Nontrivial scaling in the loss of prediction information with aggregation in hourly precipitation occurrences

Predicting the occurrence of rainfall from past patterns of rain/no rain sequences is an issue that has recently regained attention through the application of information theory and dynamical systems, claiming the existence of an underlying complexity of deterministic origin. The present work reports a rather unexpected facet that appeared in the study of hourly precipitation occurrence pattern prediction: the temporal scale invariance of the probability of prediction failure for scales up to the order of magnitude of a storm duration.