<![CDATA[Indonesia’s complex tropical climate, diverse topography, and strong influence from large-scale climate variability such as ENSO and MJO pose persistent challenges for accurate weather prediction. This study reviews recent developments in weather forecasting research in Indonesia, focusing on model approaches, predictor variables, and performance across different climatic regions. The literature shows a clear transition from conventional statistical methods, such as ARIMA and regression models, toward machine learning, deep learning, hybrid, and ensemble frameworks. Deep learning models (e.g., LSTM, Bi-LSTM, and CNN) generally achieve higher accuracy when sufficient quality data are available, while hybrid and ensemble methods improve robustness under heterogeneous climatic and data conditions. Key predictors include rainfall, temperature, humidity, wind variables, atmospheric pressure, satellite-based products, and climate indices such as Niño3.4 and the Dipole Mode Index. Performance is strongly influenced by regional climate characteristics, data availability, and model configuration. Despite these advances, major limitations remain, including uneven observational coverage, challenges in extreme weather prediction, sensitivity to sparse and noisy data, high computational demands, and limited operational deployment. These findings highlight the need for adaptive, computationally efficient forecasting frameworks and stronger integration between data infrastructure and real-time operational systems.]]>
