(collaborative research) The investigators will implement a stochastic backscatter scheme into the Community Atmosphere Model and explore how to improve the internal variability and, in particular, the prediction of extreme events on decadal and regional scales. Only a small number of publications apply Extreme Value Theory to climate models and many questions remain open. While some work has been conducted comparing model and reanalysis with constant greenhouse forcing, the major body of published work in this area focuses on the occurrence of extreme events in a changing climate and on the robustness of climate trends of extreme events across di^erent low-resolution models. The emphasis of the investigators is very di^erent: they will look at the internal variability of models under a constant greenhouse forcing. The investigators focus both on the ability of low-resolution climate models to realistically predict extreme events on decadal time-scales and global spatial scales, and on the feasibility of replacing the missing variability due to low-resolution with a stochastic model error scheme and if such a scheme can improve the decadal prediction of extreme events. Model integrations with and without a stochastic backscatter scheme would be conducted and extreme value statistics be used to determine the impact of the scheme onto the occurrence of extreme events. For comparison, the same statistic would be computed using the the ERA40, the ERA-Interim analysis and/or the NCEP/NCAR reanalysis as best proxy for multi-decadal observations.
Objectives of Research: This grant funds interdisciplinary research to address the following broad questions: What is the likelihood of extremely hot temperature in summer and extremely cold temperatures in winter? How do the extrema vary over land and ocean and with the seasons? Are climate models able to predict these extreme temperatures in a statistical sense, i.e. are they able to predict how long it will take on average until a certain threshold temperature is exceeded. To which degree are droughts and hot temperatures linked and does knowing one help predict the other? Project Findings: We compared the statistics of extremes in observations and climate simulations from the Community Atmosphere Model (CAM) general circulation model. All models have errors and one way to represent model error is by stochastic parameterizations. Here we have studied the effects of a particular stochastic parameterization, namely a stochastic kinetic-energy backscatter scheme. These schemes are routinely used to assess uncertainty in numerical weather prediction, but their use in climate models is a recent development. The impact of this model-error representation was studied with regard to mean climate error, climate variability and extreme events. We find that adding a stochastic parameterization improves the low-frequency variability in the Northern Hemisphere without introducing large biases. The model with and without stochastic parameterization has large errors in the mean climate. However when these are subtracted, the climate model simulations capture climate variability and extreme hot and cold temperatures very well. Some remaining discrepancies over Australia and Alaska are significantly reduced in the simulations with stochastic perturbations.
