<![CDATA[An accurate rainfall prediction is essential to help mitigate natural disasters that may disrupt human activity. Diversity of the Indonesian climate makes it harder to predict rainfall with high accuracy, and this situation also occurs in South Sumatera. Therefore, this research aims to assess Long Short-Term Memory (LSTM) model to predict dekadal rainfall at per-coordinate level, spanning 18 dekads ahead, and across 2821 gridded coordinates in South Sumatera. The model also uses climate indices (Southern Oscillation Index (SOI), Dipole Mode Index (DMI), Niño Sea Surface Temperature (SST) anomaly) and wind streamline, together with South Sumatera dekadal rainfall to enhance the model predictive performance. The best performing model achieved by data that includes wind streamline at 925 millibars, with per-month performance of Root Mean Squared Error (RMSE) between 31.25 to 56.50 and rainfall level category accuracy between 53.26% to 78.47%. This indicates the LSTM model is able to predict dekadal rainfall on multiple coordinates throughout South Sumatera, although requiring significant improvement to be able to be incorporated to main weather prediction system.]]>
Copyrights © 2026
