MODIS land cover product validation in the Iberian Peninsula

Abstract

The goal of this study is the post-launch validation of MODIS IGBP Land Cover Product for the Mediterranean landscape. Southern Portugal, mostly a Mediterranean ecosystem due to climate and geomorphology, is characterized by large homogeneous patches. For this reason, it was chosen as the most promising region where global products could be used. The most recent 2002 MODIS Land Cover data set was used. The selection of the test sites was conducted through a method supported by Landsat5-TM images, to derive homogeneous patches. Moreover, field test sites accommodated the guidelines from the System for Terrestrial Ecosystem Parameterization (STEP) document. Confusion matrix results in an overall accuracy of about 36% and a kappa statistic of 0.09. Croplands, which constitute the majority of the study area, is the most well classified feature, with an accuracy of 77%, although the MODIS classifies it largely by excess. The evergreen broadleaf forest is the most confused theme, with several land cover classes - evergreen needleaf forest, mixed forest, woody savannas, savannas and closed shrublands - classified as such. However, there isn't any classified evergreen broadleaf forest pixel. Therefore, it seems that there is a consistent misclassification of broadleaf evergreen forest. We strongly believe that the reason for the misclassification of the Mediterranean landscape relies in its weak representation within the training sites framework. MODIS land cover maps, produced by processing the 32-day database using decision tree and artificial neural network classification algorithms, are greatly dependent on the core sites data. Therefore, the classification of Mediterranean likely landscapes could never occur properly. Adding Mediterranean ecosystem like training sites could greatly improve the results. This study shows how careful the adoption of global products to regional studies should be. The validation of the MODIS IGBP Land Cover Product for Mediterranean landscape revealed that more regional work should be taken into account during the production of global products in order to reflect this biome.