Deep learning of backpropagation neural network algorithm for long-term predicting rainfall in the Kapuas Hulu, West Kalimantan province of Indonesia

Climate change and global warming significantly impact rainfall patterns in various regions. This can lead to more frequent and intense flooding and an increased risk of landslides. As a result, it causes unstable rain variability patterns in various regions, including Kapuas Hulu, West Kalimantan Province, Indonesia. Almost every year, the area experiences floods and landslides. The area, directly adjacent to the Indonesia-Malaysia region, can potentially disrupt community activities, including military operations guarding the border, which require a lot of manpower. This study aimed to minimize future disasters as it is vital to anticipate rainfall patterns based on previous data from databases. The Backpropagation Neural Network (BPNN) approach is one of the best at predicting long-term rainfall. Rainfall data from NASA was utilized from January 2003 through December 2020, totalling 216 data sets. The input or training data ranges from January 2003 to December 2010, whereas the training goal data is from January 2011 to December 2015. The validation data was also determined from January 2016 to December 2020. With a learning rate of 0.3 and an Epoch of 9,999, the best predictive architecture model was 8-6-9-6-5. The prediction accuracy was pretty excellent, with a mean square error (MSE) of 0.012157 and a mean absolute percentage error (MAPE) of 24.026. The highest rainfall was recorded in December 2019 at 606.672 mm/month. The prediction results are expected to serve as a reference for mitigating disasters such as floods and landslides to facilitate security operations in border areas.