Homogeneous & non-homogeneous hidden markov downscaling model for projection of hydroclimate changes in Tanzania

Climate change poses a number of problems to water resources. The precipitation change, expressed in amount of rainfall and occurrence of wet/dry days, could cause seasonal and annual shifts in the climatology of the area and bring up an extreme events intensification. Therefore, the aim of the proposed work is to develop a Statistical Downscaling Methodology on the entire Tanzania territory, where water supply by superficial and underground sources play a key role. So, Any change in hydrologic cycle could constitute an hazard and adversely affect the sustainability and also the future economic development of the area. A Hidden Markov Model (HMM) is here used to describe annual daily rainfall occurrence at 11 gauge stations in Tanzania, in east Africa, yearly along the period 1950-1990. The model assumes that rainfall occurrence is governed by a few discrete states, with Markovian daily transitions between them. Five ‘‘hidden’’ rainfall states are identified. These states are able to capture the typical Tanzanian seasonality conditions; and further the occurrence and amount of precipitations proper of the different stations. Moreover, a Non-homogeneous Hidden Markov Model (NHMM) is then applied to downscale daily precipitation occurrence at the 11 stations, using daily large scale predictors extracted from the NCAR-NCEP reanalysis General Circulation Model (GCM) dataset as: Geo Potential Height (GPH), Temperature (T), Zonal and Meridional Winds (ZW & MW) and Vertical Equator Winds from 10 to 1000 hPa (ZEW). The calibration (1950-1980) and validation (1981-1990) tests, for different predictor combinations, evidence that a considerably betterment in fitting the historical data is obtained by using the NHMM, which results able to simulate the seasonal rainfall pattern, characteristic of Tanzania. Then, the NHMM provides a useful diagnostic and predictive tool (a) in linking the statistics of daily rainfall occurrence and amount at the station level to the large-scale atmospheric patterns and (b) it can be used with the goal to make future projections of the downscaled precipitation by using the GCM’s simulations (CMIP5) under different global warming scenarios.