An investigation of national tree canopy assessments applied to urban forestry

Two nation-wide estimates of tree canopy cover, one based on 1991 AVHRR imagery and the second the USGS's NLCD 2001 tree canopy layer, will be used for urban forest assessments. Since only 3.1% of the coterminous U.S. land area is urban in 2000, the applicability of these national datasets to assess urban forests was investigated. This dissertation, consisting of three manuscripts, was an investigation of potential errors in national urban tree canopy assessments and how well these national products are suited for analysis of urban tree canopy cover.; The first manuscript presents several methods used to assess the accuracy of sub-pixel classifications including difference measures, linear regression statistics and an index of agreement. Since individual pixel errors can vary widely in sub-pixel tree canopy cover estimates, aggregating several pixels together into a minimum assessment unit allows for proper use of the resultant classification and explicitly sets spatial parameters on its end use. The second manuscript assesses the accuracy of the NLCD 2001 tree canopy in urban areas of western New York State revealing an over prediction bias of tree cover for low canopied cities. Mapped tree cover was compared with reference data derived from photo-interpretation over citywide areas. When combined with photo-interpretation, non-point-specific, or wide-area, methods could be a more efficient means to generate reference data for sub-pixel classifications. The last paper compares the two national tree cover assessments to photo-interpreted reference values. Urban forest change analyses should not be done using these datasets.; The results reveal that the AVHRR-based estimates of urban tree canopy cover are highly variable and should not be used if better estimates are available. The NLCD tree cover maps represent a substantial leap forward from the AVHRR-based imagery but could suffer from systematic biases like over-prediction. If the systematic bias is consistent across the U.S., national estimates of urban forest benefits like carbon sequestration or compensatory value from these data could be in error.