10 August 2021 in Communications Earth & Environment
A barrier to utilizing machine learning in seasonal forecasting applications is the limited sample size of observational data for model training. To circumvent this issue, here we explore the feasibility of training various machine learning approaches on a large climate model ensemble, providing a long training set with physically consistent model realizations. After training on thousands of seasons of climate model simulations, the machine learning models are tested for producing seasonal forecasts across the historical observational period (1980-2020). For forecasting large-scale spatial patterns of precipitation across the western United States, here we show that these machine learning-based models are capable of competing with or outperforming existing dynamical models from the North American Multi Model Ensemble. We further show that this approach need not be considered a ‘black box’ by utilizing machine learning interpretability methods to identify the relevant physical processes that lead to prediction skill.
14 April 2017 in Journal of Geophysical Research: Atmospheres
8 November 2022 in Weather and Climate Extremes
2 January 2025 in Journal of Advances in Modeling Earth Systems
Anticipating climate impacts and risks in present or future climates requires predicting the statistics of high‐impact weather events at fine‐scales. Direct numerical simulations of fine‐scale weather are computationally too expensive for many applications. While deterministic‐based (deep‐learning or statistical) downscaling of low‐resolution climate simulations are several orders of magnitude faster than direct numerical simulations, it suffers from several limitations. These limitations include the tendency to regress to the mean, which produces excessively smooth predictions and underestimates the magnitude of extreme events. They also fail to preserve statistical measures that are key for climate research. We use a conditional GAN (cGAN) architecture to downscale daily precipitation as a Regional Climate Model (RCM) emulator. The cGAN generates plausible residuals on top of the predictable expectation state produced by a deterministic deep learning algorithm. The skill of cGANs is highly sensitive to a hyperparameter known as the weight of the adversarial loss (), where the value of required for accurate results varies with season and performance metric, casting doubt on the reliability of cGANs as usually implemented. However, by applying a simple intensity constraint to the loss function, it is possible to obtain reliable performance results across spanning two orders of magnitude. CGANs are considerably more skillful in capturing climatological statistics, including the distribution and spatial characteristics of extreme events. With this modification, we expect cGANs to be readily transferable to other applications and time periods, making them a useful weather generator for representing extreme event statistics in present and future climates.
1 July 2019 in Journal of Hydrometeorology
Climate model evaluation is complicated by the presence of observational uncertainty. In this study we analyze daily precipitation indices and compare multiple gridded observational and reanalysis products with regional climate models (RCMs) from the North American component of the Coordinated Regional Climate Downscaling Experiment (NA-CORDEX) multimodel ensemble. In the context of model evaluation, observational product differences across the contiguous United States (CONUS) are also deemed nontrivial for some indices, especially for annual counts of consecutive wet days and for heavy precipitation indices. Multidimensional scaling (MDS) is used to directly include this observational spread into the model evaluation procedure, enabling visualization and interpretation of model differences relative to a “cloud” of observational uncertainty. Applying MDS to the evaluation of NA-CORDEX RCMs reveals situations of added value from dynamical downscaling, situations of degraded performance from dynamical downscaling, and the sensitivity of model performance to model resolution. On precipitation days, higher-resolution RCMs typically simulate higher mean and extreme precipitation rates than their lower-resolution pairs, sometimes improving model fidelity with observations. These results document the model spread and biases in daily precipitation extremes across the full NA-CORDEX model ensemble. The often-large divergence between in situ observations, satellite data, and reanalysis, shown here for CONUS, is especially relevant for data-sparse regions of the globe where satellite and reanalysis products are extensively relied upon. This highlights the need to carefully consider multiple observational products when evaluating climate models.
28 March 2017 in Environmental Research Letters