Computational Learning in Climate Change Adaptation Support

Abstract

Natural disasters such as floods and droughts resulting from climate variability and weather extremes cause catastrophes like food insecurity which is most severe in the drier parts of Uganda like Karamoja region. This study examined the effectiveness of computational learning as applied
in the forecast of stimulus to increase in agricultural crop prices and subsequent prediction of food insecurity. Secondary rainfall data from 1986 to 2008 were obtained from Meteorology Department for two weather stations and six-years data on consumer price index of maize were obtained from
Uganda Bureau of Statistics. Auto Regressive Integrated Moving Average time series analysis algorithm was used to forecast the liner pattern of rainfall and crop price while artificial neural networks (ANN), a computational model inspired by functional aspect of biological neural network,
was used to forecast the non-liner pattern. Cross validation was performed using training and validation data to evaluate learning capability of the algorithms. Prediction accuracy of the two algorithms were compared and the hybrid model produced better results than the single model. ANN produced
high sensitivity which demonstrated the effectiveness of applying computational learning in the prediction of catastrophes such as food insecurity. The prediction results can be used by decision makers for informed decisions on climate change adaptation where the local community still has
low adaptation capability. It is recommended that the local community should participate in planning interventions to address disasters such as famine to enhance their understanding of the disaster and increase communication between them and disaster managers.