Abstract An improved drought management must rely on an accurate monitoring and forecasting
of the phenomenon in order to activate appropriate mitigation measures. In this study, several
homogenous Hidden Markov Models (HMMs) were developed to forecast droughts using the
Standardized Precipitation Index, SPI, at short-medium term. Validation of the developed models
was carried out with reference to precipitation series observed in 22 stations located in the upper
Blue Nile river basin. The performance of the HMM was measured using various forecast skill
criteria. Results indicate that Hidden Markov Model provides a fairly good agreement between
observed and forecasted values in terms of the SPI time series on various lead time. Results seem
to confirm the reliability of the proposed models to discriminate between events and non-events
relatively well, thus suggesting the suitability of the proposed procedure as a tool for drought
management and drought early warning.