Long Temporal Autocorrelations in Tropical Precipitation Data and Spike Train Prototypes

Abstract. Temporal precipitation autocorrelations drop slower than exponentially at long lags, and there is a range from tens to thousands of minutes where it is relevant to ask if a scale-free process might underlie the long autocorrelations. A simple stochastic model in which precipitation appears as variable-length spikes provides a reasonable prototype for this behavior. In both observations and the model, separating the component of the autocorrelation within wet events from the inter-event contribution suggests long auto-correlation behavior is primarily associated with the latter. When precipitation spikes are short compared to dry events, a true power law is obtained with analytical exponent -0.5 and precipitation autocorrelation is determined by dry-spell model parameters. In more realistic cases, wet-spell termination is also important. Although a variety of apparent power law exponents can be obtained for different parameters, the fundamental long-lag process appears to be that of the inter-event correlation.

Citation:
Abbott, T. H., S. N. Stechmann and J. D. Neelin, 2016: Geophys. Res. Lett., 43, doi:10.1002/2016GL071282.