Neuro estimador para o fluxo de gases entre a superfície terrestre e a atmosfera / A neuro estimator for the flux of gases between the surface land and the atmosphere

AUTOR(ES)
FONTE

IBICT - Instituto Brasileiro de Informação em Ciência e Tecnologia

DATA DE PUBLICAÇÃO

21/11/2011

RESUMO

Studies show that global warming may be irreversible in the medium or long term, and is a major cause of air pollution. However, this framework can be reversed or mitigated by estimating the amount of emissions, since the mapping of pollution sources could facilitate the work of public authorities in relation to environmental monitoring and the development of public policies related to the environment. Thus, this work develops a model, based on the methodology of inverse problems, whose first model is the direct source-receptor Lagrangian technique (LAMBDA code) and the second is the SCIATRAN code, which simulates the radiance measurements of SCIAMACHY sensor, aboard the satellite ENVISAT . It then uses the coupling of neural networks to estimate the concentration of carbon dioxide between the soil and the atmosphere in various scenarios. Finding a suitable network for this type of problem, the best parameters and the best activation function, is essential to confirm the effectiveness of the technique used in this work to obtain the estimation of sources and sinks. The neural network Multilayer Perceptron with the backpropagation algorithm was used to solve the problem concerning the estimation of the rate of pollution sources. Thus, this paper presents several experiments made with meteorological data obtained from the Copenhagen Experiment and model BRAMS. The meteorological data such as: temperature, wind speed and direction, are used in the LAMBDA code simulation for dispersion of pollutants in the region of interest and the chosen date. The meteorological data of the Copenhagen Experiment were used in experiments that simulate the sources (or sinks) of CO$_{2}$, with concentrations of data ``in situ and from remote sensing. For the experiments with remote sensing data, which deal with the regions of forest and pasture in Rondônia, the input parameters required by code LAMBDA were obtained from the model BRAMS. The methodology developed in this thesis, to estimate the sources and/or sinks of pollutants, makes the coupling of two artificial neural networks, in which the first has as input the radiances generated with the model SCIATRAN and generates as output the gas concentration profiles, which are used for the second neural network to estimate the sources (or sinks) of carbon dioxide on the surface. The experiments showed satisfactory results, showing that the artificial neural network can be considered an appropriate technique for estimating pollutants.

ASSUNTO(S)

redes neurais artificiais estimação de fontes e sumidouros de gases problema inverso artificial neural networks estimation of sources and sinks of gases inverse problem

ACESSO AO ARTIGO

Documentos Relacionados