Neural Network and Principal Component Analysis on Statistical Downscaling for Local Rainfall Forecasting

Abduh Riski, Alfian Futuhul Hadi, Okit Tazkiyah, Dian Anggraeni

This paper talks about Principal Component Analysis (PCA) and Neural Network (NN) in the Statistical Downscaling (SD) scheme forecasting the local rainfall at Jember regency, East Java, Indonesia for the next 2019-2020. The SD connected the global scale of climate conditions in the form of General Circulation Model (GCM) data relates to the smaller local scaled observed data based on a mathematical functional relationship. The GCM data known as the computer-based model simulated the global climate variables in each grid and atmospheric layer. Then by borrowing the strength of the functional relationship model, the SD extrapolates the predicted value to forecast the local data for the long-term next period ahead. The GCM-SD was available for long-term climate patterns and considered to provide a higher resolution. Firstly, we were determining the best grid size for each PCA and NN in the training set according to the Root Mean Square Error (RMSE) of the SD scheme framework. Finally, we develop an alternative of PCA-NN combination in the SD scheme. Even though its RMSE is not the lowest one, we prefer PCA-NN because it was potential to be further developed for extreme values forecasting. According to our PCA-NN forecasting, we will have a dry season for four months from June until September for both 2019 and 2020.

Volume 12 | 02-Special Issue

Pages: 821-826

DOI: 10.5373/JARDCS/V12SP2/SP20201138