Time series analysis of monthly rainfall in Tanzania.

Time series analysis of monthly rainfall series of Tanzania was carried out in order to develop the parameters of stochastic models. Error free monthly rainfall records from 109 rainfall stations distributed all over Tanzania are used. The data used in the study covers a 40 year period, 1951-1990. Use is made of the 17 homogeneous rainfall regions established by the method of Principal Component Analysis (PCA), Odiyo (1994). The most representative station for each region was identified and used for the subsequent analysis. The models were to be used for the generation and study of general monthly rainfall characteristics. However, the hypothesis that Tanzania monthly rainfall series could be modelled stochastically failed. The autocorrelation analysis results showed that the historical series was correlated while the standardized series was not. Therefore the deterministic type of models could not be developed. The conclusion from this study is that the monthly rainfall data can well be described by probabilistic models. It was established using the method of L-moments that the random standardized series could be fitted (depending on the region considered) to the the three parameter log logistic (LL), Pearson type three (P3), the General Extreme Value (GEV) and the Generalized Pareto (GPA) probability distribution functions. Thus probabilistic models following these distributions for each region were established and used to simulate the seasonal rainfall series in Tanzania. The simulated rainfall was analysed to check for any statistical resemblance with the historical time series. The analysis of the generated data verified that the models could reproduce, statistically, the historical statistics such as the means, variances and skewness. The results as obtained were satisfactory. It was also shown that these distributions fitted the standardized series well as the length of the record generated was progressively increased indicating the advantages of using longer data series for more reliable fittings than the use of several particular distributions on short historical data. It is concluded that the developed models can be used to simulate the monthly rainfall series of any length for use in regional water resources assessment and management in Tanzania.