It’s encouraging to note the growing interest for regional climate information for society and climate adaptation, such as recent advances in the World Climate Research Programme (WCRP), the climate adaptation summit CAS2021, and the new Digital Europe. These efforts are likely to boost the Global Framework for Climate Services (GFCS) needed as a guide to decision-makers on matters influenced by weather and climate.
These new moves underscore the understanding that we must start to act on mitigation and adaptation now. But I think we also need to keep in mind that there still are some unresolved issues when it comes to adaptation and there is still a need for more research and development.
For instance, one question remains about the physical consistency in regional climate models (RCMs). These models provide a more refined representation of the regional climate than the coarser driving global climate models (GCMs), but that also implies changes to the shortwave and longwave fluxes at the top of the atmosphere compared to the GCMs. This is a mere consequence of a better representation of clouds and rainfall patterns.
The improved simulations by RCMs can result in different moisture contents and fluxes in the RCM and the GCM. There are also different surface representations between the two, and sometimes different representations of aerosols. One question is whether these differences can explain the biases seen in the RCM results and if these biases are stationary during global warming. Another question is whether such biases could be related to different parametrization schemes used in GCMs and RCMs. Or do these differences matter at all?
It is possible to get some answers to these questions, for instance by plotting the differences between aggregated downscaled (“upscaled downscaled results”) and the GCM results used to drive the RCM. I think there is also a need to involve empirical-statistical downscaling (ESD) and compare such results with the RCMs. ESD and RCMs have different strengths and weaknesses, and they use different and independent sources of information; for the RCMs, it is from the code representing the physical processes on a higher resolution, whereas for ESD, the information is taken from empirical observations from the past. There are many examples where regional climate modelling only involves RCMs and not ESD. Including both may help us address uncertainties.
The concept of “cascading uncertainties” is sometimes discussed in the downscaling community, acknowledging additional uncertainties introduced with a new level of models. I think the term itself, and the way it often is presented, are misleading since each new level of models also introduces new information or constraints as well as uncertainties. If each additional level merely added uncertainties, then they would have no added-value and there would be no point with the extra levels. The information-to-uncertainty can be optimised if both RCMs and ESD are used in the downscaling stage.
Models’ minimum skillful scales is another issue important for regional climate modelling. What is it exactly? When I googled “minimum skillful scale” I got 228 hits so it doesn’t appear to be a much discussed topic. Meteorological phenomena such as El Niño Southern Oscillation (ENSO) have certain spatial patterns which the models may not reproduce perfectly. The spatial extent of such oscillations may be slightly too big or too narrow in the model results. Or the models may produce an incorrect shift in storm tracks or the zone of freezing temperatures (e.g. due to smooth topography), or a slight misrepresentation of the ITCZ (“double ITCZ”). Such details may not have a big global effect but are nevertheless important for the regional or local climate where they are located.
The models’ minimum skillful scales may be due to the models’ coarse spatial resolution (e.g. representation of ocean currents), a mix of imperfect representation of land properties, the use of discrete mathematics to solve mathematical expressions, imperfect algorithms, parameterisation of unresolved processes (e.g. clouds), and incomplete understanding of all relevant physical processes. The GCMs are designed to reproduce large-scale climatological phenomena, which they do impressively well, but not the local details.
The models’ minimum skillful scale is the reason for downscaling: we take the scales that the models can skillfully reproduce and the information on the dependency between the large and small-scale climate to infer the local changes. This is different to so-called bias-adjustment for which the scale dependency and the minimum skillful scale aren’t relevant.
Even if the models were perfect, their results are often limited by the law of small numbers. The reason is the pronounced presence of stochastic regional internal variations on decadal time scales (Deser et al., 2012). We can address such limitations though large model ensemble, but one question that has often been discussed is how to interpret multi-model ensembles of available models – an ensemble of opportunity.
There is no doubt a risk of mal-adaptation if one uses results from only one RCM or from downscaling of a small ensemble of distinct GCM simulations. Or if the numbers are interpreted incorrectly. One solution for overcoming the limitations connected to ensemble size may be through the use of ‘hybrid downscaling’ and ‘pseudo reality’ to emulate a large number of GCM projections. However, these results are as good as the RCM simulations at best and may also need bias-adjustment.
Climate adaptation needs to build on reliable, salient and relevant information, which requires a ‘distillation’ process that accounts for the models minimum skillful scales, multi-model ensembles, and different and independent sources of information. It will involve observations, different types of models, ensembles, evaluation, statistical theory (e.g. how numbers “behave” and how to deal with statistical samples) and artificial intelligence. There are also sensible actions such as making use of sensitivity tests to explore what is most important and how different choices (e.g. subselection of ensembles) affect the final outcome.
In summary, climate adaptation is not merely a “plug-and-play” process where data is downloaded from a data portal and somehow ensures optimal decisions. An analogy is that we don’t expect that a drug store is sufficient for people to get the right medicine, we also need ‘doctors’ who can provide proper guidance and consultation. Hence, some of the most important ingredients of climate services are the climate experts with up-to-date skills giving science-based advice to decision-makers.