HOLT-WINTERS DALAM PREDIKSI ANOMALI OLR

Abstract

One of the pivotal study when analyzing the impact of the Madden-Julian Oscillation (MJO) phenomenon on rainfall anomalies in the Western part of Indonesia region is knowing the characteristics and the prediction of Outgoing Longwave Radiation (OLR) as the main parameter. This paper, focused to develop a predictive model OLR using the Box-Jenkins (ARIMA) method, which compared by the Holt-Winters method. This analysis focused on the average five days (pentad) OLR data in the Western part of Indonesia region, precisely at the position of 120oE for period of January 2007 to December 2009. Based on the comparison of the MSE (Mean Squares of Errors), and the MAPE (Mean Absolute Percentage Error) obtained from the Box-Jenkins (ARIMA) and Holt-Winters methods can be concluded that appropriate to predict the value of pentad OLR data on the position 120oE is the Box-Jenkins (ARIMA) method, because it has smaller value of MSE and MAPE comparing with the Holt-Winters method. Box-Jenkins model is the model ARIMA (1.0.1)(0.1.1)9 according to the results of significance analysis, autocorrelation in the residuals (white noise), also normality of residuals, where the process is repeated until model is really suitable and best meet the assumptions of model selection. The models can be assessed by the signal tracking value at each forecasting model. This assessment showed two of models in accepted boundary value, i.e ±5. This shows forecasting model can still be used to predict the 4 periods of the OLR pentad anomaly.

Keywords: Box-Jenkins (ARIMA), Holt-Winters, OLR, and Prediction Model

ABSTRAK

Salah satu kajian penting ketika menganalisis dampak dari fenomena Madden-