Projection Pursuit Regression on Statistical Downscaling Using Artificial Neural Network and Support Vector Regression for Rainfall Forecasting in Jember

DOI

10.2991/acsr.k.220202.038How to use a DOI?

Keywords

General Circulation Model (GCM); Statistical Downscaling (SD); Projection Pursuit (PP); Projection Pursuit Regression (PPR); Artificial Neural Network (ANN); Support Vector Regression (SVR)

Abstract

Information about rainfall is very necessary for the country of Indonesia which bears the title of an agricultural country. This is because the agricultural sector is very vulnerable to climate change, where rainfall is one indicator of climate change-related to crops. Therefore, an accurate rainfall forecasting model is needed to assist farmers in determining planting time, cropping patterns, and others by utilizing information from GCM outputs. However, the information provided by GCM is still on a global scale and has a low resolution for local scale forecasting. However, GCM output information can still be utilized by using statistical downscaling techniques. Statistical Downscaling is a technique that connects GCM output as a predictor variable with local rainfall in Jember Regency as a response variable with the intermediary of a functional model. The response variable, namely local rainfall in Jember, was taken from January 2005 to December 2018 with a total of 168 data. As for the GCM output response variables, there are three types of variables used in this study, namely precipitation, sea surface pressure, and air temperature with a 3×3 domain to a 10×10 domain. The two data will be split with data from January 2005-December 2017 as training data to build the model and data from January 2018 to December 2018 as testing data used for model validation. In this study, rainfall forecasting in Jember Regency was carried out using two combined methods, the first method was Projection Pursuit Regression followed by the Artificial Neural Network method. For the second method, using the projection results from PPR as a dimension reducer of a large predictor variable, namely PP and followed by the Support Vector Regression algorithm. At the modeling stage with PPR, the optimum domain and many functions will be determined, where the chosen domain is a 6×6 domain and the number of optimum functions is m=5. Furthermore, it will be modeled using two rainfall forecasting methods, namely ANN and SVR. The results of model validation using RMSE show that the PP+SVR method has a smaller RMSE value of 65.61 compared to the PPR+ANN method with an RMSE value of 67.48. This shows that the performance for the PP+SVR model is better than the PPR+ANN model.

Copyright

© 2022 The Authors. Published by Atlantis Press International B.V.

Open Access

This is an open access article under the CC BY-NC license.

Download article (PDF)

TY  - CONF
AU  - Chandrika Desyana Putri
AU  - Ema Fahma Farikha
AU  - Alfian Futuhul Hadi
AU  - Yuliani Setia Dewi
AU  - I Made Tirta
AU  - Firdaus Ubaidillah
AU  - Dian Anggraeni
PY  - 2022
DA  - 2022/02/08
TI  - Projection Pursuit Regression on Statistical Downscaling Using Artificial Neural Network and Support Vector Regression for Rainfall Forecasting in Jember
BT  - Proceedings of the  International Conference on Mathematics, Geometry, Statistics, and Computation (IC-MaGeStiC 2021)
PB  - Atlantis Press
SP  - 204
EP  - 210
SN  - 2352-538X
UR  - https://doi.org/10.2991/acsr.k.220202.038
DO  - 10.2991/acsr.k.220202.038
ID  - Putri2022
ER  -