GIS and Remote Sensing Analyses of Forest Plantations in Costa Rica’s Atlantic Lowland Area

With the recent implementation of the Kyoto Protocol (Kyoto 1997), carbon credits may afford a new economic market associated with forested landscapes (IPCC 2000). To assess the potential for carbon sequestration in an important part of those forested landscapes, tree plantations, it is not only important to determine the extent to which tree species differ in carbon storage, but also to know the areal extent of these plantations. Here, two different types of images were used to develop a method for classifying and geo-referencing tree plantations in the Costa Rican lowlands. The first, an ALI multispectral image, yielded some success with 59% of the expected plantations found to be correctly identified. The second type, Hydice images with much smaller areal coverage, was found to be almost useless due to sensor noise. For the multispectral ALI images, applying a Minimum Noise Fraction, MNF, filter to decorrelate noise and determine the dimensionality of the image was followed by a supervised Maximum Likelihood classification to distinguish plantations from primary and secondary forests in the surrounding areas. For the hyperspectral Hydice images, the MNF filter was again used followed instead by a Spectral Angle Mapper classification. The Spectral Angle Mapper was chosen due to this algorithm’s ability to classify even with large variances in lighting (Lumme 2004). This method was found to be the most successful method of distinguishing plantations of differing species from the surrounding vegetation. No definitive information on the areal extent of plantations in the Costa Rican lowlands was found, but a map of plantations that were ground-truthed was successfully created for future researchers to use more recent and higher resolution image sets, such as the CARTA II mission, to attempt similar or new methods.