Drought is a natural disaster with widespread impacts on agriculture and water availability, particularly in the Gajah Mungkur Reservoir area of Wonogiri Regency, Indonesia. Rainfall instability driven by global climate change and local climate variability is the primary cause of this disaster. Accurate drought prediction is essential for formulating sustainable mitigation strategies. This study aims to analyze drought characteristics in the Gajah Mungkur Reservoir, Wonogiri Regency, using the Standardized Precipitation Evapotranspiration Index (SPEI) and to compare the performance of three prediction models: Long Short-Term Memory (LSTM), Bidirectional LSTM (Bi-LSTM), and Prophet in predicting SPEI. The dataset includes monthly rainfall and air temperature data from 1995 to 2024. The analysis reveals that longer SPEI time scales tend to show more temporally concentrated drought patterns. At the 6-month SPEI scale, which represents long-term drought, a total of 55 drought months were detected between 1995 and 2024, with major drought episodes occurring in 1996–1997, 2000–2007, 2019, and 2023–2024. Model performance evaluation shows a numerical trend in which Bi-LSTM outperforms others for 1-month SPEI prediction, while LSTM performs better at the 3- and 6-month scales. However, statistical significance testing indicates that the performance differences among the three models are not significant (p > 0,05), suggesting that other factors such as computational efficiency may be important considerations in practical applications.