Comparison of Heavy Rain Prediction with 1-hour and 2-hour Prediction Using Himawari-8 Data Based on LSTM Algorithm

Weather is one of the most unpredictable things in Indonesia. One of the factors is that Indonesia is an archipelago flanked by two large oceans, the Indian and Pacific Oceans. In addition, the fact that Indonesia is called the “land of the equator” creates uncertainty and rapid changes in weather movements. So, the Meteorology, Climatology, and Geophysics Agency (BMKG) still have difficulty accurately predicting the weather. This paper uses the Deep Learning method for weather data analysis and processing. The deep learning algorithm is the Long-Short Term Memory (LSTM) algorithm, often used to process time series data. This paper proposes a combination of 13 rain detection parameters with a comparison of 1 -hour and 2-hour heavy rain prediction models using LSTM. The resulting output with the accuracy level of the heavy rain prediction model for 2 hours has better accuracy than the 1-hour prediction model. The $\mathbf{2}$-hour prediction model is more recommended than the 1 -hour model, with the best prediction model results being a combination of parameters $1,2,4,10$, and 11 with an accuracy rate of $87 \%$. Further modeling using 3-hour modeling on parameter combinations $1,2,4,10$, and 11 resulted in an accuracy of $\mathbf{8 8 \%}$ on the new test data. Modeling for a long time provides promising, accurate results but can provide significant noise.