Analysis of agricultural drought in Iiuni, Eastern Kenya: application of a Markov model

In semi-arid Kenya, episodes of agricultural droughts of varying severity and duration occur. The occurrence of these agricultural droughts is associated with seasonal rainfall variability and can be reflected by seasonal soil moisture deficits that significantly affect crop performance and yield. The objective of this study was to stochastically simulate the behaviour of dry and wet spells and rainfall amounts in Iiuni watershed, Kenya. The stochastic behaviour of the longest dry and wet spells (runs) and largest rainfall amounts were simulated using a Markov (order 1) model. There were eight raingauge stations within the watershed. The entire analysis was carried out using probability parameters, i.e. mean, variance, simple and conditional probabilities of dry and rain days. An analysis of variance test (ANOVA) was used to establish significant differences in rainfall characteristics between the eight stations. An analysis of the number of rain days and rainfall amount per rain day was done on a monthly basis to establish the distribution and reliability of seasonal rainfall. The graphic comparison of simulated cumulative distribution functions (Cdfs) of the longest spells and largest rainfall amounts showed Markovian dependence or persistence. The longest dry spells could extend to 24 days in the long rainy season and 12 in the short rainy season. At 50% (median) probability level, the largest rainfall amounts were 91 mm for the long rainy season and 136 mm for the short rainy season. The short rains were more reliable for crop production than the long rains. The Markov model performed well and gave adequate simulations of the spells and rainfall amounts under semi-arid conditions.