Non-Linear Dimensionality Reduction With a Variational Encoder Decoder to Understand Convective Processes in Climate Models.

Deep learning can accurately represent sub-grid-scale convective processes in climate models, learning from high resolution simulations. However, deep learning methods usually lack interpretability due to large internal dimensionality, resulting in reduced trustworthiness in these methods. Here, we use Variational Encoder Decoder structures (VED), a non-linear dimensionality reduction technique, to learn and understand convective processes in an aquaplanet superparameterized climate model simulation, where deep convective processes are simulated explicitly. We show that similar to previous deep learning studies based on feed-forward neural nets, the VED is capable of learning and accurately reproducing convective processes. In contrast to past work, we show this can be achieved by compressing the original information into only five latent nodes. As a result, the VED can be used to understand convective processes and delineate modes of convection through the exploration of its latent dimensions. A close investigation of the latent space enables the identification of different convective regimes: (a) stable conditions are clearly distinguished from deep convection with low outgoing longwave radiation and strong precipitation; (b) high optically thin cirrus-like clouds are separated from low optically thick cumulus clouds; and (c) shallow convective processes are associated with large-scale moisture content and surface diabatic heating. Our results demonstrate that VEDs can accurately represent convective processes in climate models, while enabling interpretability and better understanding of sub-grid-scale physical processes, paving the way to increasingly interpretable machine learning parameterizations with promising generative properties.

Deep Learning Based Cloud Cover Parameterization for ICON.

A promising approach to improve cloud parameterizations within climate models and thus climate projections is to use deep learning in combination with training data from storm-resolving model (SRM) simulations. The ICOsahedral Non-hydrostatic (ICON) modeling framework permits simulations ranging from numerical weather prediction to climate projections, making it an ideal target to develop neural network (NN) based parameterizations for sub-grid scale processes. Within the ICON framework, we train NN based cloud cover parameterizations with coarse-grained data based on realistic regional and global ICON SRM simulations. We set up three different types of NNs that differ in the degree of vertical locality they assume for diagnosing cloud cover from coarse-grained atmospheric state variables. The NNs accurately estimate sub-grid scale cloud cover from coarse-grained data that has similar geographical characteristics as their training data. Additionally, globally trained NNs can reproduce sub-grid scale cloud cover of the regional SRM simulation. Using the game-theory based interpretability library SHapley Additive exPlanations, we identify an overemphasis on specific humidity and cloud ice as the reason why our column-based NN cannot perfectly generalize from the global to the regional coarse-grained SRM data. The interpretability tool also helps visualize similarities and differences in feature importance between regionally and globally trained column-based NNs, and reveals a local relationship between their cloud cover predictions and the thermodynamic environment. Our results show the potential of deep learning to derive accurate yet interpretable cloud cover parameterizations from global SRMs, and suggest that neighborhood-based models may be a good compromise between accuracy and generalizability.

Context for interpreting equilibrium climate sensitivity and transient climate response from the CMIP6 Earth system models.

For the current generation of earth system models participating in the Coupled Model Intercomparison Project Phase 6 (CMIP6), the range of equilibrium climate sensitivity (ECS, a hypothetical value of global warming at equilibrium for a doubling of CO2) is 1.8°C to 5.6°C, the largest of any generation of models dating to the 1990s. Meanwhile, the range of transient climate response (TCR, the surface temperature warming around the time of CO2 doubling in a 1% per year CO2 increase simulation) for the CMIP6 models of 1.7°C (1.3°C to 3.0°C) is only slightly larger than for the CMIP3 and CMIP5 models. Here we review and synthesize the latest developments in ECS and TCR values in CMIP, compile possible reasons for the current values as supplied by the modeling groups, and highlight future directions. Cloud feedbacks and cloud-aerosol interactions are the most likely contributors to the high values and increased range of ECS in CMIP6.

Causal networks for climate model evaluation and constrained projections.

Global climate models are central tools for understanding past and future climate change. The assessment of model skill, in turn, can benefit from modern data science approaches. Here we apply causal discovery algorithms to sea level pressure data from a large set of climate model simulations and, as a proxy for observations, meteorological reanalyses. We demonstrate how the resulting causal networks (fingerprints) offer an objective pathway for process-oriented model evaluation. Models with fingerprints closer to observations better reproduce important precipitation patterns over highly populated areas such as the Indian subcontinent, Africa, East Asia, Europe and North America. We further identify expected model interdependencies due to shared development backgrounds. Finally, our network metrics provide stronger relationships for constraining precipitation projections under climate change as compared to traditional evaluation metrics for storm tracks or precipitation itself. Such emergent relationships highlight the potential of causal networks to constrain longstanding uncertainties in climate change projections.

