Estimativa dos componentes do balanço de radiação a partir de sensoriamento remoto, observação de superfície e redes neurais artificiais
Arquivos
Data
2016-07-27
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Federal de Viçosa
Resumo
Restingas são ecossistemas costeiros localizados no domínio da Mata Atlântica que hospedam uma grande diversidade de plantas, incluindo muitas espécies endêmicas. Atualmente, encontram-se altamente ameaçados devido à expansão da infraestrutura de transporte e desenvolvimento imobiliário. Para tomar medidas mitigatórias é necessário desenvolver estudos que permitam compreender o funcionamento deste ecossistema. Como parte deste esforço, o presente trabalho teve como objetivo estudar os componentes do balanço de radiação na restinga de Marambaia –RJ, obtidos a partir de medições em campo, sensoriamento remoto e redes neurais artificias. Os dados de campo foram coletados e avaliados no período de março de 2015 a fevereiro de 2016. Os resultados mostraram que a rede neural para calcular a radiação de onda longa atmosférica, foi mais eficiente que os modelos clássicos (r 2 > 0,83, RMSE < 6 W m -2 , MAE < 5 W m -2 e d > 0,94). No cálculo do balanço de radiação, a rede mostrou melhor desempenho para dias de céu claro (r 2 = 0,90, RMSE = 81,67 W m -2 , MAE = 64,96 W m -2 , d = 0,96) do que para dias de céu nublado (r 2 = 0,74, RMSE = 74,30 W m -2 , MAE = 40,51 W m -2 , e d = 0,83). Os resultados de sensoriamento remoto, mostraram uma subestimação no cálculo dos componentes do balanço de radiação. No entanto, no cálculo do ciclo diurno do balanço de radiação, este apresentou desempenho similar da rede neural artificial para dias de céu claro, r 2 = 0,93, RMSE = 85,81 W m -2 , MAE = 74,50 W m -2 , e d = 0,96. De modo geral, consideram-se ambas as metodologias como alternativas interessantes no esforço de calcular os componentes do balanço de radiação, a partir de um mínimo de variáveis meteorológicas ou para áreas sem nenhuma instrumentação.
Restingas are coastal ecosystems located in the area of Atlantic Forest that host a wide variety of plants, including many endemic species. Currently, they are highly threatened due to expansion of the transport infrastructure and real estate development. To take mitigation measures is necessary to develop studies to understand the functioning of this ecosystem. As part of this effort, this study aimed to study the radiation balance components in the restinga of Marambaia -RJ obtained from field measurements, remote sensing and artificial neural networks. Field data were collect and evaluated from March 2015 to February 2016. The results showed that the neural network to calculate the atmospheric longwave radiation was more efficient than the classical models (r 2 > 0.83, RMSE < 6 Wm- 2 , MAE < 5 Wm-2, and d > 0.94). In calculating the net radiation, the network showed better performance for a clear day (r 2 = 0.90, RMSE = 81.67 Wm- 2 , MAE = 64.96 Wm- 2 , d = 0.96) than for days of cloudy sky (r 2 = 0.74, RMSE = 74.30 Wm- 2 , MAE = 40.51 Wm- 2 , d = 0.83). The results of remote sensing showed an underestimation in the calculation of the components of the radiation balance. However, in calculating the diurnal cycle of net radiation, it showed similar performance of the artificial neural network for a clear day, r2 = 0.93, RMSE = 85.81 Wm- 2 , MAE = 74.50 Wm- 2 , d = 0.96. In general, both methods were consider as interesting alternatives in an effort to calculate the radiation balance components from a minimum of meteorological variables or areas without any instrumentation.
Restingas are coastal ecosystems located in the area of Atlantic Forest that host a wide variety of plants, including many endemic species. Currently, they are highly threatened due to expansion of the transport infrastructure and real estate development. To take mitigation measures is necessary to develop studies to understand the functioning of this ecosystem. As part of this effort, this study aimed to study the radiation balance components in the restinga of Marambaia -RJ obtained from field measurements, remote sensing and artificial neural networks. Field data were collect and evaluated from March 2015 to February 2016. The results showed that the neural network to calculate the atmospheric longwave radiation was more efficient than the classical models (r 2 > 0.83, RMSE < 6 Wm- 2 , MAE < 5 Wm-2, and d > 0.94). In calculating the net radiation, the network showed better performance for a clear day (r 2 = 0.90, RMSE = 81.67 Wm- 2 , MAE = 64.96 Wm- 2 , d = 0.96) than for days of cloudy sky (r 2 = 0.74, RMSE = 74.30 Wm- 2 , MAE = 40.51 Wm- 2 , d = 0.83). The results of remote sensing showed an underestimation in the calculation of the components of the radiation balance. However, in calculating the diurnal cycle of net radiation, it showed similar performance of the artificial neural network for a clear day, r2 = 0.93, RMSE = 85.81 Wm- 2 , MAE = 74.50 Wm- 2 , d = 0.96. In general, both methods were consider as interesting alternatives in an effort to calculate the radiation balance components from a minimum of meteorological variables or areas without any instrumentation.
Descrição
Palavras-chave
Radiação solar, Restingas, Sensoriamento remoto, Redes neurais artificiais
Citação
ZULUAGA ARISTIZÁBAL, Cristian Felipe. Estimativa dos componentes do balanço de radiação a partir de sensoriamento remoto, observação de superfície e redes neurais artificiais. 2016. 43 f. Dissertação (Mestrado em Meteorologia Aplicada) - Universidade Federal de Viçosa, Viçosa. 2016.