MODIS and Landsat TM image fusion using the SIFuLAP method for mapping the Brazilian savannas (Cerrado)

Abstract

The Cerrado is the second largest biome in Brazil, covering an area of about 2 million km2. However, recent estimates revealed that more than one third of the Cerrado was already converted into other land cover types. This paper presents a multi-temporal data fusion experiment where one Landsat TM image (30 m) was fused with a series of MODIS images (250-500 m). Moreover, a simplified data fusion method based on the Laplacian pyramid concept (SIFuLAP) is illustrated. The objective is to assess the ability of the fused images to reflect the seasonal patterns of the Cerrado. In order to do that, NDVI temporal profiles were derived from the fused images and subsequently compared with the ones that were obtained from the original MODIS images. In addition, the temporal fused products were used in an attempt to improve classification accuracy of the Cerrado. The results showed that the NDVI temporal profiles, derived from the fused images, were efficient to assess the seasonal behaviour of the Cerrado. Moreover, classification accuracy was improved using the temporal fused images as additional information into the classifier.