الفهرس 
Basic conceptsOnly 14 pages are availabe for public view
 |  | from | 145 |  |  |
| | | from | 145 | | |
Abstract
Numerical models along with some data mining techniques can represent a major goal of atmospheric modelling to provide a better understanding of processes determining the dynamical and chemical state of the atmosphere.
The forecasting of air pollution, should provide a good simulation for the vertical fluxes diffusion of pollutants in the whole column of the air within the boundary layer.
In this work an experiment was accomplished to study the using of the numerical model RegCM4.7 to predict the PM10 mass concentration levels originated by two cases of sand storm. As the pollutant dispersion occurred within the planetary boundary layer PBL, it was focused on three PBL schemes for running the model (Holtslag, UW and GFS). The results showed that the model gave a good representation for the case study. Concerning the PM10 concentration, the results of Holslag scheme came to be more accurate than UW and GFS schemes. The consistency of the model was also examined and the results of the relation among the wind speed, PBLH and PM10 concentration proved during the sand storm. Hence we can conclude that strong wind speed and thick PBL produce a good dispersion of pollutants which consequently led to decrease the concentration of the PM10, and give healthier conditions and produce higher ventilation index. The results of the numerical model was interpreted by using data mining techniques to provide the conditions and the suitable criteria of the occurrence of different dust phenomena according to numerical model’s output data.
The thesis consists of five chapters,
In chapter one we introduced all the necessary definitions, mathematical background and tools on which the study is based on.
In chapter two we presented the definition of data mining, and its main concepts. We studied some of data mining techniques like Decision tree, K- nearest neighbor and naïve biased. We also presented some techniques to examine the model accuracy such as: holdout method, random sampling method, K-cross validation method and confusion matrix. We also presented a definition of Rapidminer software which we used to accomplish our experiments in chapter (5).
In chapter three we studied of some numerical methods used in solving differential equations and stability of the solution, classification of partial differential equations, the definition of the two concepts existence and uniqueness for the solution. Finite difference method in both time and space is presented. We also studied the discretisation of the advection equation using a centred difference both in time (leap frog) and space. The basic concepts of the regional climate models, aim of such models, its historical background, and its components were also presented.
In chapter four we studied the three planetary boundary layer schemes mathematical back ground within regional climate model. We applied the three schemes on two real cases of a sand storm. We predicted the concentration of particulate matter PM10, boundary layer height, and wind speed. We also compared the results of the three schemes to find the best accurate one among them.
In chapter five we applied some of the data mining techniques to build a model to predict different weather phenomena and we tested the accuracy of the model, then we used that model to analyze and interpret the output data we get from experiments in chapter (4).