Projected land photosynthesis constrained by changes in the seasonal cycle of atmospheric CO2.

Uncertainties in the response of vegetation to rising atmospheric CO2 concentrations contribute to the large spread in projections of future climate change. Climate-carbon cycle models generally agree that elevated atmospheric CO2 concentrations will enhance terrestrial gross primary productivity (GPP). However, the magnitude of this CO2 fertilization effect varies from a 20 per cent to a 60 per cent increase in GPP for a doubling of atmospheric CO2 concentrations in model studies. Here we demonstrate emergent constraints on large-scale CO2 fertilization using observed changes in the amplitude of the atmospheric CO2 seasonal cycle that are thought to be the result of increasing terrestrial GPP. Our comparison of atmospheric CO2 measurements from Point Barrow in Alaska and Cape Kumukahi in Hawaii with historical simulations of the latest climate-carbon cycle models demonstrates that the increase in the amplitude of the CO2 seasonal cycle at both measurement sites is consistent with increasing annual mean GPP, driven in part by climate warming, but with differences in CO2 fertilization controlling the spread among the model trends. As a result, the relationship between the amplitude of the CO2 seasonal cycle and the magnitude of CO2 fertilization of GPP is almost linear across the entire ensemble of models. When combined with the observed trends in the seasonal CO2 amplitude, these relationships lead to consistent emergent constraints on the CO2 fertilization of GPP. Overall, we estimate a GPP increase of 37 ± 9 per cent for high-latitude ecosystems and 32 ± 9 per cent for extratropical ecosystems under a doubling of atmospheric CO2 concentrations on the basis of the Point Barrow and Cape Kumukahi records, respectively.

The effect of future ambient air pollution on human premature mortality to 2100 using output from the ACCMIP model ensemble.

Ambient air pollution from ground-level ozone and fine particulate matter (PM2.5) is associated with premature mortality. Future concentrations of these air pollutants will be driven by natural and anthropogenic emissions and by climate change. Using anthropogenic and biomass burning emissions projected in the four Representative Concentration Pathway scenarios (RCPs), the ACCMIP ensemble of chemistry-climate models simulated future concentrations of ozone and PM2.5 at selected decades between 2000 and 2100. We use output from the ACCMIP ensemble, together with projections of future population and baseline mortality rates, to quantify the human premature mortality impacts of future ambient air pollution. Future air pollution-related premature mortality in 2030, 2050 and 2100 is estimated for each scenario and for each model using a health impact function based on changes in concentrations of ozone and PM2.5 relative to 2000 and projected future population and baseline mortality rates. Additionally, the global mortality burden of ozone and PM2.5 in 2000 and each future period is estimated relative to 1850 concentrations, using present-day and future population and baseline mortality rates. The change in future ozone concentrations relative to 2000 is associated with excess global premature mortality in some scenarios/periods, particularly in RCP8.5 in 2100 (316 thousand deaths/year), likely driven by the large increase in methane emissions and by the net effect of climate change projected in this scenario, but it leads to considerable avoided premature mortality for the three other RCPs. However, the global mortality burden of ozone markedly increases from 382,000 (121,000 to 728,000) deaths/year in 2000 to between 1.09 and 2.36 million deaths/year in 2100, across RCPs, mostly due to the effect of increases in population and baseline mortality rates. PM2.5 concentrations decrease relative to 2000 in all scenarios, due to projected reductions in emissions, and are associated with avoided premature mortality, particularly in 2100: between -2.39 and -1.31 million deaths/year for the four RCPs. The global mortality burden of PM2.5 is estimated to decrease from 1.70 (1.30 to 2.10) million deaths/year in 2000 to between 0.95 and 1.55 million deaths/year in 2100 for the four RCPs, due to the combined effect of decreases in PM2.5 concentrations and changes in population and baseline mortality rates. Trends in future air pollution-related mortality vary regionally across scenarios, reflecting assumptions for economic growth and air pollution control specific to each RCP and region. Mortality estimates differ among chemistry-climate models due to differences in simulated pollutant concentrations, which is the greatest contributor to overall mortality uncertainty for most cases assessed here, supporting the use of model ensembles to characterize uncertainty. Increases in exposed population and baseline mortality rates of respiratory diseases magnify the impact on premature mortality of changes in future air pollutant concentrations and explain why the future global mortality burden of air pollution can exceed the current burden, even where air pollutant concentrations decrease.

