Climate models are one of our most important tools to help us understand likely changes to extreme rainfall under a warmer climate. These models are complex
mathematical representations of the world’s climate system, and simulate the circulation of the earth’s atmosphere and oceans. They are closely related to the numerical weather prediction models which are used to develop weather predictions, except that they are run over longer timescales (often decades at a time) and therefore need to be run at coarser resolutions.
Given the widespread interest in using these models for developing projections of how extreme rainfall will change in the future, a large amount of effort is being devoted to diagnosing model performance, both to evaluate how good these models are at simulating historical rainfall patterns and to identify possible areas for improvement.
To this end, a colleague (Jason Evans) and I used a measure of sub-daily rainfall variability to evaluate the performance of a regional climate model in southeast Australia. As those living in the tropics will attest, rainfall patterns can vary significantly over the course of the day, with heavy showers often occurring in the late afternoon due to convective rainfall activity. This daily variability is known as the diurnal cycle, and occurs because of the close relationship between the daily cycle of atmospheric temperature and the associated rainfall.
On closer inspection, the diurnal cycle turns out to be quite a complex phenomena, with stronger diurnal variability during summer compared with winter, and in the tropics compared with the subtropics. Over the oceans, the diurnal cycle is often the opposite to that over land. Furthermore, ‘diurnal marches’ have been observed in the USA whereby an early afternoon rainfall maxima occurring near the coast will translate to later maxima further inland. This makes it a difficult metric for a climate model to ‘get right for the right reasons’.
When evaluating the Weather Research and Forecasting (WRF) model, we found that like most climate models, the frequency of rainfall occurrence was overestimated and the intensity was underestimated. When looking at diurnal variability, however, the model performed remarkably well. In particular the timing of the daily maximum and minimum occurrence (i.e. the probability of rainfall occurring), intensity (i.e. the intensity of rainfall when rainfall does occur) and total rainfall amount were similar between the model and the observations. The seasonality was also similar, with summer having a much stronger cycle compared with winter. And as shown in the figure above, the spatial distribution of the cycle was very accurate as well.
Studies like these add confidence that any rainfall projections from such models are credible. Having concluded that the model does reasonably well, we next investigated how and why the model simulated this diurnal rainfall variability. We looked at a range of different known precipitation triggering mechanisms, including atmospheric instability, thermal convection, topographic lifting, sea breezes, large-scale moisture convergence, frontal systems and tidal variations in atmospheric pressure. What we found was that atmospheric instability and thermal convection were most important – which makes sense given the link with the daily variation in land-surface and atmospheric temperature – as well as large-scale moisture convergence at 700 hPa.
We hope that studies like these will allow us to better understand the mechanisms which give rise to rainfall (including the most extreme rainfall events), and in doing so better understand both the current climate as well as the climate we might expect in the future.