Evidence shows a continuing increase in the frequency and severity of global heatwaves^1,2, raising concerns about the future impacts of climate change and the associated socioeconomic costs^3,4. Here we develop a disaster footprint analytical framework by integrating climate, epidemiological and hybrid input–output and computable general equilibrium global trade models to estimate the midcentury socioeconomic impacts of heat stress. We consider health costs related to heat exposure, the value of heat-induced labour productivity loss and indirect losses due to economic disruptions cascading through supply chains. Here we show that the global annual incremental gross domestic product loss increases exponentially from 0.03 ± 0.01 (SSP 245)–0.05 ± 0.03 (SSP 585) percentage points during 2030–2040 to 0.05 ± 0.01–0.15 ± 0.04 percentage points during 2050–2060. By 2060, the expected global economic losses reach a total of 0.6–4.6% with losses attributed to health loss (37–45%), labour productivity loss (18–37%) and indirect loss (12–43%) under different shared socioeconomic pathways. Small- and medium-sized developing countries suffer disproportionately from higher health loss in South-Central Africa (2.1 to 4.0 times above global average) and labour productivity loss in West Africa and Southeast Asia (2.0–3.3 times above global average). The supply-chain disruption effects are much more widespread with strong hit to those manufacturing-heavy countries such as China and the USA, leading to soaring economic losses of 2.7 ± 0.7% and 1.8 ± 0.5%, respectively.

Current moist physics parameterization schemes in general circulation models (GCMs) are the main source of biases in simulated precipitation and atmospheric circulation. Recent advances in machine learning make it possible to explore data‐driven approaches to developing parameterization for moist physics processes such as convection and clouds. This study aims to develop a new moist physics parameterization scheme based on deep learning. We use a residual convolutional neural network (ResNet) for this purpose. It is trained with 1‐year simulation from a superparameterized GCM, SPCAM. An independent year of SPCAM simulation is used for evaluation. In the design of the neural network, referred to as ResCu, the moist static energy conservation during moist processes is considered. In addition, the past history of the atmospheric states, convection, and clouds is also considered. The predicted variables from the neural network are GCM grid‐scale heating and drying rates by convection and clouds, and cloud liquid and ice water contents. Precipitation is derived from predicted moisture tendency. In the independent data test, ResCu can accurately reproduce the SPCAM simulation in both time mean and temporal variance. Comparison with other neural networks demonstrates the superior performance of ResNet architecture. ResCu is further tested in a single‐column model for both continental midlatitude warm season convection and tropical monsoonal convection. In both cases, it simulates the timing and intensity of convective events well. In the prognostic test of tropical convection case, the simulated temperature and moisture biases with ResCu are smaller than those using conventional convection and cloud parameterizations.

In order to investigate the impact of different treatments for the contact angle (α) in heterogeneous ice nucleating properties of natural dust and black carbon (BC) particles, we implement the classical-nucleation-theory-based parameterization of heterogeneous ice nucleation (Hoose et al., 2010) in the Community Atmospheric Model version 5 (CAM5) and then improve it by replacing the original single-contact-angle model with the probability-density-function-of-α (α-PDF) model to better represent the ice nucleation behavior of natural dust found in observations. We refit the classical nucleation theory (CNT) to constrain the uncertain parameters (i.e., onset α and activation energy in the single-α model; mean contact angle and standard deviation in the α-PDF model) using recent observation data sets for Saharan natural dust and BC (soot). We investigate the impact of the time dependence of droplet freezing on mixed-phase clouds and climate in CAM5 as well as the roles of natural dust and soot in different nucleation mechanisms. Our results show that, when compared with observations, the potential ice nuclei (IN) calculated by the α-PDF model show better agreement than those calculated by the single-α model at warm temperatures (T; T > −20 °C). More ice crystals can form at low altitudes (with warm temperatures) simulated by the α-PDF model than compared to the single-α model in CAM5. All of these can be attributed to different ice nucleation efficiencies among aerosol particles, with some particles having smaller contact angles (higher efficiencies) in the α-PDF model. In the sensitivity tests with the α-PDF model, we find that the change in mean contact angle has a larger impact on the active fraction at a given temperature than a change in standard deviation, even though the change in standard deviation can lead to a change in freezing behavior. Both the single-α and the α-PDF model indicate that the immersion freezing of natural dust plays a more important role in the heterogeneous nucleation than that of soot in mixed-phase clouds. The new parameterizations implemented in CAM5 induce more significant aerosol indirect effects than the default parameterization.

A team effort to develop a Community Integrated Earth System Model (CIESM) was initiated in China in 2012. The model was based on NCAR Community Earth System Model (Version 1.2.1) with several novel developments and modifications aimed to overcome some persistent systematic biases, such as the double Intertropical Convergence Zone problem and underestimated marine boundary layer clouds. Aerosols' direct and indirect effects are prescribed using the MACv2‐SP approach and data sets. The spin‐up of a 500‐year preindustrial simulation and three historical simulations are described and evaluated. Prominent improvements include alleviated double Intertropical Convergence Zone problem, increased marine boundary layer clouds, and better El Niño Southern Oscillation amplitude and periods. One deficiency of the model is the significantly underestimated Arctic and Antarctic sea ice in warm seasons. The historical warming is about 0.55 °C greater than observations toward 2014. CIESM has an equilibrium climate sensitivity of 5.67 K, mainly resulted from increased positive shortwave cloud feedback. Our efforts on porting and redesigning CIESM for the heterogeneous Sunway TaihuLight supercomputer are also introduced, including some ongoing developments toward a future version of the model.

The Plant‐Craig (PC) stochastic convective parameterization scheme is implemented into the National Center for Atmospheric Research Community Atmosphere Model version 5 (CAM5) to couple with the Zhang‐McFarlane deterministic convection scheme. To evaluate its impact on tropical precipitation simulation, two experiments are conducted: one with the standard CAM5 and the other with the stochastic scheme incorporated. Results show that the PC stochastic parameterization decreases the frequency of weak precipitation and increases the frequency of strong precipitation, resulting in better agreement with observations. The most striking improvement is in the probability distribution function (PDF) of precipitation intensity, with the well‐known too‐much‐drizzle problem in CAM5 largely eliminated. In the global tropical belt, the precipitation intensity PDF from the simulation agrees remarkably well with that of Tropical Rainfall Measuring Mission observations. The stochastic scheme also yields a similar magnitude of intraseasonal variability of precipitation to observations and improves the simulation of the eastward propagating intraseasonal signals of precipitation and zonal wind.