Abstract

Combination forecasting methods will try to minimize the resulting error to provide more accurate predictions and reduce uncertainty about the future when it is not known whether the historical data pattern will repeat or differ from the past. This study aims to test a combination forecasting method of mixed linear regression (LMM) method and support vector regression (SVR) method. The combination forecasting method is applied to obtain a model of rice phenology. The dependent variable is the polarization index RPI (Ratio polarization index), NDPI (Normalised Polarization Index), and API (average polarisation Index) extracted from sentinel 1 satellite images. The response variable is the age of rice obtained from field surveys in the rice field area of PT Sang Hyang Seri Subang West Java in planting season 1 (rainy) and planting season 2 (dry). It is assumed that the variation that occurs in both seasons is different so the growing season is set as a random effect in the model. The combination forecasting method trialled in this study was the Variance-Covariance (VACO) method. The results showed that individually, the LMM method was able to produce R2=78.4% accuracy while the SVR method was able to produce R2=84.57%. The combination of the two methods using the VACO method showed an accuracy of R2 = 84.57%. The accuracy of the model is almost exactly the same as the accuracy of the SVR method. This is because the residual variance of the SVR method is smaller than the residual variance of the LMM method so that the weight of the SVR method is greater than the weight of the LMM method.