Probabilistic forecasting of drought class transitions in Sicily (Italy) using Standardized Precipitation Index and North Atlantic Oscillation Index

Abstract

Summary Since the mid-90s the Standardized Precipitation Index (SPI) has found widespread use to monitor drought periods at different time scales. Recently, some efforts have been made to analyze the role of SPI for drought forecasting, as well as to estimate transition probabilities between SPI drought classes. In the present paper probabilistic models for short and middle term forecasting of SPI drought class transition probabilities are presented and extended in order to include information provided by an exogenous variable, such as an index of large scale atmospheric circulation pattern like, for instance, the North Atlantic Oscillation index (NAO). In particular, the proposed models result from evaluating conditional probability of future SPI classes with respect to current SPI (and NAO) classes or current SPI (and NAO) values, under the hypothesis of multivariate normal distribution of the underlying joint variables. SPI series are computed on average areal precipitation in Sicily region (Italy). As a significant negative correlation exists between NAO and SPI series in Sicily during recent decades, the proposed models are calibrated on the period from 1979 to 2008. Both SPI and NAO values are categorized in four classes. Transition probabilities to future SPI classes are evaluated based on SPI and NAO current classes or values and compared to the corresponding probabilities when NAO is neglected. Results indicate that drought transition probabilities in Sicily are generally affected by NAO index. In particular, transition probabilities related to persisting or worsening drought conditions significantly increase as NAO index tends toward extremely positive values. On the other hand transition probabilities to a less severe drought class decrease as NAO values increase. Furthermore, application of a simple score approach to quantitatively assess the skill in forecasting of the proposed models shows that assessing transition probabilities to future SPI classes from current SPI and NAO values leads to better results than considering current classes.