Classificação e monitoramento fenológico foliar da cobertura vegetal na região da floresta Nacional do Tapajós - Pará, utilizando dados multitemporais do sensor

Jane Delane Verona

The relative importance of accuracy in forest cover mapping is given by the necessity to obtain improvement in the elaboration of a management plan of natural resources and in the definition of priority areas for conservation, as well as in landscape analysis. Other aspect that has raised even more interest in the scientific community is concerned to modeling of biogeochemical cycles and global changes. Studies related to forest phenology have offered promising results to help the researches of ecological subject, by directing important questions with respect to global modeling, monitoring and climate changes. The main contribution of this work was to identify and quantify the fluctuation of spectral response throughout of seven distinct months, as a consequence of forest phenology related to climate variations, and, from this, to direct the choice of images more adequate for discriminating physiognomies in tropical forest areas. The study area is located in the north region of the Tapajós National Forest, Pará State. Multitemporal Landsat-5 TM images, corresponding to the months selected in the period from May 1997 to August 1999 were utilized. Initially, these images were pre-processed involving procedures of geometric rectification and image registration, as well as radiometric rectification. Besides 3, 4, 5, and 7 bands of each TM image, it was also generated synthetics bands such as NDVI, 5/4 ratio, and shade, soil and vegetation fraction images. Following, some tasks were developed to guarantee reliable samples of vegetation classes to perform the statistical tests: preliminary analysis with 1999 TM image (forest and non forest mask); change detection between two dates, 1986 and 1999 (to guarantee the presence of forest cover in the latest date)and cloud masks for all dates. Seven forest samples were selected, with forest in the high and low plateau, ""babaçu"", regeneration areas with 21 years and scarp among them. With the objective to verify the presence of phenology at terrestrial level, field campaigns were performed, where floristic and structural information were collected, as well as measurements of leaf area index variation, with LAI-2000, in three different epochs, in some primary and secondary forest transects. The results were not satisfactory. However, at orbital level, the correlation between precipitation and the bands of each TM image for the seven vegetation classes were elaborated, confirming the presence of seasonal variation, considering that a correlation of 0.94 between precipitation and vegetation fraction image was achieved. The interesting thing was that the NDVI presented a very low correlation, maybe due to the fact that NDVI values saturate rapidly in the forest environment. Based on these results, the work was pursued by selecting the best dates and processing to classify the vegetation classes. So, two statistical approaches were performed: the anomaly test and the stepwise discriminant analysis. Both tests selected the same months, September, October, December, and May, as the ideal for classifying the highest number of vegetation samples, and the selected bands by both statistical approaches were shade fraction image, band 3, vegetation fraction image, and band 5. In addition, the anomaly test identified the band 7, while the NDVI was selected in the stepwise discriminant analysis. The unitemporal classification approach discriminated a maximum of 56.61 percent of the vegetation samples, while the multitemporal approach achieved values greater than 90 pecent of classification, utilizing 3 or more dates and several bands. The adopted methodology achieved successfully the objectives of this work and will be useful for future multitemporal classification of forest cover in the tropical environment.

AUTOR(ES)

DATA DE PUBLICAÇÃO

RESUMO

ASSUNTO(S)

ACESSO AO ARTIGO

Documentos Relacionados