Combinação de modelos de previsão climática

• Autor(a) principal: Moura, Isabelle
• Data de Publicação: 2018
• Tipo de documento: Trabalho de conclusão de curso
• Idioma: por
• Título da fonte: Repositório Institucional da UFRN
• Texto Completo: https://repositorio.ufrn.br/handle/123456789/34306
• topic: Aprendizado de Máquina; Machine Learning; Floresta Aleatória; Random Forest; Máquina de vetores de suporte; Support Vector Machine

Precipitation is one of the most important variables to describe the future climate, since it describes any type of phenomenon related to the fall of water from the sky. There are several models of climate simulation to forecast precipitation today. The models adopted in this work will be BCC-CSM1.1, CCSM4, CESM1, CPC-NOAA, NorESM1-ME and MRI-CGCM3 in three distinct regions Amazonia, Rio de la Plata Basin and Northeast Brazil. Three different types of scenarios are considered: RCP 4.5, RCP 6.0, and RCP 8.5. These scenarios have different characteristics in relation to the variables of emission of gases, concentration of gases of greenhouse effect, and information of type of terrestrial coverage. Previous studies show that the combination of models makes forecasting more accurate. Due to this, two methods of combining models, Random Forest llamods and Support Vector Machine (SVM) were adopted. Following the mean square error was used for the comparison between the two methods, after this comparison it was observed that the two methods can make an approximate prediction of the real values, with Random Forest having the lowest mean square error in all the different types of scenarios and considered regions.