Analysis of fragmented time directionality in time series to elucidate feedbacks in climate data
Résumé
Identifying and characterizing feedbacks in environmental processes may help in improving predictions for some environmental systems. The statistical study of time series is a manner to approach these feedbacks. Here, we consider feedbacks that are induced by occasional extreme events and that locally disturb the probabilistic behavior of climate time series. For example, intense rainfalls may induce biophysical feedback processes and, consequently, influence the occurrence or intensity of daily rainfalls afterwards. In this article, we associate such eventual perturbations in time series to the concept of fragmented time directionality, and we present the R package FeedbackTS that contains a statistical exploratory toolbox for investigating fragmented time directionality. The toolbox mostly consists of simple randomization tests. The use of the package is illustrated with historical Australian rainfall data: we show the existence of feedback and identify temporal and spatial variation in feedback.