Land cover mapping, fire regeneration, and scaling studies in the Canadian boreal forest with 1 km AVHRR and Landsat TM data

Abstract

A multitemporal 1 km advanced very high resolution radiometer (AVHRR) land cover analysis approach was used as the basis for regional land cover mapping, fire disturbance‐regeneration, and multiresolution land cover scaling studies in the boreal forest ecosystem of central Canada. The land cover classification was developed by using regional field observations from ground and low‐level aircraft transits to analyze spectral‐temporal clusters that were derived from an unsupervised cluster analysis of monthly normalized difference vegetation index (NDVI) image composites (April‐September 1992). Quantitative areal proportions of the major boreal forest components were determined for a 821 km × 619 km region, ranging from the southern grasslands‐boreal forest ecotone to the northern boreal transitional forest. The boreal wetlands (mostly lowland black spruce, tamarack, mosses, fens, and bogs) occupied approximately 33% of the region, while lakes accounted for another 13%. Upland mixed coniferous‐deciduous forests represented 23% of the ecosystem. A SW‐NE productivity gradient across the region is manifested by three levels of tree stand density for both the boreal wetland conifer and the mixed forest classes, which are generally aligned with isopleths of regional growing degree days. Approximately 30% of the region was directly affected by fire disturbance within the preceding 30–35 years, especially in the Canadian Shield Zone where large fire‐regeneration patterns contribute to the heterogeneous boreal landscape. Intercomparisons with land cover classifications derived from 30‐m Landsat Thematic Mapper (TM) data provided important insights into the relative accuracy of the 1 km AVHRR land cover classification. Primarily due to the multitemporal NDVI image compositing process, the 1 km AVHRR land cover classes have an effective spatial resolution in the 3–4 km range; therefore fens, bogs, small water bodies, and small patches of dry jack pine cannot be resolved within the wet conifer mosaic. Major differences in the 1‐km AVHRR and 30‐m Landsat TM‐derived land cover classes are most likely due to differences in the spatial resolution of the data sets. In general, the 1 km AVHRR land cover classes are vegetation mosaics consisting of mixed combinations of the Landsat classes. Detailed mapping of the global boreal forest with this approach will benefit from algorithms for cloud screening and to atmospherically correct reflectance data for both aerosol and water vapor effects. We believe that this 1 km AVHRR land cover analysis provides new and useful information for regional water, energy, carbon, and trace gases studies in BOREAS, especially given the significant spatial variability in land cover type and associated biophysical land cover parameters (e.g., albedo, leaf area index, FPAR, and surface roughness). Multiresolution land cover comparisons (30 m, l km, and 100 km grid cells) also illustrated how heterogeneous landscape patterns are represented in land cover maps with differing spatial scales and provided insights on the requirements and challenges for parameterizing landscape heterogeneity as part of land surface process research.