Forecasting seasonal rainfall for agricultural decision-making in northern Nigeria

Least absolute deviation (LAD) multiple regression equation is proposed as a model for forecasting rainfall in the savanna agro-ecological region of northern Nigeria. The model is employed in forecasting seasonal rainfall, in order to assess the climatological success range for improving the production of three staple cereals grown in the region. Historic climate data spanning 35 years are used to generate a computer-based statistical model, which utilizes November, December and January (NDJ) values of identified climate controlling variables in the region, and the previous year's rainfall as predictors. Nineteen synoptic stations in the study area are selected, and because of high inter-annual variability of rainfall in the region, grouped into three sub-regions using cluster analysis. Seasonal rainfall was forecasted in a probabilistic fashion for respective sub-regions and a measure of the anticipated variability about the forecasted value established using confidence bounds. The model was tested in recommending rainfed potentiality for growing corn, sorghum and millet, during 1991-1995, by deriving probabilistic estimates of receiving critical rainfall threshold for growing each cereal. Result indicated with high probability, 0.98 and 0.91, that millet and sorghum, respectively, can be grown without irrigation under rainfed conditions, while corn which required more seasonal rainfall, had the lowest probability at 0.60. (C) 2001 Elsevier Science B.V. All rights reserved.