Evolving Drought Dynamics in Barcelona: Leveraging a Bayesian Ensemble Algorithm for Insightful Analysis and a Bidirectional Long Short-Term Memory Network for Predictive Modeling

In recent times, the growing influence of climate change has emphasized the significance of examining hydrological patterns for efficient planning and management of water resources. This study proposes an investigation of the Standard Precipitation Index (SPI) trends and abrupt changes, at time scales of 12 and 24 months, for the municipality of Barcelona, Spain. The overall trend of SPI was assessed based on the seasonal Mann-Kendall (MK) test. The severity and duration of drought events, considering the entire time series and twenty-year intervals from 1820–1840 to 2000–2020, were also evaluated. Then, the Bayesian Changepoint Detection and Time Series Decomposition (BEAST) algorithm was employed to identify abrupt changes in trend along the SPI time series. The seasonal MK analysis reveals a rising trend, indicating a positive shift in precipitation patterns over time. On the other hand, the BEAST analysis presents a more intricate scenario, where recent decades demonstrate a simultaneous presence of short-term positive shifts alongside prolonged negative trends, indicating a shift toward drought conditions. Furthermore, the effectiveness of a Bi-LSTM-based model in forecasting the SPI with a temporal horizon of up to 6 months was evaluated. The forecasting model displayed a decline in performance as the forecasting horizon extended, with the most precise predictions achieved for a 1-month lead time, with R ² up to 0.899 for SPI-24.