Climate-invariant machine learning.

Projecting climate change is a generalization problem: We extrapolate the recent past using physical models across past, present, and future climates. Current climate models require representations of processes that occur at scales smaller than model grid size, which have been the main source of model projection uncertainty. Recent machine learning (ML) algorithms hold promise to improve such process representations but tend to extrapolate poorly to climate regimes that they were not trained on. To get the best of the physical and statistical worlds, we propose a framework, termed "climate-invariant" ML, incorporating knowledge of climate processes into ML algorithms, and show that it can maintain high offline accuracy across a wide range of climate conditions and configurations in three distinct atmospheric models. Our results suggest that explicitly incorporating physical knowledge into data-driven models of Earth system processes can improve their consistency, data efficiency, and generalizability across climate regimes.

High-resolution (1 km) Köppen-Geiger maps for 1901-2099 based on constrained CMIP6 projections.

We introduce Version 2 of our widely used 1-km Köppen-Geiger climate classification maps for historical and future climate conditions. The historical maps (encompassing 1901-1930, 1931-1960, 1961-1990, and 1991-2020) are based on high-resolution, observation-based climatologies, while the future maps (encompassing 2041-2070 and 2071-2099) are based on downscaled and bias-corrected climate projections for seven shared socio-economic pathways (SSPs). We evaluated 67 climate models from the Coupled Model Intercomparison Project phase 6 (CMIP6) and kept a subset of 42 with the most plausible CO2-induced warming rates. We estimate that from 1901-1930 to 1991-2020, approximately 5% of the global land surface (excluding Antarctica) transitioned to a different major Köppen-Geiger class. Furthermore, we project that from 1991-2020 to 2071-2099, 5% of the land surface will transition to a different major class under the low-emissions SSP1-2.6 scenario, 8% under the middle-of-the-road SSP2-4.5 scenario, and 13% under the high-emissions SSP5-8.5 scenario. The Köppen-Geiger maps, along with associated confidence estimates, underlying monthly air temperature and precipitation data, and sensitivity metrics for the CMIP6 models, can be accessed at www.gloh2o.org/koppen .

Humidity's Role in Heat-Related Health Outcomes: A Heated Debate.

BACKGROUND: As atmospheric greenhouse gas concentrations continue to rise, temperature and humidity will increase further, causing potentially dire increases in human heat stress. On physiological and biophysical grounds, exposure to higher levels of humidity should worsen heat stress by decreasing sweat evaporation. However, population-scale epidemiological studies of heat exposure and response often do not detect associations between high levels of humidity and heat-related mortality or morbidity. These divergent, disciplinary views regarding the role of humidity in heat-related health risks limit confidence in selecting which interventions are effective in reducing health impacts and in projecting future heat-related health risks. OBJECTIVES: Via our multidisciplinary perspective we seek to a) reconcile the competing realities concerning the role of humidity in heat-related health impacts and b) help ensure robust projections of heat-related health risks with climate change. These objectives are critical pathways to identify and communicate effective approaches to cope with present and future heat challenges. DISCUSSION: We hypothesize six key reasons epidemiological studies have found little impact of humidity on heat-health outcomes: a) At high temperatures, there may be limited influence of humidity on the health conditions that cause most heat-related deaths (i.e., cardiovascular collapse); b) epidemiological data sets have limited spatial extent, a bias toward extratropical (i.e., cooler and less humid), high-income nations, and tend to exist in places where temporal variations in temperature and humidity are positively correlated; c) analyses focus on older, vulnerable populations with sweating, and thus evaporative, impairments that may be further aggravated by dehydration; d) extremely high levels of temperature and humidity (seldom seen in the historical record) are necessary for humidity to substantially impact heat strain of sedentary individuals; e) relationships between temperature and humidity are improperly considered when interpreting epidemiological model results; and f) sub-daily meteorological phenomena, such as rain, occur at high temperatures and humidity, and may bias epidemiological studies based on daily data. Future research must robustly test these hypotheses to advance methods for more accurate incorporation of humidity in estimating heat-related health outcomes under present and projected future climates. https://doi.org/10.1289/EHP11807.

