UQ in Earth System Sciences#
The Earth system is an integrated combination of four main sub-systems, namely: the geosphere (interior and surface of Earth), the biosphere (all living organisms), the hydrosphere (water, ice, and vapor), and the atmosphere. These subsystems are interconnected by intricate processes and cycles. Earth’s climate is an example, which consists of many interconnected processes and cycles whose response time varies widely from days to weeks for the atmosphere to thousands of years for the oceans, ice sheets and the lithosphere.
Stratospheric ozone, as is well known, is necessary for life on Earth. However, the emission of ozone depleting substances (ODSs) and greenhouse gases (GHGs) into the atmosphere have had a negative effect on ozone concentrations in the stratosphere, thus increasing harmful UV radiation near the surface. The Montreal protocol has limited and prohibited the use of several ODSs and ecovery of stratospheric ozone has been observed since 2000 (see IPCC AR5 report or Scientific Assessment of Ozone Depletion: 2018).
To simulate Earth’s climate, numerical models, also known as general circulation models (GCMs), are used. However, in order to better understand the evolution of future ozone in the atmosphere, a new generation of climate models, also known as Chemistry-Climate Models (CCM’s), are used for future climate projections. They interactively calculate chemistry (including ozone) for prescribed future scenarios of GHGs and ODSs. Quantifying uncertainties in future projections of climate and ozone has significant impact on the gauging of milestones, e.g. when will ozone be back to 1980 levels - or when will temperature pass a certain warming threshold. Better understanding of uncertainties can help researchers, political decision makers and stakeholders in developing efficient and robust climate strategies.
There are different GCMs being used in Earth System Modeling (ESM) communties. With regards to the Helmholtz-uncertainty quantification project, ICON-ART is used for climate simulations at KIT while the model AWI-CM is used at AWI. A short description about this project is given below. At the core of climate/atmospheric models lie multiple systems of mathematical equations such as Navier-Stokes equations for atmospheric and ocean motions, transport equations for gases, particles and liquids, parameterizations for unresolved processes, radiative transfer (for solar and infrared radiation), chemistry and etc.
ICON-ART#
ICON (ICOsahedral Nonhydrostatic) model is the next generation weather forecasting and global climate modeling framework. ICON-ART (Aerosol and Reactive Trace gases), developed at KIT, is an extension of ICON to simulate chemical tracers and the transport of aerosols like volcanic ash, atmospheric dust, and others.
AWI-CM#
AWI-CM is a coupled climate model. There are different variants: AWI-CM1 uses the atmospheric model ECHAM6 in combination with the finite-element ocean model FESOM, while AWI-CM3 simulates the atmosphere using Open-IFS and the ocean using the finite-volume ocean model FESOM2.
In the following, the uncertainties that occur in Earth system models, specifically in climate/atmospheric model, are defined briefly.
Uncertainties in Climate Projections#
There are three main types of uncertainties associated with chemistry-climate models (CCMs) while performing simulations.
Emission Scenario Uncertainty
Structural Uncertainty
Natural Internal variability
Emission Scenario Uncertainty#
Representative Concentration Pathways (RCPs) suggest four different pathways of greenhouse gas emissions, air pollution and land use for the 21st century. These scenarios are used in climate model projections. So far, RCP scenarios are considered as a useful range of estimates. However, there are uncertainties in the future evolution of these scenarios. fig_uq_in_ESM taken from the IPCC AR5 report illustrates the future pathways.
<Figure 1: Image source: IPCC AR5, Box 2.2>
Structural Uncertainty#
Structural uncertainty can be analyzed by comparing forecasts/projections of different CCMs (or comparing different simulations/realizations of the same model), while e.g. certain processes can be represented using different descriptions and/or parameterizations (impact of a small change). It has also been observed that uncertainty in the projections arise when certain numerical schemes are used to solve the same set of prognostic equations (integration in time). In addition, the grid resolution for climate models is an important factor contributing to the uncertainties. Moreover, there are usually many uncertain parameters and/or coefficients in chemistry modules of CCMs. As CCMs are becoming better (more complex) with each generation, the number of uncertain paramter is growing as well, making CCMs projections potentially more uncertain.
Natural Internal Variability#
CCMs also exhibit internal variability which impacts future climate projections. For this type of uncertainty, multiple simulations are needed with slightly different initial and boundary conditions. However, there are some natural processes which are difficult to model due to the complex processes involved and insufficient historical data for the validation of such processes. A list of some of those processes is given below.
UQ methods in ESM#
Climate models are expensive to compute which makes it harder to apply standard methods of uncertainty quantification like Monte Carlo methods. However, methods such as ensemble data assimilation and multi-level Monte Carlo are viable approaches in uncertainty quantification for Earth System Modeling. These approaches not only offer potential solutions but are actively employed in practice. Ensemble data assimilation involves integrating observational data into model simulations, enabling us to estimate certain model parameters and improve the assessment of the current model state. Similarly, multi-level Monte Carlo methods contribute to the quantification and potential reduction of uncertainties in model projections and predictions.
In current research, there is a growing emphasis on investigating uncertainties in ESMs and implementing efficient methods to estimate and manage them. These UQ methods play a crucial role in enhancing the reliability of climate projections. For example, ensemble data assimilation allows for a more realistic integration of real-world observations, leading to a refined understanding of the model’s behavior. Simultaneously, multi-level Monte Carlo methods offer a sophisticated means to tackle uncertainties inherent in complex Earth system processes.