Summary Description
Climate projection datasets are collections of outputs from climate models that simulate the climate of the past as well as future climate conditions. Future simulations are based on future greenhouse gas emission scenarios. These datasets are used to assess potential impacts of climate change on various sectors, including agriculture, water resources, energy systems or ecosystems. Most climate projection datasets presented in this guide have been bias-adjusted to account for systematic deviations of the simulations. The strengths and limitations discussed on this page refer to these bias-adjusted datasets. See CMIP6 and CORDEX for strengths and limitations of the raw datasets that serve as input for bias-adjusted datasets.
Global Climate Models
Climate models represent the natural processes of the Earth based on physical principles of fluid dynamics and thermodynamics, broken down onto three-dimensional grids to represent interactions of terrain, atmosphere, ice, water, and vegetation (lithosphere, atmosphere, cryosphere, hydrosphere, biosphere). Global climate models (GCMs) or Earth system models (ESMs) can be interpreted as numerical replicas of planet Earth built to generate sequences of global atmospheric circulation and weather events. The resulting time series do not and are not meant to reproduce historical weather sequences or events. They are distinct from the observed record of weather events of the real world but reproduce well long-term climate and its evolution on Earth. Climate models are often used to simulate the response of climate to variations in external forcing, i.e. processes external to the climate that drive changes in the Earth’s energy balance. Exploring the Earth system’s response to changes in external forcing due to increasing greenhouse gas (GHG) concentrations in the atmosphere is the most prominent application of GCMs.
Regional Climate Models
Any region’s climate is inevitably linked to the global atmospheric circulation, therefore any simulation with a Regional climate model (RCMs) requires information about atmospheric conditions at its boundaries. Within a regional domain, RCMs operate on the same principles as GCMs; however, they are limited to smaller geographical regions. The conditions at the regions boundaries are taken from GCM simulations or global reanalysis of observational data to drive a regional model. With their smaller spatial coverage, RCM simulations can be run at higher spatial resolution than GCM simulations, thus providing added value where small-scale features impact local climate, such as mountains or coastlines. The process of running an RCM with the input from a driving GCM is often referred to as “dynamical downscaling,” since the higher spatial resolution output from the RCM is intrinsically tied to the coarser driving data. Generations of ensembles of RCM simulations become available later than the GCM simulations, as they depend on the latter as their driving data. Hence, a CMIP generations data from GCMs are published first, followed by those from RCMs. Usually, RCMs will be driven by a relatively small subset of available GCM simulations, which modifies RCM ensemble characteristics compared to GCM ensembles. A regions RCM ensemble will be defined by a combination of RCMs and GCMs which can be represented as a matrix. The North America CORDEX-CMIP5 simulation matrix provides an example of such an ensemble based on the GCMs modeling centers selected to drive their RCM(s).
Strengths and Limitations
Key Strengths of bias-adjusted datasets
| High spatial resolution |
Resolves effects of regional topography and local climate patterns better than GCMs. |
| Supports impact studies |
Designed to be used for vulnerability, impact, and adaptation (VIA) studies. |
| Adjustement of biases |
Some of GCM’s biases are sustantially reduced. |
Key Limitations of bias-adjusted datasets
| Reference bias |
Biases present in the reference dataset will also appear in bias-adjusted datasets. |
| Biases remain |
Bias-adjustment cannot completely remove all biases. Different methods will target different characteristics of a dataset. |
| No-physics at the small scale |
The fine resolution information only come from the reference and not a physical simulation of the future. |
| Bias stationnarity |
Bias-adjustment assumes that the bias are constant over time. |
Expert Guidance
Multi-Model Ensembles
To be further developed Research institutions around the globe run global climate models to generate simulations of past climate and projections of future climate conditions. The uncertainties inherent to this process, including structural model uncertainty and natural internal climate variability, are best represented by ensembles of these simulations.
Structural model uncertainty arises from the limitations and approximations made in climate models to represent processes. Climate models, while based on physical principles of fluid dynamics and thermodynamics, parameterize many processes at scales smaller than their spatial resolution, such as cloud formation or thunderstorms. This means that these small-scale processes are represented by simplified approximations rather than resolved explicitly by the physics of the model. Different climate modelling groups use different parameterizations and model structures, leading to different responses to the same external forcings, such as green house gas emissions. The diversity among models is beneficial for capturing a range of possible outcomes but also introduces uncertainty in climate projections. The other important source of uncertainty, natural climate variability, refers to the chaotic nature of the climate system, and leads to different sequences of weather events even under the same external forcings.
Good practice for climate assessment uses ensembles of climate simulations to enhance the robustness and reliability of an assessment through the combination of outputs of multiple climate models. This allows to factor in uncertainty from model structure and to account for the climate system’s natural variability in climate analysis for the electricity sector.
- Ensemble selection / ensemble reduction techniques.
Emission Scenarios
To be further developed
Bias adjustement
**ToDo: To be added: - Frequency of extremes is not adjusted. - makes the assumption that the relative spatial patterns will remain constant under future climate change.
Given that GCMs operate at relatively coarse spatial resolutions (typically 100-250 km), their outputs are often inadequate for regional scale modeling. Because electricity planning is typically sensitive to finer-scale temperature variations, it is essential that climate inputs local meteorological conditions with sufficient detail. To address this, two primary categories of downscaling techniques are widely applied in the literature: statistical downscaling and dynamical downscaling.
Statistical downscaling establishes empirical relationships between large-scale GCM variables and local climate observations, using historical data as a training base. These methods are computationally efficient and have been widely used where computational resources or high-resolution models are limited. In contrast, dynamical downscaling embeds higher-resolution Regional Climate Models (RCMs) within GCMs to explicitly simulate regional climate processes. This approach is particularly beneficial for capturing terrain-driven weather dynamics and localized extremes.
Nevertheless, dynamical downscaling is not without limitations. Bias correction is typically necessary because RCM outputs may deviate from observed climatology. Moreover, the selection of GCM-RCM combinations can introduce additional variability into the projections, as a concern emphasized by Chilkoti et al. (2017). To address this, ensemble approaches using multiple model combinations have become more common to capture the spread of plausible futures and quantify projection uncertainty, although this adds additional computational costs.
Global Warming Levels
To be further developed Traditionally, climate projections have been presented as the change of a future 20-year or 30-year period (e.g. 2071-2100) with respect to present climate conditions. However, during the last IPCC cycle, the WG1 introduced the concept of Warming Levels (WLs), wherein climate change projections are presented relative to the change in Global Mean Temperature (GMT) with respect to preindustrial conditions (e.g. 2◦C above preindustrial level), rather than tied to a designated future period. In this framework, a WL is first selected (e.g. 2◦C) and then, for each GCM of an ensemble, the 20-year or 30-year period centered on the predetermined change in GMT is selected for comparison. Because the GCMs warm at different rates, the GCMs will reach the selected level of warming at different times. Consequently, different future time periods will usually be selected for each GCMs.
Projection Temporal Horizons
To be developed
Delta Method
To be developed
