The impact of stochastic physics on tropical rainfall variability in global climate models on daily to weekly time scales

Many global atmospheric models have too little precipitation variability in the tropics on daily to weekly time scales and also a poor representation of tropical precipitation extremes associated with intense convection. Stochastic parameterizations have the potential to mitigate this problem by representing unpredictable subgrid variability that is left out of deterministic models. We evaluate the impact on the statistics of tropical rainfall of two stochastic schemes: the stochastically perturbed parameterization tendency scheme (SPPT) and stochastic kinetic energy backscatter scheme (SKEBS), in three climate models: EC-Earth, the Met Office Unified Model, and the Community Atmosphere Model, version 4. The schemes generally improve the statistics of simulated tropical rainfall variability, particularly by increasing the frequency of heavy rainfall events, reducing its persistence and increasing the high-frequency component of its variability. There is a large range in the size of the impact between models, with EC-Earth showing the largest improvements. The improvements are greater than those obtained by increasing horizontal resolution to approximate to 20km. Stochastic physics also strongly affects projections of future changes in the frequency of extreme tropical rainfall in EC-Earth. This indicates that small-scale variability that is unresolved and unpredictable in these models has an important role in determining tropical climate variability statistics. Using these schemes, and improved schemes currently under development, is therefore likely to be important for producing good simulations of tropical variability and extremes in the present day and future. Plain Language Summary Simulations from climate models have been found to lack day-to-day variability in tropical rainfall, with there being too many rainy days and not enough days with very heavy rainfall. A possible contributor to this problem is that the schemes the models use to predict rainfall try to predict the average rainfall that would be expected for given large-scale conditions. In reality, unpredictable small-scale features like eddies and gravity waves may contribute to the formation of severe storms or prevent them from developing. We test whether using stochastic methods to represent the effectively random impact of these small-scale features improves the variability of tropical rainfall simulated by three climate models. We find evidence that it does, and this indicates that treating the prediction of tropical rainfall probabilistically rather than deterministically will give improvements in climate simulations.