Research brief: ARCCSS researchers improve confidence in regional climate model projections

A study involving researchers from the ARC Centre of Excellence for Climate System Science (ARCCSS) have taken a new approach on how to combine results from multiple climate models to test if it improves projections of our future climate. To do this they focused on maximum summer temperatures likely to be experienced over central North America, which currently show a large uncertainty.

Climate models have projected an increase of 3°C to 7°C (5°C average) in maximum summer temperatures in this region by 2100 under the highest concentration pathway (RCP 8.5).This wide range makes it very difficult for today’s policymakers to formulate responses to future temperature increases across central North America.

The reason for this range is that different climate models often produce different results when projecting future climate mainly due to variations in how different climate models get their results.

It has been suggested that giving increased weight to models whose historical simulations reproduce observations of our climate well will improve the accuracy of projections. This approach assumes climate models that perform well in historical simulations also have more skill in projecting future changes.

However, some climate models have a shared development history, which means they are likely to use very similar approaches to potentially get relatively similar projections. Simply taking the average of those dependent models has the potential to create biased projections.

Therefore, the researchers needed to introduce independence weights to obtain an independent set of models to test whether using this weighted ensemble increased the accuracy of projections. They then focused this approach on central North America’s summer maximum temperatures.

For their analysis, the researchers selected their models from the rigorously analysed Coupled Model Intercomparison Project phase 5 (CMIP5) using the criteria of model independence and accuracy in simulating historical climate.

This accuracy didn’t just relate to how each model simulated maximum summer temperatures in historical runs but included other “diagnostics” to ensure the accuracy of models. The researchers looked at climate influences that had the largest impact on maximum summer heat changes – trends in shortwave radiation (sunlight), average rainfall, changes in sea surface temperature and the general variability and trends in hottest days. Models were selected by comparing how closely these diagnostics and maximum summer temperatures matched real world observations in historical runs.

Accurate observational datasets were vital in making these determinations.

The researchers found that as more of these diagnostics were used to select independent and better models the more robust the results in the future became. Three different observational datasets were used and with more diagnostics the differences between using these various  observational datasets decreased, suggesting an improvement.

When these (weighted) climate models were then used to project changes in maximum summer temperatures out to 2100 for central North America, it saw the average decrease fractionally by 0.05°C - 0.45°C.

While the projection changes for central North America were relatively small, the researchers were able to conclude that model independence, the selection of multiple variables from historical runs and accurate real world observations of these combined variables together improved confidence in climate projections at a regional scale.

This opens further areas for investigation around the ideal number of diagnostics required to improve projection accuracy and yet again stresses the importance of the need for ongoing, high quality observational data to accurately project future climate states.

UNSW logo ANU logo Monash logo UMelb logo UTAS logo