Evaluación de imágenes del sensor MODIS para la cartografía de la cobertura del suelo en una región altamente diversa de México

Abstract

In recent decades, the use of arable land for agriculture has expanded to occupy nearly 40 % of the world's land surface, thereby greatly impacting the biodiversity of our planet. In order to understand and manage these changes, it is indispensable to have updates on land use/land cover generated with tools that allow us to obtain information over larger areas with greater frequency. Images derived from moderate resolution sensors such as MODIS represent an alternative to high resolution imagery, though we lack a precise understanding of the accuracy of the land characterization provided by this sensor at regional levels. The aim of this work is to contribute to the knowledge about the most ideal type of remote sensing data needed to generate land cover/land use information and the methodologies that can produce a more detailed legend while still retaining an acceptable level of accuracy. The study area is the region of Tancitaro, Michoacan, Mexico and is represented by temperate and dry tropical forests, pasture lands and croplands. Three kinds of MODIS data were tested: vegetation indices, spectral reflectance eight day composites, and daily spectral reflectance images. These data were analyzed through two different approaches; maximum likelihood and neural networks. We also applied ancillary data to compare the results of the classifications with and without the ancillary data. The results obtained show it is possible to achieve acceptable levels of accuracy using moderate resolution imagery if a simple classification scheme is used.