Mineração de dados para previsão de eventos extremos de precipitação

• Autor(a) principal: Charles, Anderson Cordeiro
• Data de Publicação: 2015
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Biblioteca Digital de Teses e Dissertações da UERJ
• Texto Completo: http://www.bdtd.uerj.br/handle/1/13811
• topic: Data mining; Artificial neural networks; Decision trees; Weather. instability rates; Mineração de dados (Computação); Redes neurais artificiais; Árvores de decisão; Meteorologia; Índices de instabilidade; Precipitação (Meteorologia) Métodos de simulação; Controle de inundações; Chuvas Modelos matemáticos; Inundações Previsão; CNPQ::CIENCIAS EXATAS E DA TERRA::GEOCIENCIAS::METEOROLOGIA::METEOROLOGIA APLICADA

Data mining classification models were applied in this work in order to predict extreme climatic events one hour in advance. Data (2008 2012) extracted from the surface meteorological station located at the Polytechnic Institute of Rio de Janeiro State University in Nova Friburgo (RJ) were used. The knowledge discovery in databases (KDD) process was applied, comprising the data preparation, its mining and its post processing. Artificial neural networks and decision tree algorithms were used to identify patterns that indicates accumulation of precipitation above 10 mm in the following hour. It was observed that the use of weather observational data in micro scale was susceptible to high false positive rates in the short term prediction. In order to solve this problem, historical data of predictions performed by the ETA Model with resolution of 15 km were inserted into the model construction process. This data has been made available from the Weather Prediction and Climate Studies Center of the National Institute for Space Research - CPTEC/INPE. The work included the calculation of the instability index related to the formation of a severe convective situation at the region of Nova Friburgo and its store at a database, joining the micro and meso scale records. The obtained results demonstrated that the union between databases was extremely important to reduce the high false positive rates. This was a very important contribution to the meteorological studies performed in surface meteorological stations. The model with the highest accuracy was used to develop a real time alert system which verifies the possibility of torrential rains occurrences (10 mm or more) one hour ahead.