Even though there is no doubt that global temperatures are rising due to the anthropogenic emission of greenhouse gases, for other types of climate hazards – such as certain types of flooding or droughts – it is less clear what the future might bring. Imagine 100 scientists in a room and they all say “It’s going to be wetter”, “It’s going to be dryer”, “It is going to be this, it will be that”. Who should you listen to and what should you do? How do you make a decision?
Global climate models (GCMs) are often used to explore future conditions, but the variability of projections among GCMs at the local scale complicates regional climate risk assessments. This variability in future projections is only partly explained by the often-used emission scenarios: model uncertainty and internal variability play a major role in the outcomes of projected meteorological conditions, especially for local precipitation patterns. As precipitation is a key driver for hazards such as floods, droughts, and wildfires, local assessment of resulting risks using emission-based multi-model means probably leads to limited impact exploration.
In a paper published in the Environmental Research: Climate journal, Ted Buskop, Frederiek Sperna Weiland and Bart van den Hurk (Deltares) propose a method to select more impact-relevant scenarios by determining regionally relevant climatic impact drivers (CIDs) and clustering GCMs on their projected changes in these drivers. Taking the flood risk in the Latvian Lielupe basin as a case study, the authors quantify the effectiveness of this approach, expressed as an “exploratory amplification” factor, by comparing future impacts covered by multi-model means per emission scenario.
“There are about 30 different global climate models. If you look at the impacts these climate models generate on the ground in terms of flooding, you see that one model will say your flooding will decrease by 30%, another one by 20%, and so on. And you also have models that will say your floodings will increase 30% or 20% and everything in the middle” points out Buskop. All of these climate models are validated by climate scientists and each one can tell us something about the future. But we still end up with a huge variation in what the future might bring.
Indeed, models run at different emission scenarios, and that creates some signal in how impacts might change. However, there are also differences across these models that are programmed differently to take into account various physical aspects. “These models are also dependent on natural variability. This means that some of these include, just because of randomness, a bunch of dry days in their data set while other models have a lot. This fluctuation also influences the fluctuation in the range of future impacts” continues Buskop.
The Lielupe basin is small and, therefore, subject to large internal variability in the observed and modelled weather, getting a climate signal can be challenging. Instead of basing the future weather just on emissions, the authors based their scenarios on what matters locally. Flooding in this region is really dependent on specific CIDs, such a spring and summer time precipitation amounts and their intensities.
“With those climate variables, we go to these climate models and ask «Climate model one, what do you think is the change in this climate variable?» then we ask the next one, and so on. And based on that, on their climate model behavior locally, we create scenarios that allow exploring a lot more of what the future might bring in terms of impact than if you would simply aggregate all projected climate signals based on the emissions” explains Buskop.
Indeed, if you simply aggregate all the information based on the emissions, you may get a change of zero percent of rainfall in the region. Half of the models, for example, may predict less rain in the summer but the other half would say the opposite. “This does not depend on the emission scenario but on how the models are structured, coded, and which physical processes they take into account. But if you separate models according to their behavior, you will find two distinct groups of dry and wet models that have a lot of a clearer signal of what the future might bring” comments Buskop.
The method results in locally relevant climate scenarios that significantly improve regional exploration of future climate impacts. Such scenarios provide targeted risk information that can be used in adaptation planning: each synthesis of each model bucket is a separate climate scenario that can be used to assess future discharges, extremes, and flood changes in a specific region.
