by Baño-Medina et al (2019) in terms of model depth, that is, the number of convolution layers, the models in use did not exploit recent design patterns, like skip or residual connections (e.g., Srivastava et al, 2015;He et al, 2016) or the fully-convolutional U-Net-like architecture (Ronneberger et al, 2015), which enable network models to achieve state-of-the-art results in computer vision tasks.…”