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 (\(\lambda_{adv}\)), where the value of  \(\lambda_{adv}\) 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 \(\lambda_{adv}\) 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.

This work presents a new approach to defining drought, establishing an empirical relationship between historical droughts (and wet spells) documented in impact reports, and a broad range of observed drought-related climate features. A Random Forest (RF) algorithm was trained to identify the particular combinations of predictors – such as precipitation, soil moisture and potential evapotranspiration – that led to categorical, documented drought or non-drought events. Unlike traditional drought definitions, the new RF drought indicator combines meteorological, hydrological, agricultural, and socioeconomic drought, providing drought information for all impacted sectors. The metric also quantifies the conditional probability of drought (rather than being threshold-based), considering multiple climate features and their interactive effect, and can be used for forecasting. The approach was validated out-of-sample across several random selections of training and testing datasets, and demonstrated better predictive capabilities than commonly used drought indicators in a range of performance metrics. Furthermore, it showed a comparable performance to the (expert elicitation-based) US Drought Monitor (USDM) which is the current state-of-the-art record of historical drought in the USA. As well as providing an alternative historical drought indicator to USDM, the RF approach offers additional advantages by being automated, by providing drought information at the grid-scale, and by having predictive capacity. As a proof-of-concept case, the RF drought indicator was trained on Texan climate data and droughts, and validated in all Texas ecoregions. However, the introduced approach can be easily implemented to develop a RF drought indicator for new regions if adequate information on historical droughts is available.