Mapping chlorophyll in the Amazon floodplain lakes with MERIS
Barbosa, Claudio¹; Novo, Evlyn¹; Carvalho, Lino^2
1National Institute for Space Research – INPE, Brazil, BRAZIL; ²INPE, BRAZIL

The Amazon drainage basin is one of the most important river systems in the world covering approximately 6.5 x 10⁶km², of which around 17% are wetlands. The main stem floodplain in central Amazonia has approximately 300 x 10³km² and contain almost 8000 lakes larger than 100 ha. The primary productivity of these lakes, whose dynamic is affected by both land use and the flood pulse, has been the subject of several studies. Many of these floodable areas are remote, inaccessible and formed by large lakes, making remote sensing a feasible tool for their study and monitoring. The photosynthetic pigment chlorophyll-a chl-a is a key indicator of primary productivity and it causes changes in water color that can be registered by remote sensors and transformed into useful information. In this context, the aim of this research was to estimate chlorophyll-a concentration in lakes of complex and turbid waters of a large Amazon basin floodplain using MERIS/Envisat images and empirical models. In situ measurements (spectroradiometric and limnological) taken before, simultaneously and after MERIS/Envisat images acquisition were used to fit two and three band empirical models. Three different input data were used to build the models: a) In situ spectra, tuned to best wavelengths to fit the models, which resulted R² of 0.91 for two band model and R² of 0.95 for three band model; b) MERIS bands simulated from in situ spectra, which resulted R² of 0.87 for two band model and R² 0.94 of for three band model; c) MERIS image bands, which resulted R² of 0.77 for two band model and R² 0.75 of for three band model. The results indicate that spectral bands and spatial resolution of MERIS are suitable for mapping chlorophyll in lakes along the Amazon basin. The spatial resolution of 300 meters is not a constrain for the large lakes at the Amazon floodplain. A visual comparison between simulated MERIS spectra with those extracted from image at the same location suggests that atmospheric correction and footprint differences between in situ measurements and the MERIS sensor are responsible for the decrease in the accuracy of models based on MERIS images. A time lag of two days before and after between the image acquisition and some in situ measurements used to build the model also contributed for the accuracy decrease.