DROUGHT PREDICTION USING LSTM MODEL WITH STANDARDIZED PRECIPITATION INDEX ON THE NORTH COAST OF CENTRAL JAVA

Fluctuating weather can trigger hydrometeorological disasters, especially affecting farmers and fishermen on the north coast of Central Java. Weather predictions including drought are very important to anticipate drought disasters. Deep learning-based prediction models such as Long Short Term Memory (LSTM) are used in an effort to reduce the impact of drought. The purpose of this study is to prove the level of accuracy of the LSTM model and determine the drought index with the Standardized Precipitation Index (SPI). The LSTM model is used to predict drought based on the SPI, while the SPI acts as a drought index that considers precipitation (rainfall) for a period of 1, 3, and 6 months. Predictions use rainfall data obtained from online data from the Central Java BMKG UPT Indonesia for the period 2010-2023 in the Tegal City and Semarang City station areas. The results of data treatment with LSTM can effectively analyze and capture complex patterns in meteorological data to predict drought events accurately. The effectiveness of the model is shown by the relatively small MAE and RMSE results, namely MAE 0.163 - 0.352 and RMSE 0.247-0.515. The best prediction result is the 3-month SPI in the Semarang area with MAE 0.163 and RMSE 0.274. While the prediction result with the largest error is the 1-month SPI in the Tegal area. Drought modeling using LSTM has been successfully implemented for the northern coast of Central Java using the Streamlit Framework and can process and visualize the drought prediction system well.