Temperature (maximum) and Rainfall Prediction Using FNN and RNN Models over Kaba Aye Station, Yangon Region, Myanmar

Rainfall and temperature are essential roles in weather forecasting in a tropical country like Myanmar. In this study, we explored data-driven feed-forward neural networks (FNN) and recurrent neural networks in terms of long short-term memory (RNN-LSTM) utilized for predicting temperature (maximum) and rainfall at Kaba Aye station. We trained both FNN and RNN-LSTM models separately to predict, and explored the models’ prediction capabilities. The FNN models are trained with 9 years of monthly data which includes temperature (maximum, and minimum), rainfall, relative humidity, pressure, and nino1.2. Whereas, the LSTM-based RNN model is trained with 9 years of monthly data which includes monthly maximum temperature and rainfall. The result of the outputs was the monthly maximum temperature and rainfall of the next year. respectively. For the given station, the expected monthly maximum temperature and rainfall in each month of 2010 are predicted in this research. The results of temperature prediction show that the RNN-LSTM model provided better results with the mean absolute error (MAE=1.33), mean squared error (MSE=2.57) and root mean squared error (RMSE=1.6). Generally, both FNN and RNN-LSTM models examined overestimate in rainfall, however, the FNN model performed better with (MAE=70.6), and (RMSE=107.08)