Factsheet 14: Model evaluation
Before using climate models to understand the impacts of future climate change, we need to evaluate how well they simulate the present-day climate. This is important to provide confidence that the models capture regional climate features and are suitable for understanding future climate impacts. Model evaluation also highlights areas where future model development can improve. The climate models available on the Queensland Future Climate Dashboard have all been thoroughly evaluated.
This factsheet provides an overview of the evaluation of the downscaled CMIP6 models. A full description is available in this peer-reviewed research paper.
How did we evaluate the models?
Climate models produce a range of different outputs that can be evaluated. As part of our evaluation, we examined mean climate, extreme climate, and seasonal cycles. The dynamically downscaled models were assessed by comparing them to observational data from the Australian Gridded Climate Dataset (AGCD) and to their host global climate model to evaluate whether the downscaling improved the representation of climate. Note that he AGCD is a dataset developed by the Bureau of Meteorology based on daily temperature and precipitation measured at weather stations. We used the period from 1981 to 2010 to evaluate performance and used a number of statistical measures, including bias, correlation, variability, and the Perkins skill score. These statistical measures tell us how similar the mean climate and climate extremes in the models are to the observations and how much the climate varies over time.
We also calculated 'added-value', which is how much improvement there is in performance in the downscaled high-resolution models compared to the coarse-resolution host models.
How well did the models perform?
We found the high-resolution downscaled climate models improved on the performance of the global climate models, with improvements particularly noticable for the extremes and in mountainous and coastal areas. The global climate models generally represent the mean climate well, but are limited when it comes to capturing climate extremes, and downscaled models tend to outperform them. The downscaled models generally show an improvement on extreme high and low temperatures, the number of rainy days, and extreme precipitation. The downscaled models also generally improved on the representation of the seasonal cycle. We found these improvements across all of Australia (Figure 1).
How well did the models perform in Queensland?
In Queensland, the downscaled models added value over the host models in all regions. The largest improvements were found in the South East Queensland region. This evaluation shows that the downscaling of the models improved performance for temperature and precipitation, and this dataset should be particularly useful when looking at climate extremes and when preparing regional climate hazard assessments.
