PRAKIRAAN CUACA DENGAN METODE AUTOREGRESSIVE INTEGRATED MOVING AVERAGENEURAL NETWORK, DAN ADAPTIVE SPLINES THRESHOLD AUTOREGRESSION DI STASIUN JUANDA SURABAYA

Abstract

The need of weather forecasting is primary to support activities in various sectors, so the efforts of development for forecast methods to improve the precision and the accuracy of the weather information are very important. Various weather forecasting models by engineering or stochastic model approach have been developed, although each method has both weaknesses and strengths, the efforts for developing techniques or methods to get the best model have to be done. What is elaborated in this article represent the result of testing in three statistical methods to obtain the best weather forecasting models. Three methods as mentioned before are: the Autoregressive Integrated Moving Average (ARIMA), Neural Network (NN), and Adaptive Splines Threshold Autoregression (ASTAR) to forecast the temperature, humidity, and daily rainfall. The performance of these three methods are evaluated by correlation values and Root Mean Square Error (RMSE). The good performance characterized by a high correlation between actual and forecast values, and also has a small RMSE. The results of this research indicate that ASTAR method produces better signed by a higher correlation, lower RMSE values and the constant forecasting from the first day until the thirtieth. The correlation in ASTAR method for Tmax and RHmin respectively are 0,70 and 0,75, for ARIMA method are 0,31 and 0,47, for NN method are 0,02 and -0,06. The three methods have poor performance for Tmin, RHmax and RRR. Keywords: Weather forecast, ARIMA, ASTAR, Neural Network