Most climate models agree that high levels of atmospheric carbon will increase precipitation intensity—by an average of approximately 6 percent per degree temperature rise. These models also predict an increase in total precipitation; however, this growth is smaller, only 1 percent to 2 percent per degree temperature rise.
Understanding changes in storm behavior that might explain this gap have remained elusive. In the past, climate simulations were too coarse in resolution (100s of kilometers) to accurately capture individual rainstorms. More recently, high-resolution simulations have begun to approach weather-scale, but analytic approaches had not yet evolved to make use of that information and evaluated only aggregate shifts in precipitation patterns instead of individual storms.
To address this discrepancy, postdoctoral scholar Won Chang (now an assistant professor at the University of Cincinnati) and co-authors Michael Stein, Jiali Wang, V. Rao Kotamarthi and Moyer developed new methods to analyze rainstorms in observational data or high-resolution model projections. First, the team adapted morphological approaches from computational image analysis to develop new statistical algorithms for detecting and analyzing individual rainstorms over space and time. The researchers then analyzed results of new ultra-high-resolution (12 km) simulations of U.S. climate performed with the Weather Research and Forecasting Model at Argonne National Laboratory.
Analyzing simulations of precipitation in the present (2002-2011) and future (years 2085-2094), the researchers detected changes in storm features that explained why the stronger storms predicted didn’t increase overall rainfall as much as expected. Individual storms become smaller in terms of the land area covered, especially in the summer. (In winter, storms become smaller as well, but also less frequent and shorter.)
“It’s an exciting time when climate models are starting to look more like weather models,” Chang said. “We hope that these new methods become the standard for model evaluation going forward.”
The team also found several important differences between model output and present-day weather. The model tended to predict storms that were both weaker and larger than those actually observed, and in winter, model-forecast storms were also fewer and longer than observations. Assessing these model “biases” is critical for making reliable forecasts of future storms.
“While our results apply to only one model simulation,” Moyer said, “we do know that the amount-intensity discrepancy is driven by pretty basic physics. Rainstorms in every model, and in the real world, will adjust in some way to let intensity grow by more than total rainfall does. Most people would have guessed that storms would change in frequency, not in size. We now have the tools at hand to evaluate these results across models and to check them against real-world changes, as well as to evaluate the performance of the models themselves.”
New precipitation forecasts that include these changes in storm characteristics will add important details that help assess future flood risk under climate change. These results suggest that concerns about higher-intensity storms causing severe floods may be tempered by reductions in storm size, and that the tools developed at UChicago and Argonne can help further clarify future risk.
Citation: “Changes in spatio-temporal precipitation patterns in changing climate conditions,” Journal of Climate, Dec. 1, 2016. doi/abs/10.1175/JCLI-D-15-0844.1
Funding: National Science Foundation