Autocorrelation-A Simple Diagnostic for Tropical Precipitation Variability in Global Kilometer-Scale Climate Models

We propose the lag-1 autocorrelation of daily precipitation as a simple diagnostic of tropical precipitation variability in climate models. This metric generally has a relatively uniform distribution of positive values across the tropics. However, selected land regions are characterized by exceptionally low autocorrelation values. Low values correspond to the dominance of high frequency variance in precipitation, and specifically of high frequency convectively coupled equatorial waves. Consistent with previous work, we show that CMIP6 climate models overestimate the autocorrelation. Global kilometer-scale models capture the observed autocorrelation when deep convection is explicitly simulated. When a deep convection parameterization is used, though, the autocorrelation increases over land and ocean, suggesting that land surface-atmosphere interactions are not responsible for the changes in autocorrelation. Furthermore, the metric also tracks the accuracy of the representation of the relative importance of high frequency and low frequency convectively coupled equatorial waves in the models. Rainfall in the tropics is influenced by many atmospheric processes that depend on geographic location. We use the lag-1 autocorrelation as a metric for the day-to-day persistence of rainfall. We find that rainfall is very persistent in most parts of the tropics with a few exceptions over land, for example, the Sahel, where high frequency rainfall events dominate. Our results show that models with a horizontal resolution of a few kilometers reproduce the autocorrelation, in contrast to coarser climate models. We also analyze atmospheric waves and find that they are important for the autocorrelation pattern in the observations and the simulations. The lag-1 autocorrelation pattern of daily precipitation in the tropics is robust across different observation-based data sets The lag-1 autocorrelation reflects the relative variance of high frequency and low frequency convectively coupled equatorial waves Kilometer-scale models capture the observed autocorrelation, but models with parameterized deep convection overestimate it