64th ISI World Statistics Congress - Ottawa, Canada

64th ISI World Statistics Congress - Ottawa, Canada

Predicting precipitation using transfer function models of the spatiotemporal variabilities in the arid and sub-humid regions of Southern Africa

Author

LC
Mr Lyson Chaka

Co-author

  • M
    Mohamed A. M. Abd Elbasit

Conference

64th ISI World Statistics Congress - Ottawa, Canada

Format: CPS Abstract

Keywords: dynamic-linear-regression, forecasting, regression, spatio-temporal, time-series, time_series

Abstract

The impact of global warming on coastal regions of Southern Africa has contributed to a series of unusual rainfall patterns and floods in the past decade. The region is experiencing severe damage and loss to infrastructure and productive landscape. Currently, the factors associated with these adverse rainfall patterns, and the relationships among these factors are not well known. The spatial and temporal variability in monthly rainfall in the arid, semi-arid and sub-humid areas in South Africa are analysed to generate a mechanism for the prediction of precipitation in the regions. Transfer function models are appropriate tools to model precipitation patterns that are assumed to be influenced by other spatiotemporal variations in climatic conditions. These models provide a basis for explaining the relationships between precipitation patterns and other climatic factors for a specified period. Precipitation predictions are useful for planning of agricultural activities, making decisions in farming projects and disaster management purposes in the region. Decision-making in agriculture and other sectors of economy rely mostly on weather patterns and forecasts. We propose a time-series modelling approach that uses dynamic linear regression models to forecast monthly average precipitation in South Africa regions based on shortwave radiation, sea surface temperature anomalies and air temperature. We demonstrate the concept using the 1983-2021 data collected from multi-satellites located in the Indian and Atlantic Oceans bordering the target regions.