Performance Evaluation of a Simple Feed-forward Deep Neural Network Model Applied to Annual Rainfall Anomaly Index (RAI) Over Indramayu, Indonesia

Indramayu is a district in West Java that is known for being the leading producer of rice and brackish salt. The production of these two commodities is strongly influenced by hydroclimatological conditions, making accurate and reliable long-term estimates crucial. In this study, we evaluated a simple feed-forward deep neural network (DNN) model that could potentially be used as a candidate for statistical guidance to improve the accuracy of a mesoscale numerical climate model. We used the spatial average of the accumulated annual rainfall of the Climate Hazards Group InfraRed Precipitation with Station (CHIRPS) data as an input time series with a time range from 1981 to 2022. This data was then processed into annual rainfall anomaly index (RAI) data. The Annual RAI was divided into training and test sets, and the feed-forward DNN model was fitted to the annual RAI in the training set. The accuracy of the model was then tested in the test set using the root-mean-square error (RMSE) metric. Our study shows that the feed-forward DNN model is unsuitable for estimating the annual RAI over Indramayu. The RMSE values are significantly high in the training and test sets.