MARKOV REGIME-SWITCHING AUTOREGRESSIVE MODEL FOR MODELING RAINFALL IN THE UPPER EAST REGION OF GHANA

Abstract

The economy of Ghana largely depends on agriculture for its growth. In Ghana, agricultural productivity largely relies on rainfall. This study seeks to explore the rainfall pattern using data from the Ghana Meteorological Agency from January 1999 to December 2019 for the Navrongo and Bolgatanga weather substations. A univariate two-state Markov switching autoregressive model was used to characterize the distinct seasonal regime behavior of the rainfall in the study area. The data sets were found to be non-stationary. Hence, were differenced to attain stationarity. MS (2)-AR (1) was obtained as the best model for the Navrongo data as it had the least of goodness-of-fit measures. Also for the Bolgatanga data sets, MS (2)-AR (1) was obtained as the best model. The transitional probabilities for the Navrongo weather substation are P11= 0.80, P12 =0.20 , P21 =0.35, and P22 =0.65 with a 5.1 months expected duration of rainfall and at least 3.05 months of dry season every year. The transitional probabilities for the Bolgatanga weather substation are P11 =0.82, P12 =0.18 , P21 =0.27, and P22 =0.73. The expected duration for rainfall in the study area is 5.5 months and that for low or no rainfall is estimated to be at least 3.7 months for every year. Based on the variations in the expected duration for the various regimes, it is recommended that government assists farmers with early maturing seedlings and drought resistant seedlings. Due to uneven distribution of the rainfall pattern observed, the study recommends farmers practice good water management in their farms as it is essential to crop survival and the maximization of crop yield potential.