Preparing for environmental risks requires estimating the frequencies of extreme events, often from data records that are too short to confirm them directly. This requires fitting a statistical distribution to the data. To improve precision, investigators often pool data from neighboring sites into single samples, referred to as “superstations,” before fitting. We demonstrate that this technique can introduce unexpected biases in typical situations, using wind and rainfall extremes as case studies. When the combined locations have even small differences in the underlying statistics, the regionalization approach gives a fit that may tend toward the highest levels suggested by any of the individual sites. This bias may be large or small compared to the sampling error, for realistic record lengths, depending on the distribution of the quantity analysed. The results of this analysis indicate that previous analyses could potentially have overestimated the likelihood of extreme events arising from natural weather variability.

Global climate change is often thought of as a steady and approximately predictable physical response to increasing forcings, which then requires commensurate adaptation. But adaptation has practical, cultural and biological limits, and climate change may pose unanticipated global hazards, sudden changes or other surprises, as may societal adaptation and mitigation responses. We outline a strategy for better accommodating these challenges by making climate science more integrative, in order to identify and quantify known and novel physical risks even–or especially–when they are highly uncertain, and to explore risks and opportunities associated with mitigation and adaptation responses by engaging across disciplines. This improves the chances of anticipating potential surprises and identifying and communicating “safe landing” pathways that meet UN Sustainable Development Goals and guide humanity toward a better future.

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.

Two assumptions commonly applied in convection schemes—the diagnostic and quasi-equilibrium assumptions—imply that convective activity (e.g., convective precipitation) is controlled only by the large-scale (macrostate) environment at the time. In contrast, numerical experiments indicate a “memory” or dependence of convection also on its own previous activity whereby subgrid-scale (microstate) structures boost but are also boosted by convection. In this study we investigated this memory by comparing single-column model behavior in two idealized tests previously executed by a cloud-resolving model (CRM). Conventional convection schemes that employ the diagnostic assumption fail to reproduce the CRM behavior. The memory-capable org and LMDZ cold pool schemes partially capture the behavior, but fail to fully exhibit the strong reinforcing feedbacks implied by the CRM. Analysis of this failure suggests that it is because the CRM supports a linear (or superlinear) dependence of the subgrid structure growth rate on the precipitation rate, while the org scheme assumes a sublinear dependence. Among varying versions of the org scheme, the growth rate of the org variable representing subgrid structure is strongly associated with memory strength. These results demonstrate the importance of parameterizing convective memory, and the ability of idealized tests to reveal shortcomings of convection schemes and constrain model structural assumptions.

Convection is usually parameterized in global climate models, and there are often large discrepancies between results obtained with different convection schemes. Conventional methods of comparing convection schemes using observational cases or directly in 3D models do not always clearly identify parameterization strengths and weaknesses. In this paper we evaluate the response of parameterizations to various perturbations rather than their behavior under particular strong forcing. We use the linear response function method proposed by Kuang (2010) to compare twelve physical packages in five atmospheric models using single-column model (SCM) simulations under idealized radiative-convective equilibrium conditions. The models are forced with anomalous temperature and moisture tendencies. The temperature and moisture departures from equilibrium are compared with published results from a cloud-resolving model (CRM). Results show that the procedure is capable of isolating the behavior of a convection scheme from other physics schemes. We identify areas of agreement but also substantial differences between convection schemes, some of which can be related to scheme design. Some aspects of the model linear responses are related to their RCE profiles (the relative humidity profile in particular), while others constitute independent diagnostics. All the SCMs show irregularities or discontinuities in behavior that are likely related to switches or thresholds built into the convection schemes, and which do not appear in the CRM. Our results highlight potential flaws in convection schemes and suggest possible new directions to explore for parameterization evaluation.

Key Points: • The concept of precipitation efficiency is broad, and can be related to many proposed cloud feedback mechanisms • Microphysical precipitation efficiency of tropical clouds likely increases with warming, but bulk precipitation efficiency and precipitation efficiency of midlatitude clouds could decrease • The impacts of precipitation efficiency on clouds and feedbacks deserve further study and require better evaluation against observations A number of studies have demonstrated strong relationships between precipitation efficiency, particularly its changes under warming, and climate sensitivity. In this chapter, we review the evidence for these relationships, including how they depend on the definition of precipitation efficiency. We identify six mechanisms by which changes in precipitation efficiency may affect Earth’s net climate feedback, and also discuss evidence for an inverse relationship between present-day precipitation efficiency and climate sensitivity based on several perturbed physics ensembles. This inverse relationship hints at the possibility of developing emergent constraints on climate sensitivity using precipitation efficiency, though it is put in doubt by studies varying convective entrainment rates, which have found the opposite relationship. More work is required to refine our understanding of the mechanisms linking changes in precipitation efficiency to climate sensitivity and more observational data is needed to validate model results. In particular, the precipitation efficiency of mid-latitude clouds has been relatively understudied, but deserves more attention in light of the importance of extratropical cloud feedbacks for the high climate sensitivities of CMIP6 models.

Single-column models (SCMs) are often used to evaluate model physics and aid parameterization development. However, few studies have systematically compared the results obtained using 1D setups with those of their corresponding 3D models, and examined what factors potentially impact their comparability. This paper addresses these questions. We focus on the application of SCMs under idealized RCE conditions and use a multi-column model (MCM) setup as stepping stone for a 3D model. We find that convective organization in the MCM depends at least as much on the convection scheme used as on other mechanisms known to organize convection (e.g., radiative feedback). Moreover, convective organization emerges as a robust factor affecting SCM-MCM comparability, with more aggregated states in 3D associated with larger behavior deviations from the 1D counterpart. This is found across five convection schemes and applies to simulated mean states, linear responses to small tendency perturbations, and adjustments to doubled-CO2 forcing. Applying a “model-as-truth” approach, we find that even when convection is organized, behavior differences between pairs of schemes in the SCM are largely preserved in the MCM. This indicates that when model physics produces accurate behavior in a 1D setup, it will be more likely to do so in a 3D setup. We also demonstrate the practical value of linear responses by showing that they can accurately predict an SCM’s tropospheric adjustment to doubled-CO2 forcing.