Predictions of onset and withdrawal using Markov chain in Semarang residency

Abstract

As part of the monsoon region, the prediction of the onset of the rainy and dry seasons in Indonesia is very important. The purpose of this paper is to predict the onset of the rainy season and the withdrawal of the dry season using Markov Chains. The prediction test uses rainfall data from the Semarang Residency, which is a highly developed but flood-prone area, namely Demak Regency, Semarang, Kendal and Semarang City. The results of the study show that 17% of the total data did not have a defined beginning for both the rainy and dry seasons, and this condition was excluded from the analysis. Meanwhile, 83% of the data that had onset and withdrawal were found that the average deviation of dry season prediction (AMK) was -2 decades or 20 days late which occurred in Demak, Kendal, and Semarang City. While for Semarang Regency the value is 0 decades or the accuracy is very precise. However, 0 decades does not mean exactly on the same day because the determination of decades in Indonesia uses a 10-day accumulated rainfall parameter so it is looser compared to other countries, especially India. This study also found that the accuracy of onset and withdrawal predictions differed based on elevation. In the withdrawal prediction, the median accuracy was 56%, but the values were 52%, 71%, and 54% in Semarang, Semarang Regency, Kendal. While for onset, the accuracy averaged 59%, with regions like Kendal and Semarang Regency having more precise predictions. The errors in onset predictions were balanced between early and late predictions. The study highlights that while predictions in some regions were generally accurate, there were notable outliers and instances of predictions significantly deviating from the observed data, suggesting room for improvement in forecast models.