| dc.description.abstract | Abstract: Many small-scale farmers require adequate forecasts to help them plan for the rainfall. The National
Meteorological Service provides forecasts seasonally, monthly and weekly. The forecasts are qualitative in nature hence
inform, but cannot be directly used with decision support models. It is therefore important to consider forecast methods that
researchers can use to generate quantitative data that can be applied in the models. In particular, an increasing need for
forecasting daily rainfall data. In this study, the ARIMA and VAR models have been used to forecast five time period data for
daily, monthly and seasonal rainfall data. The objective was to find the model parameters that best fit the three time periods.
Fifty-year data from Kenya Meteorological Station, Kisumu, was used for the analysis. For each time period, five events were
used as the test dataset. The ARIMA model was found to be best for forecasting daily rainfall in comparison to the VAR model,
while SARIMA was best for monthly and seasonal data. One difference was done for the seasonal rainfall total, but not for
monthly and monthly rainfall data. The VAR models included the available daily minimum and maximum temperatures.
However, forecasted daily rainfall deviated from the test data, while monthly and seasonal data deviated even more. | en_US |
