Evolving drought dynamics in Barcelona: leveraging a Bayesian ensemble algorithm for insightful analysis and a bidirectional long short-term memory network for predictive modeling

Drought dynamics, intensified by climate change, pose critical challenges to sustainable water resource management, particularly in Mediterranean regions such as Barcelona, Spain. Addressing these challenges requires tools capable of detecting evolving trends and predicting future conditions with precision. This study investigated drought trends using the Standardized Precipitation Index (SPI) at 12- and 24-month scales, employing advanced methodologies to overcome limitations in traditional analyses. Specifically, the Seasonal Kendall (SK) test was applied to identify long-term trends, while the Bayesian Changepoint Detection and Time Series Decomposition (BEAST) algorithm was utilized to detect abrupt changes in SPI patterns. This dual approach revealed an intricate scenario: while long-term trends suggested an increase in precipitation, recent decades demonstrated short-term positive shifts coexisting with prolonged negative trends, indicative of worsening drought conditions. To complement the trend analysis, a bidirectional Long Short-Term Memory (Bi-LSTM) network was developed for multi-step ahead SPI forecasting, achieving high accuracy for short prediction horizons (R2 up to 0.899 for SPI-24). This integration of BEAST and Bi-LSTM represented a novel framework for capturing the complexities of drought evolution, addressing both trend detection and forecasting in a unified approach. By applying this methodology to a city experiencing acute water stress, the study provided innovative tools and actionable insights for enhancing drought preparedness and resilience. These findings underscored the importance of advanced data-driven methods in mitigating the socio-economic impacts of evolving drought dynamics.