Arima and vector autoregressive model evaluation in forecasting rainfall: a case of Kisumu

Abstract

Time Series Analysis has been used over the decades in data analysis and forecast ing. Auto Regressive Integrated Moving Average (ARIMA) models have been fit on economic data and engineering data. The models have also been used in analysis of climate data. Previous studies have focussed on temperature data from National Mete orological Stations where summarized monthly values were used. In this study, we used daily rainfall data from Kenya Meteorological Services Station in Kisumu. The objec tives included univariate time series modelling using ARIMA on long term rainfall data for daily, monthly, seasonal and annual data and forecasting rainfall for the different time periods. The other objective was to compare forecast from univariate ARIMA to Vector Autoregression (VAR) when rainfall, minimum and maximum temperature values are included in model. ARIMA models were fit on the KMS rainfall data, and VAR models were fit on temperature, minimum and maximum rainfall data from KMS. Finally, farm ers’ local rainfall data was compared to that of KMS for independence. Results showed that forecasts under VAR did not give a more precise forecast of future rainfall than ARIMA. Further, that there was not enough statistically significant evidence to suggest that rainfall data from KMS and farmers’ locale were independent.