RAINFALL PREDICTION USING SEASONAL AUTOREGRESSIVE INTEGRATED MOVING AVERAGE AND GEOGRAPHIC INFORMATION SYSTEM APPROACH

Abstract

Rainfall is one indicator to determine the estimated adequacy of groundwater on agricultural land. The groundwater availability produced by rain can determine cropping patterns in an area. The availability of rainfall data depends on the accuracy of information on current climate conditions. This case causes the related parties to find difficulty determining the classification of cropping patterns in the future. Accurate rainfall prediction models are needed to overcome the problem of shifting rain patterns. Rainfall prediction models in determining cropping patterns are recommended by FAO, such as linear regression, which is still widely used today. This study aims to develop a new model of rainfall prediction by using the method SARIMA to determine cropping patterns to increase crop yields. Rainfall data was used from 2010 to 2020 from seven rainfall collection stations in Sleman Regency, and they are used as training data to predict future rainfall. The output of the data analysis is a prediction of rainfall in the range of January-April, which is predicted to be high, May-August, which is predicted to be low; and September-December, which is predicted to be moderate. In addition, based on the identified cropping patterns, recommendations can be given to farmers to set cropping schedules and strategies to increase the productivity of the farmland. The testing of accuracy forecasting used relative mean absolute error (RMAE) for 12 months. The results of the forecasting accuracy test for 12 months in Sleman Regency showed RMAE average of 1.46 was considered low, for it was still below 10%.