INTERNATIONAL JOURNAL OF CLIMATOLOGY Int. J. Climatol. (2016) Published online in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/joc.4632 Projecting changes in Tanzania rainfall for the 21st century F. Ciofi, a,b * F. Conticello a and U. Lall b a Department of Civil, Construction and Environmental Engineering, University of Rome “La Sapienza”, Italy b Department of Earth and Environmental Engineering, Columbia University of New York, NY, USA ABSTRACT: A non-homogeneous hidden Markov model (NHMM) is developed using a 40-year record (1950–1990) of daily rainfall at 11 stations in Tanzania and National Centers for Environmental Prediction-National Center for Atmospheric Research (NCEP-NCAR) re-analysis atmospheric ields of a number of meteorological variables. The following atmospheric ields, temperature at 1000 hPa, geo-potential height at 1000 hPa, meridional winds and zonal winds at 850 hPa, and zonal winds at the equator from 10 to 1000 hPa, in a region deined by 25 ∘ S–25 ∘ N and 25 ∘ –75 ∘ E are identiied as appropriate predictors for the downscaling of the seasonal regime of daily rainfall in Tanzania. The NHMM is used to predict future rainfall patterns under a comparatively high greenhouse gas emissions scenario [Representative Concentration Pathway 8.5 (RCP8.5)], using predictors from the CMCC-CMS (Centro Mediterraneo sui Cambiamenti Climatici) simulations from 1950 to 2100. Instead of pre-specifying a ixed rainy season, the model considers seasonality of precipitation to be determined by the 21st century simulations of the atmospheric variables used as predictors. The future downscaled precipitation simulations for the RCP8.5 scenario indicate that in the 21st century Tanzania may experience: (1) a slight decrease in the number of wet days and seasonal rainfall in MAM and JJAS, but not in OND; (2) a reduction of annual total rainfall; and (3) an intensiication of the frequency and intensity of extreme rainfall, as identiied by 90th, 95th, and 99th percentiles. KEY WORDS NHMM; Tanzania; rainfall; climate change; downscaling model Received 28 May 2015; Revised 23 November 2015; Accepted 1 December 2015 1. Introduction In recent years, Tanzania has experienced a number of extreme hydrological events such as looding and droughts that triggered crop failures, livestock deaths, and inten- siication of diseases (Shemsanga et al., 2010). This has led to a discussion about climate change, whose effects could be widespread and with potential outcomes for agriculture, forests, water resources, coastal resources, human health, as well as energy, industry, and transport (Shemsanga et al., 2010). An evaluation of how signiicant these outcomes are – e.g. by using detailed hydraulic models (Ciofi and Gallerano, 2006, 2012) – requires a local projection of precipitation. Unfortunately, the future trends of precip- itation in Tanzania are quite uncertain (Agrawala et al., 2003; Tumbo et al., 2010). Because of the coarse spatial resolution of general circulation models (GCMs), the statistics of precipitation at the local scale can be strongly biased in retrospective simulations. Consequently, dynamical and statistical downscaling schemes are employed. Hewitson and Crane (1996), Murphy (1999), Zorita and Von Storch (1999), and Giorgi and Mearns (2002) discuss the relative merits and * Correspondence to: F. Ciofi, Department of Civil, Construc- tion and Environmental Engineering, University of Rome “La Sapienza”, Rome, Italy. E-mail: francesco.ciofi@uniroma1.it; federicorosario.conticello@uniroma1.it shortcomings of different procedures. These inter- comparisons vary widely with respect to predictors, pre- dictands, and measures of skill (Von Storch et al., 2000). Statistical downscaling methods are preferred where a large number of simulated sequences of the climate vari- able are needed for application with a hydrological model. GCM simulations of large-scale upper-air ields are gen- erally better constrained than those for precipitation, and an appropriate selection of these variables can provide an effective set of predictors for statistical downscaling. From the different statistical approaches proposed in lit- erature, we chose to use the hidden Markov model (HMM) and non-homogeneous hidden Markov model (NHMM). These have found widespread application in meteorology and hydrology, for statistical downscaling of daily pre- cipitation from observed and numerical climate model simulations (Zucchini and Guttorp, 1991; Hughes and Guttorp, 1994; Charles et al., 1999; Hughes et al., 1999; Bellone et al., 2000; Robertson et al., 2004; Betrò et al., 2008). Recently, NHMMs have been successfully used to downscale precipitation in different regions of the world (Robertson et al., 2004; Kwon et al., 2006, 2009; Khalil et al., 2010). Here, we explore several potential predictors derived from the re-analysis (NCEP-NCAR) considering a phys- ical basis. We show that temperature at 1000 hPa (T1000), geo-potential height at 1000 hPa (GPH1000), meridional winds (MW850) and zonal winds at 850 hPa (ZW850), © 2016 Royal Meteorological Society