Desempenho de um algoritmo de otimização hierárquico multiobjetivo aplicado a um modelo de superfície terrestre e ecossistemas

• Autor(a) principal: Camargos, Carla Cristina de Souza
• Data de Publicação: 2013
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: LOCUS Repositório Institucional da UFV
• Texto Completo: http://locus.ufv.br/handle/123456789/5261
• topic: Hidroclimatologia; Modelagem; Ecossistema; Algoritmo; Superfície terrestre; Hydroclimatology; Modeling; Ecosystem; Algorithm; Land Surface; CNPQ::CIENCIAS EXATAS E DA TERRA::GEOCIENCIAS::METEOROLOGIA::CLIMATOLOGIA

The performance of LSEMs (Land surface and ecosystem models) depends on the parameters of the equations representing the simulated process. However, the measurement of some parameters can be impractical or even impossible; therefore, they need to be estimated, or preferably optimized specifically for each ecosystem. When the parameters are calibrated to a single variable (mono-objective problem) they may not represent the reality, because the complexity of the model and its dependence on several variables (multi-objective problem). Thus, simultaneous multi-objective optimizations are indispensable. However, the optimization performance decreases as the number of variables to be optimized simultaneously increases. Furthermore, the study of simultaneous optimization using more than three objectives is a new area and not yet sufficiently studied. For simultaneous optimization of a large number of variables, there is a method that uses concepts of hierarchical systems theory in which the optimization occurs from the fastest (radiative fluxes) to the slowest process (carbon allocation). This study evaluates the performance of the hierarchical optimization using the index D (the average of the ratios between the individual outputs of multi-objective optimization and monoobjective). Understanding how the performance index D varies with respect to the number of objective functions optimized and to the number of hierarchical levels is important for the development of this research area. Two steps are necessary to achieve the study goals. First, a sensitivity analysis was performed to determine the output variables sensitivity to the model parameters. After, simulations were made using all possible combinations among the seven micrometeorological variables available (PARo, fAPAR, Rn, u *, H, LE, NEE) taking into account the hierarchy of processes. The results indicate that for up to three objective functions, hierarchical multi-objective optimization generates better results than the simultaneous multiobjective optimization (one hierarchical level), provided that the parameters distribution among hierarchical levels is consistent with the sensitivity analysis. Another important result shows that for the same number of outputs optimized, the greater the number of hierarchical levels the better the performance of the optimized model. However, the model performance falls quickly as the number of objective functions increases, evidencing that the power of hierarchy calibration that use a high number of objective functions is highly dependent on the removal of some constraints for model s performance.