Global air quality and climate.

Emissions of air pollutants and their precursors determine regional air quality and can alter climate. Climate change can perturb the long-range transport, chemical processing, and local meteorology that influence air pollution. We review the implications of projected changes in methane (CH(4)), ozone precursors (O(3)), and aerosols for climate (expressed in terms of the radiative forcing metric or changes in global surface temperature) and hemispheric-to-continental scale air quality. Reducing the O(3) precursor CH(4) would slow near-term warming by decreasing both CH(4) and tropospheric O(3). Uncertainty remains as to the net climate forcing from anthropogenic nitrogen oxide (NO(x)) emissions, which increase tropospheric O(3) (warming) but also increase aerosols and decrease CH(4) (both cooling). Anthropogenic emissions of carbon monoxide (CO) and non-CH(4) volatile organic compounds (NMVOC) warm by increasing both O(3) and CH(4). Radiative impacts from secondary organic aerosols (SOA) are poorly understood. Black carbon emission controls, by reducing the absorption of sunlight in the atmosphere and on snow and ice, have the potential to slow near-term warming, but uncertainties in coincident emissions of reflective (cooling) aerosols and poorly constrained cloud indirect effects confound robust estimates of net climate impacts. Reducing sulfate and nitrate aerosols would improve air quality and lessen interference with the hydrologic cycle, but lead to warming. A holistic and balanced view is thus needed to assess how air pollution controls influence climate; a first step towards this goal involves estimating net climate impacts from individual emission sectors. Modeling and observational analyses suggest a warming climate degrades air quality (increasing surface O(3) and particulate matter) in many populated regions, including during pollution episodes. Prior Intergovernmental Panel on Climate Change (IPCC) scenarios (SRES) allowed unconstrained growth, whereas the Representative Concentration Pathway (RCP) scenarios assume uniformly an aggressive reduction, of air pollutant emissions. New estimates from the current generation of chemistry-climate models with RCP emissions thus project improved air quality over the next century relative to those using the IPCC SRES scenarios. These two sets of projections likely bracket possible futures. We find that uncertainty in emission-driven changes in air quality is generally greater than uncertainty in climate-driven changes. Confidence in air quality projections is limited by the reliability of anthropogenic emission trajectories and the uncertainties in regional climate responses, feedbacks with the terrestrial biosphere, and oxidation pathways affecting O(3) and SOA.

Global-mean temperature change from shipping toward 2050: improved representation of the indirect aerosol effect in simple climate models.

We utilize a range of emission scenarios for shipping to determine the induced global-mean radiative forcing and temperature change. Ship emission scenarios consistent with the new regulations on nitrogen oxides (NO(x)) and sulfur dioxide (SO(2)) from the International Maritime Organization and two of the Representative Concentration Pathways are used as input to a simple climate model (SCM). Based on a complex aerosol-climate model we develop and test new parametrizations of the indirect aerosol effect (IAE) in the SCM that account for nonlinearities in radiative forcing of ship-induced IAE. We find that shipping causes a net global cooling impact throughout the period 1900-2050 across all parametrizations and scenarios. However, calculated total net global-mean temperature change in 2050 ranges from -0.03[-0.07,-0.002]°C to -0.3[-0.6,-0.2]°C in the A1B scenario. This wide range across parametrizations emphasizes the importance of properly representing the IAE in SCMs and to reflect the uncertainties from complex global models. Furthermore, our calculations show that the future ship-induced temperature response is likely a continued cooling if SO(2) and NO(x) emissions continue to increase due to a strong increase in activity, despite current emission regulations. However, such cooling does not negate the need for continued efforts to reduce CO(2) emissions, since residual warming from CO(2) is long-lived.

Climate impact of biofuels in shipping: global model studies of the aerosol indirect effect.

