Time-Series Classification of High-Temporal Resolution AVHRR NDVI Imagery of Mexico

Abstract

Time-series data from wide-field sensors, acquired for the period of a growing season or longer, capitalize on phenological changes in vegetation and make it possible to identify vegetated land cover types in greater detail. Our objective was to examine the utility of time-series data to rapidly update maps of vegetation condition and land cover change in Mexico as an input to biodiversity modeling. We downloaded AVHRR NDVI 10-day composites from the USGS EROS Data Center for 1992-1993 and adjusted for cloud contamination by further aggregating the data. In the first phase of our analysis, we selected training sites for various land cover types using a land cover map created by the Mexican National Institute of Statistics, Geography, and Informatics (INEGI) as a guide. Since there is a high degree of spectral variability within many of the vegetated land cover types, we subjected the spectral response patterns to cluster analysis. We then used the statistics of the clusters as training data in a supervised classification. We also compared unsupervised and univariate decision tree approaches, but these provided unsatisfactory results. Best results were achieved with a 19-class map of land use/land cover employing a supervised approach.