Statistical Downscaling of GCM Data using Support Vector Regression to Predict Monthly Rainfall in Indramayu.

Abstract

The knowledge about weather and climate pattern, especially rainfall is required by many sectors such as, agricultural, plantation, transportation, and so on. Recently, the model that is used to observe the impact and to predict the climate change is the global circulation model (GCM). However, the current GCM-resolution data is too low and it is difficult to predict the local climate pattern that requires high resolution. However, it is possible to obtain regionalscale information if it is combined with statistical downscaling (SD) methods. The main objective of this research is to develop SD models by using support vector regression (SVR) in forecasting monthly rainfall in Indramayu, in order to get the accurate climate information based on GCM data. The results of this research showed that overall, the model is good enough to predict rainfall with normal conditions. But for extreme circumstances, although the prediction model was able to follow the pattern of the observational data, but the value of the resulting predictions have not managed to predict accuratelly the actual observed values. Variations of the evaluation of estimated results by the model for 13 rain stations in Indramayu district showed that the location of rain stations in certain areas can influence and determine the accuracy of the monthly rainfall forecast. Rain observations points located farther from the sea tend to have better estimation results than the observation points located close to the sea.

Iklim merupakan gejala alamiah yang sangat penting dan berpengaruh bagi kehidupan manusia. Pengetahuan tentang pola cuaca dan iklim terutama curah hujan, sangat dibutuhkan di banyak sektor seperti pertanian, perkebunan, transportasi, dan lain-lain. Untuk mengetahui dan memahami sistem iklim sehingga dapat digunakan dalam memprediksi jumlah curah hujan bulanan di suatu daerah, diperlukan suatu model/alat yang dapat menyimulasikan iklim. Sampai saat ini, model/alat yang digunakan dalam kajian utama untuk mempelajari dampak dan menduga perubahan iklim adalah global circulation model (GCM). Akan tetapi resolusi dari data luaran GCM yang dianggap terlalu rendah menyulitkan dalam melakukan prediksi dengan mempelajari pola iklim regional/lokal yang membutuhkan resolusi yang tinggi. Namun GCM masih mungkin digunakan untuk memperoleh informasi skala regional hingga lokal bila dipadukan dengan teknik statistical downscaling (SD).