A Bayesian Quantile Clustering Approach of Spatio-Temporal Climate Time-Series

In the face of climate change, the profound transformation in the distribution of climate variables, including temperature and precipitation, goes beyond mere changes in their mean values. These complex changes underscore the need for robust methods capable of capturing the nuanced spatial and temporal patterns that emerge. Traditional clustering algorithms that rely solely on measures of central tendency risk misinterpretation and produce erroneous results when applied to such dynamic datasets. In response, we present a new approach to spatiotemporal clustering that utilizes quantiles within a Bayesian framework. Central to our method is the incorporation of a spatial dependence layer based on a Markov random field, which allows for a more nuanced understanding of the underlying patterns. We provide illustrative examples of the application of our method, focusing on the monthly precipitation in the northeast of Italy, a region affected by the ongoing effects of climate change. We perform different classifications according to the quantile of interest to provide valuable insights into the evolving dynamics of climate change of precipitation regime. Our work can improve the understanding of climate variability and facilitate more informed decision-making in adaptation and mitigation efforts.