Capturing Climatic Variability: Using Deep Learning for Stochastic Downscaling

TL;DR

The paper tackles the problem of local-scale climate downscaling under nonstationarity by advocating stochastic sampling from a high-resolution distribution conditioned on low-resolution inputs. It introduces a Wasserstein conditional GAN with latent-space noise injection, dual covariate streams, and probabilistic loss (CRPS), along with two training strategies (frequency separation and stochastic sampling) to improve calibration. On synthetic data, noise injection markedly improves marginal and conditional distributions; in a realistic wind downscaling task, combining noise injection with stochastic training and CRPS loss (S_full_CRPS) yields the best calibration and extreme-value representation. This approach offers a computationally efficient path to better quantify uncertainty and extremes in local climate projections, aiding adaptation planning, though performance can be challenged by spatial heterogeneity and nonstationarity.

Abstract

Adapting to the changing climate requires accurate local climate information, a computationally challenging problem. Recent studies have used Generative Adversarial Networks (GANs), a type of deep learning, to learn complex distributions and downscale climate variables efficiently. Capturing variability while downscaling is crucial for estimating uncertainty and characterising extreme events - critical information for climate adaptation. Since downscaling is an undetermined problem, many fine-scale states are physically consistent with the coarse-resolution state. To quantify this ill-posed problem, downscaling techniques should be stochastic, able to sample realisations from a high-resolution distribution conditioned on low-resolution input. Previous stochastic downscaling attempts have found substantial underdispersion, with models failing to represent the full distribution. We propose approaches to improve the stochastic calibration of GANs in three ways: a) injecting noise inside the network, b) adjusting the training process to explicitly account for the stochasticity, and c) using a probabilistic loss metric. We tested our models first on a synthetic dataset with known distributional properties, and then on a realistic downscaling scenario, predicting high-resolution wind components from low-resolution climate covariates. Injecting noise, on its own, substantially improved the quality of conditional and full distributions in tests with synthetic data, but performed less well for wind field downscaling, where models remained underdispersed. For wind downscaling, we found that adjusting the training method and including the probabilistic loss improved calibration. The best model, with all three changes, showed much improved skill at capturing the full variability of the high-resolution distribution and thus at characterising extremes.

Figures (11)

• Figure 1: Architecture of GAN networks showing Residual in Residual Dense Block (RRDB) with noise injection. Green denotes locations where noise is added into the network. Rectified Linear Units (ReLU) are used to introduce non-linearity.
• Figure 2: a) Kernel density estimates (KDEs) of marginal distributions of $p(HR|LR)$ for the unimodal synthetic dataset for one example pixel (i = 5, j = 5) for the true distribution and generated distributions. KDEs are based on 500 realisations for a single conditioning field for each distribution. Dashed line shows true marginal distribution. b) Violin plot showing KS statistic values comparing generated marginal conditional distributions to ground truth distributions for all pixels. Statistics are calculated for each pixel individually, using 500 realisations of a single conditioning field. Lines show 0.25, 0.5, and 0.75 quantiles, respectively. c) CDF of rank histogram on unimodal synthetic data, with four models, showing calibration of conditional distributions. Dashed line shows reference uniform distribution. Rank histograms were calculated across 100 randomly selected conditioning fields, with 96 HR realisations generated for each. d) KDEs of marginal conditional distributions for one example pixel of a bimodal dataset, comparing true (dashed line) and generated distributions. Distributions were estimated using the same approach as in a).
• Figure 3: Spatial fields of median and 99.9 percentiles of the full distributions across samples for ground truth, and generated data from two models, using the unimodal synthetic dataset (equation 3).
• Figure 4: Radially averaged spectral power (RASP) for four models. Values are standardised to amplitudes of ground truth wavenumbers, so perfectly matched spectral power occurs at one. Solid lines and shaded regions respectively show mean and +/- one standard deviations across 1200 randomly selected samples. Dashed line indicates wavenumber corresponding to LR pixel size.
• Figure 5: CDFs of rank histograms for meridional wind components, using four different models. Rank histograms were calculated across 100 randomly selected conditioning fields, with 96 HR realisations Generator for each. Dashed line shows reference uniform CDF.