Research

Probabilistic projections with simplified models

Simple or intermediate complexity climate models can be run thousands of times to sample uncertainties in the main parameters like climate sensitivity and ocean heat uptake which affect the future temperature response. With the use of observations, it is possible to narrow down the range of model responses by selecting only those models for the future that have reproduced the past sufficiently well. Such methods are useful to generate probability distributions for global temperature for different scenarios, and to generate distributions for parameters like climate sensitivity. The models are fast and flexible to explore new ideas, but their interpretation is limited to large scales. Further difficulties lie in the choices of prior distributions, i.e. how the parameter space is sampled, which observations are chosen as constraints, and in the model itself, which does not resolve many of the important processes and feedbacks.

 

Uncertainties from multiple 3D models

Comprehensive three-dimensional coupled models resolving the atmosphere, ocean, land and sea ice can be used to make projections on smaller scales and for quantities other than temperature.

A climate model is based on a system of differential equations describing physics, fluid motions and chemistry. In order to solve the equations, the planet is divided into a 3-dimensional grid on which the equations are discretized. The time dependent solution is then calculated numerically with the aid of a computer code.

 

In some cases, however, the agreement of the climate projections of the different models is very poor. It is very likely that aggregating changes over larger regions can lead to an increased consistency of the projections across the different models. The first step is, therefore, a detailed study on the spatial scales over which models provide robust results.

The difficulty arising from working with climate projections a century into the future is that the response of the model can not be verified with observations. Therefore, one has to define what a good model is based on other quantities. Climate models have traditionally been evaluated and compared to observations mostly on the climatological mean state, and to some degree on their variability as well as on individual processes. Despite the fact that models are getting continuously better in simulating the present day mean state, their spread a century into the future has not decreased. This suggests that an accurate representation of the mean state provides little information on the ability of a model to predict the future climate. Efforts on finding an objective method on evaluating the skill of climate models are ongoing in our group.

 

In order to improve the overall results of the climate projections it seems very natural to combine the results of climate change projections. For climate models, some errors indeed appear to be random such that an average of the models results in an increase of information. But how much do we gain in averaging models? Given the differences in model performance, the problem of dependence and common bias, it seems unlikely that blindly averaging across all models with equal weight is making optimal use of the available information. Finding weights to attach to the models is one focus of our research group.

The role of the ocean in climate change

The ocean is the largest heat reservoir responding to climate change. Because of the long timescales in the ocean, there is currently an imbalance in the climate system. The ocean takes up heat and therefore ‘hides’ part of the warming signal at the surface. Understanding how the ocean responds to climate change, both with heat uptake and with its currents, is a crucial element in understanding the transient response to the climate forcing.

Carbon cycle climate system interactions

The chain from carbon emissions to concentrations to forcing to temperature and sea level response involves many components in the climate system, which interact on multiple timescales. Uncertainty is introduced at each step. If the goal is to keep climate change below a tolerable limit (for example 2°C above preindustrial), then emissions need to peak soon and then decrease dramatically. Coupled carbon cycle climate models can be used to predict an emission path that needs to be followed in order to reach a certain temperature target, and in what way such an emission path is affected by uncertainties in the physical climate system as well as in the carbon cycle. The figure below illustrates the complex interactions of carbon emissions, atmospheric CO₂ surface warming and sea level rise. As the emissions of carbon are reduced to zero in the year 2100, it takes centuries for the surface temperature and the atmospheric CO₂ to stabilize on preindustrial levels, whereas the the sea level keeps rising.

The other side of climate change