Mapping vegetation in a heterogeneous mountain rangeland using landsat data: an alternative method to define and classify land-cover units

摘要: Abstract Three major problems are faced when mapping natural vegetation with mid-resolution satellite images using conventional supervised classification techniques: defining the adequate hierarchical level for mapping; discrete land cover units discernible by satellite; and selecting representative training sites. In order to solve these problems, we developed an approach based on the: (1) definition of ecologically meaningful as mosaics or repetitive combinations structural types, (2) utilization spectral information (indirectly) define units, (3) exploration two alternative methods classify once they defined: traditional, Maximum Likelihood method, which was enhanced analyzing objective ways best sites, method Discriminant Functions directly obtained from statistical analysis signatures. The study carried out in a heterogeneous mountain rangeland central Argentina Landsat data 251 field sampling On basis our combining terrain (a matrix stands×14 attributes) stands×8 bands), defined 8 (mosaics types) mapping, emphasizing types had stronger effects reflectance. comparison through validation both showed that produced better results than traditional (accuracy 86% vs. 78%).