الفهرس 
Chapter 1 IntroductionOnly 14 pages are availabe for public view
Abstract
This thesis aims to develop a new study on the area of data assimilation via chemical transport model and aerosol remote sensing images. More specifically, we focus on enhancement of the model prediction of Aerosol Optical Depth (AOD) using satellite data, and examination of the nonlinear combination of aerosol species properties. On October, 2004, Ozone Monitoring Instruments (OMI) began their operational flight to retrieve global quantitative aerosol properties over Earth. In this work we propose a study which specifically evaluates AOD retrieved by OMI at various areas including positions and the AErosol RObotic NETwork (AERONET). The comparison result shows that, just over Europe, OMI AOD is better retrieved in the multi-wavelength retrieval than in the near UV. Initial validations show moderate correlation values due to the cloud contaminated effect and mineral dust high load during spring and summer. Also, overestimations of OMI values are detected with a positive bias. Correlations with enhancements have been done by using an easy standard to keep away from sights with very high values of AOD resulting from high aerosol loading. Bias adjustment method has been applied based on a detailed analysis of aerosol characteristic on AERONET stations. The linear bias-adjustment method is carried out to make AOD observation better (i.e. minimized errors and maximized correlation coefficients) for the entire year at almost all locations. In order to overcome the effect of the boundary layer of the absorbing aerosols, a post correction procedure was applied, depending on a large-scale climatologically study of absorbing aerosol. The results generally show bias decreasing by 15%, smaller variance, and a high correlation coefficient. A proposed method to improve OMI aerosol retrieval using combination of OMI, MODIS, and CALIPSO observations is illustrated in this study. The results are validated and indicated a great agreement with ground observations, which show a large potential of using multi satellite analysis to improve aerosol retrievals.
Also, this thesis presents a novel artificial intelligent approach that use Artificial Neural Network (ANN) to merge and enhance the retrieved satellite data. Neural network technique is used for merging multi-sensor satellite data using two techniques: stacked and Bayesian NN. Stacked NNs are used to learn the temporal and spatial drifts between data from different satellite sensors. Bayesian NN has the ability to predict and estimate the best value of AOD from different instruments. Graphics Processing Units (GPUs) can guarantee high performance computing, low-cost platform, and programmable framework for simulation of ANNs. Both NNs are trained and validated by using ground based measurements. Initial validations show a high correlation values 0.74 and 0.79 for both SNN and Bayesian NNs respectively, which is better than state-of-the-art results. The steps, strategies, processes, and optimization approaches of the GPU computations for both networks are outlined. The performance and acceleration in terms of speedup of the proposed techniques are studied on selected graphics cards. Maximum speedup 2.01 is achieved on SNN, while Bayesian NN GPU implementation speedup is 1.45. Although, the gained speedup is limited, it could enhance training time and implementation of large scale NN. The resulting data from SNNs and enhanced satellite measurement are then assimilated with a regional chemical transport model to produce three dimensional distributions of aerosols throughout Europe for the year 2011. The results show that the assimilation of AOD observations significantly improves the forecast for total mass. The errors on aerosol chemical composition are reduced and are sometimes vanished by the assimilation procedure and NN preprocessing, which shows a big contribution to the assimilation process.
A regional chemical transport model assimilated with daily mean satellite and ground based AOD observations is used to produce three dimensional distributions of aerosols throughout Europe for the year 2011. The aerosol forecasts involve two-phase process assimilation, and then a feedback correction process. During the assimilation phase, the total column AOD is estimated from the model aerosol fields. The model state is then adjusted to improve the agreement between the simulated AOD and OMI retrievals of AOD using positive matrix factorization. The results show that, the assimilation of AOD observations significantly improves the forecast for total mass. The errors on aerosol chemical composition are reduced and are sometimes vanished by the assimilation procedure, which shows a big contribution to the assimilation process.
Throughout this study, the AOD measurements of the OMI are assimilated with Polyphemus model. The performance of the model simulation is simulated on three multicore processors; NIVIDIA GPUs, the multicore shared memory Intel Quad-Core Xeon processor, and the heterogeneous CELL Broadband Engine, Multiple issues related to model analysis are studied: time partitioning to reduce truncation slip, and dimension partitioning to enhance chemical transport models’ multi-processing. Additionally, a technique to access matrix row and column randomly is discussed. The technique could be applied to a matrix that consists of different sizes from the Cell Broadband Engine and GPUs accelerator unit. Our proposed enhancements reach an average of 7.44 speedup using GPUs, speedup of 5.89 is achieved using the Cell Broadband Engine, and a maximum speedup of 4.53 on a Quad core processing shared memory with OpenMP.