Difficulties with estimating city-wide urban forest cover change from national, remotely-sensed tree canopy maps

Two datasets of percent urban tree canopy cover were compared. The first dataset was based on a 1991 AVHRR forest density map. The second was the US Geological Survey’s National Land Cover Database (NLCD) 2001 sub-pixel tree canopy. A comparison of these two tree canopy layers was conducted in 36 census designated places of western New York State. Reference data generated by photo-interpreting the tree cover on 1994 and 2002 orthoimages were compared on a city-wide basis to estimates of the tree cover derived from the AVHRR and NLCD tree canopy maps. Comparison of the AVHRR-based estimate with the 1994 photo-interpreted reference showed low accuracy and high variability. The comparison between the NLCD-derived estimate and 2002 photo-interpreted values had higher accuracy but shows a consistent under prediction bias for all tree cover values. In comparing the two photo-interpreted reference sets, little actual change was detected for this study region. Assessing change using differences between the imagery-based AVHRR and NLCD urban tree cover datasets is shown to lead to erroneous results.