Aerosol emissions from international shipping are recognized to have a large impact on the Earth's radiation budget, directly by scattering and absorbing solar radiation and indirectly by altering cloud properties. New regulations have recently been approved by the International Maritime Organization (IMO) aiming at progressive reductions of the maximum sulfur content allowed in marine fuels from current 4.5% by mass down to 0.5% in 2020, with more restrictive limits already applied in some coastal regions. In this context, we use a global bottom-up algorithm to calculate geographically resolved emission inventories of gaseous (NO(x), CO, SO(2)) and aerosol (black carbon, organic matter, sulfate) species for different kinds of low-sulfur fuels in shipping. We apply these inventories to study the resulting changes in radiative forcing, attributed to particles from shipping, with the global aerosol-climate model EMAC-MADE. The emission factors for the different fuels are based on measurements at a test bed of a large diesel engine. We consider both fossil fuel (marine gas oil) and biofuels (palm and soy bean oil) as a substitute for heavy fuel oil in the current (2006) fleet and compare their climate impact to that resulting from heavy fuel oil use. Our simulations suggest that ship-induced surface level concentrations of sulfate aerosol are strongly reduced, up to about 40-60% in the high-traffic regions. This clearly has positive consequences for pollution reduction in the vicinity of major harbors. Additionally, such reductions in the aerosol loading lead to a decrease of a factor of 3-4 in the indirect global aerosol effect induced by emissions from international shipping.

Present-day and future global bottom-up ship emission inventories including polar routes.

We present a global bottom-up ship emission algorithm that calculates fuel consumption, emissions, and vessel traffic densities for present-day (2006) and two future scenarios (2050) considering the opening of Arctic polar routes due to projected sea ice decline. Ship movements and actual ship engine power per individual ship from Lloyd's Marine Intelligence Unit (LMIU) ship statistics for six months in 2006 and further mean engine data from literature serve as input. The developed SeaKLIM algorithm automatically finds the most probable shipping route for each combination of start and destination port of a certain ship movement by calculating the shortest path on a predefined model grid while considering land masses, sea ice, shipping canal sizes, and climatological mean wave heights. The resulting present-day ship activity agrees well with observations. The global fuel consumption of 221 Mt in 2006 lies in the range of previously published inventories when undercounting of ship numbers in the LMIU movement database (40,055 vessels) is considered. Extrapolated to 2007 and ship numbers per ship type of the recent International Maritime Organization (IMO) estimate (100,214 vessels), a fuel consumption of 349 Mt is calculated which is in good agreement with the IMO total of 333 Mt. The future scenarios show Arctic polar routes with regional fuel consumption on the Northeast and Northwest Passage increasing by factors of up to 9 and 13 until 2050, respectively.

Assessment of near-future policy instruments for oceangoing shipping: impact on atmospheric aerosol burdens and the earth's radiation budget.

We apply the global climate model ECHAM5/MESSy1-MADE with detailed aerosol and cloud microphysics to study the impact of shipping on tropospheric aerosol burdens, clouds, and the radiation budget for four near-future ship emission policy scenarios for the year 2012. We compare a "No Control" scenario with global sulfur limits and regionally applied reductions. We show that, if no control measures are taken, near surface sulfate increases by about 10-20% over the main transoceanic shipping routes from 2002 to 2012. A reduction of the maximum fuel sulfur (S) content allowed within 200 nautical miles of coastal areas ("global emission control areas") to 0.5% or 0.1% (5000 or 1000 ppm S, respectively) results in a distinctive reduction in near surface sulfate from shipping in coastal regions compared with the year 2002. The model results also show that if emissions of nitrogen oxides (NO(x)) remain unabated, a reduction of the fuel sulfur content favors a strong increase in aerosol nitrate (NO3) which could counteract up to 20% of the decrease in sulfate mass achieved by sulfur emission reductions. The most important impact of shipping on the radiation budget is related to the modification of low maritime stratus clouds resulting in an increased reflectivity and enhanced shortwave cloud forcing. The direct aerosol effect from shipping is small. Our study shows that one can expect a less negative (less cooling) radiative forcing due to reductions in the current fuel sulfur content of ocean-going ships. The global annual average net cloud forcings due to shipping (year 2012) are in the range of -0.27 to -0.58 W/m2 with regional cooling occurring most over the remote oceans.

Mortality from ship emissions: a global assessment.

Epidemiological studies consistently link ambient concentrations of particulate matter (PM) to negative health impacts, including asthma, heart attacks, hospital admissions, and premature mortality. We model ambient PM concentrations from oceangoing ships using two geospatial emissions inventories and two global aerosol models. We estimate global and regional mortalities by applying ambient PM increases due to ships to cardiopulmonary and lung cancer concentration-risk functions and population models. Our results indicate that shipping-related PM emissions are responsible for approximately 60,000 cardiopulmonary and lung cancer deaths annually, with most deaths occurring near coastlines in Europe, East Asia, and South Asia. Under current regulation and with the expected growth in shipping activity, we estimate that annual mortalities could increase by 40% by 2012.