GLOBAL LAND-COVER CLASSIFICATIONS AT 8 KM SPATIAL-RESOLUTION - THE USE OF TRAINING DATA DERIVED FROM LANDSAT IMAGERY IN DECISION TREE CLASSIFIERS

This paper reports a study which aims to (i) develop methodologies for global land cover classifications that are objective, reproducible an d feasible to implement as new satellite data become available in the future and (ii) provide a global land cover classification product bas ed on the National Aeronautics and Space Administration/National Ocean ic and Atmospheric Administration Pathfinder Land (PAL) data that can be used in global change research. The spatial resolution for the land cover classification is 8 km, intermediate between our previously pub lished coarse one degree by one degree spatial resolution and the 1 km global land cover product being developed under the auspices of the I nternational Geosphere Biosphere Program. We first derive a global net work of training sites from Landsat imagery, using 156 Landsat scenes mostly from the Multispectral Scanner System, to identify over 9000 pi xels in the PAL data where we have high confidence that the labelled c over type occurs. We then use the training data to test a number of me trics that describe the temporal dynamics of vegetation over an annual cycle for potential use as input variables to a global land cover cla ssification. The tested metrics are based on: (i) the ratio between su rface temperature and Normalized Difference Vegetation Index (NDVI); ( ii) seasonal metrics derived from the NDVI temporal profile, such as l ength of growing season; (iii) a rule-based approach that determines c over type through a series of hierarchical trees based on surface temp erature and NDVI values; and (iv) annual mean, maximum, minimum and am plitude values for all optical and thermal channels in the Advanced Ve ry High Resolution Radiometer (AVHRR) (PAL) data. Highest mean class a ccuracies from a decision tree classifier were obtained using the annu al mean, maximum, minimum, and amplitude values For all AVHRR bands. F inally, we apply these metrics to 1984:PAL data at 8 km resolution to derive a global land cower classification product using a decision tre e classifier. The classification has an overall accuracy between 81.4 and 90.3%. The Landsat images used for deriving the training data and the methodology for classification of AVHRR data at 8 km resolution ca n also be applied to 1 km AVHRR data and, in the future, Moderate Reso lution Imaging Spectroradiometer (MODIS) data at 250 and 500 m resolut ion. Digital versions of the land cover dataset and detailed documenta tion can be found on the World Wide Web at http://www.geog.umd.edu/lan dcover/8km-map.html.