Tratamento e preenchimento de falhas de s?ries de dados meteorol?gicos utilizando workflows cient?ficos paralelos em ambientes de GPU

• Autor(a) principal: Silva, F?bio Cardozo da
• Data de Publicação: 2014
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Biblioteca Digital de Teses e Dissertações da UFRRJ
• Texto Completo: https://tede.ufrrj.br/jspui/handle/jspui/3124
• topic: Meteorologia; Climatologia; e-Ci?ncia; Proveni?ncia; Workflows Cient?ficos; Hidrologia; Computa??o paralela; Meteorology; Climatology; e-Science; Provenance; Scientific Workflows; Hidrolog; Paralel Computing; Matem?tica

Despite of the growing importance of researches in the field of Meteorology, especially those that handle large volumes of data focused on studies of hidrological resources, difficulties the handle datasets are increasing. Researchers are developing great efforts to obtain and store high quality data in their repositories. This dissertation aims to present a computational proposal capable to compute meteorological data and add quality control to long times series of data. The artifacts conceived and developed in this work are based on the e-Science vision. We used high performance processing features and scientific workflows to aid to automate the process of scientific research in Meteorology. Furthermore, this work integrates VisTrails scientific workflows with parallel computing environments using GPU and CUDA programming. The integration was guided to extend the capability of handling large volumes of hihg quality meteorological data. Other features of this work are the discussions about performance gains of the proposal and the representation of (raw and curated) data and retrospective provenance generated by the computational experiments according to PROV-DM specification. The main results of this work are. 87,7% of detection of errors and failures replacemente were achieved using 77 meteorological stations. We can conclude that the fusion of E-Sceince vision with CUDA parallel computing approach is viable to deal with large volumes of meterological and climatological data.