Assessing effective radiative forcing from aerosol-cloud interactions over the global ocean.

How clouds respond to anthropogenic sulfate aerosols is one of the largest sources of uncertainty in the radiative forcing of climate over the industrial era. This uncertainty limits our ability to predict equilibrium climate sensitivity (ECS)-the equilibrium global warming following a doubling of atmospheric CO2. Here, we use satellite observations to quantify relationships between sulfate aerosols and low-level clouds while carefully controlling for meteorology. We then combine the relationships with estimates of the change in sulfate concentration since about 1850 to constrain the associated radiative forcing. We estimate that the cloud-mediated radiative forcing from anthropogenic sulfate aerosols is [Formula: see text] W m-2 over the global ocean (95% confidence). This constraint implies that ECS is likely between 2.9 and 4.5 K (66% confidence). Our results indicate that aerosol forcing is less uncertain and ECS is probably larger than the ranges proposed by recent climate assessments.

Emergent Constraints on Regional Cloud Feedbacks.

Low-cloud based emergent constraints have the potential to substantially reduce uncertainty in Earth's equilibrium climate sensitivity, but recent work has shown that previously developed constraints fail in the latest generation of climate models, suggesting that new approaches are needed. Here, we investigate the potential for emergent constraints to reduce uncertainty in regional cloud feedbacks, rather than the global-mean cloud feedback. Strong relationships are found between the monthly and interannual variability of tropical clouds, and the tropical net cloud feedback. These relationships are combined with observations to substantially narrow the uncertainty in the tropical cloud feedback and demonstrate that the tropical cloud feedback is likely >0Wm-2K-1. Promising relationships are also found in the 90°-60°S and 30°-60°N regions, though these relationships are not robust across model generations and we have not identified the associated physical mechanisms.

Weaker Links Between Zonal Convective Clustering and ITCZ Width in Climate Models Than in Observations.

Strong links are seen in observations between convective clustering and several properties of the Intertropical Convergence Zone (ITCZ). These links suggest that biases in how climate models simulate the ITCZ may be related to model biases in convective clustering or that there may be biases in how models represent the relationship between clustering and the ITCZ. We investigate these issues by analyzing convective clustering, and the link between clustering and ITCZ properties in 18 climate models. We find that the links between variability in convective clustering and variability of ITCZ properties are generally weaker and less robust in models than in observations. By contrast, model biases in the climatological convective clustering explain a substantial fraction of the climatological double-ITCZ bias, though they do not explain biases in the climatological ITCZ width.

The Relationship Between Convective Clustering and Mean Tropical Climate in Aquaplanet Simulations.

Convective clustering, the spatial organization of tropical deep convection, can manifest itself in two ways: through a decrease in the total area covered by convection and/or through a decrease in the number of convective areas. Much of our current understanding of convective clustering comes from simulations in idealized radiative convective equilibrium (RCE) configurations. In these simulations the two forms of convective clustering tend to covary, and their individual effects on the climate are thus hard to disentangle. This study shows that in aquaplanet simulations with more realistic boundary conditions, such as meridional gradients of surface temperature and rotational forces, the two aspects of convective clustering are not equivalent and are associated with different impacts on the large-scale climate. For instance, reducing the convective area in the equatorial region in the aquaplanet simulations results in broader meridional humidity and rain distributions and in lower tropospheric temperatures throughout the tropics. By contrast, the number of convective regions primarily impacts the zonal variance of humidity-related quantities in the aquaplanet simulations, as the distribution of convective regions affects the size of the subsidence regions and thereby the moistening influence of convective regions. The aquaplanet simulations confirm many other qualitative results from RCE simulations, such as a reduction of equatorial tropospheric humidity when the area covered by convection diminishes.