Statistic and Probabilistic Variations with Precipitation Predictions of Jordan

This thesis deals with the analysis of monthly rainfall data gathered from 66 meteorological stations for the hydrologic years from October 1960 to September 2017 that are distributed irregularly in the Kingdom of Jordan. The area with the help of Thessien polygon is divided into twelve main districts, and the representative number of stations were detected. In order to study and analyze the rainfall (precipitation) data, the used sample size is 57. Four data quality tests Normality, Consistency, Homogeneity, and Stationary/Unit root and also the trend analysis test were studied based on parametric and non-parametric tests. The proper probability distribution function was found for each region, among the main four distributions the Normal, the log-Normal, the Gumble, and the Gamma. In order to predict the further yearly precipitation data for each meteorological district, three diverse widely used time series models were used; the Moving average (MA), the Holt-Winters, and nine different combinations of ARIMA. For this reason, the data of hydrologic years 1960-61 up to 2015-16 were used as a training period and the hydrologic year 2015-16 measured precipitation data was used for testing the trained data set where the best representative model was designated among the models which has the closest forecasted 2016-17 value to the measured (actual) rainfall value. Also the best model for each region was selected to predict the 10 consecutive upcoming hydrologic years from 2017-18 to 2026-27. Even for each region the rainfall within the months (from October to May) were studies statistically and their wetness or dryness periods and their overall trends were determined. Keywords: Rainfall, Time series models, Trends, Wet or dry spells.