Any given geographic area is often subjected to numerous mapping efforts over the course of time. Similar end products may be generated from the same data source with similar target attributes. For instance, two maps representing the land cover of Canada were produced in 1995 and 1997 with data from the advanced very high resolution radiometer (AVHRR) satellite: the Northern Biosphere Observation and Modeling Experiment (NBIOME) product produced by Natural Resources Canada, and the International Geosphere–Biosphere Programme Data and Information System (IGBP DISCover) product. The thematic and spatial agreement of the forested classes of the map area representing Canada are considered in this study. A difference image was generated for each of two scenarios, where one product identified forest and the other identified non-forest and vice versa. Standard area summaries and per-pixel analyses were used to initially identify and quantify the differences between the two map products. To enable a more comprehensive comparison of the two map products, a 50 km × 50 km grid extending over the entire area of Canada was used as a framework for analyzing the spatial autocorrelation in the difference images. Differences that are not spatially autocorrelated are considered random; conversely, differences that are spatially autocorrelated may be systematic and reflect differences in classification legends and methodologies, and in image-processing methods. The total estimates of forest area from both maps are similar, varying by 6%, yet the area of agreement between the two maps (i.e., where both mapping processes have the same result in the same location) represents 62% of the total area classified as forest in both maps, or 35% of Canada. The spatial distribution of these classification differences is captured through the introduction of ancillary data (ecozones) and the consideration of spatial autocorrelation. Predominantly, spatially autocorrelated differences are found to occur within ecozones that are transition areas between forest and non-forest and at ecozonal interfaces. These differences appear related to the heterogeneous nature of the land cover and the small size of contiguous forest stands. In this research we demonstrate a range of approaches to map comparison. These approaches enable end users of map products to make informed decisions regarding various large area land cover products and to understand the implications of using these different products as inputs for subsequent applications or models.

Estimates of stand height are an integral component of forest inventories. Lidar has been demonstrated as a tool for remotely sensing information on the vertical structure of forests, such as height. The ability to remotely sense height information for forest inventory purposes may allow for procedures such as update, audit, calibration, and validation. With current technologies, lidar data collection and processing are a resource-intensive undertaking. The ability to use a regression model to spatially extend a lidar survey from a sample to a larger area would act to decrease costs while allowing for the characterization of a larger area. In this study we address the ability to extend lidar estimates of height from sample flight lines to a greater area using segmented Landsat-5 thematic mapper (TM) data. Based upon empirical relationships between lidar-estimated height and within segment digital numbers, height is estimated for an entire landscape from a 0.48% sample. To conform to current polygon-based forest management practices, the within polygon segment-based height estimates are combined to create an updated height attribute for each polygon. The empirical relationships between lidar data and forest inventory polygon attributes (coefficient of determination (r(2)) = 0.23; standard error (SE) = 4.15) and within polygon spectral values (r(2) = 0.26; SE = 4.06) indicated a need to develop more representative models. To this end, we developed a regression model to produce a relationship between quantile-based estimates of mean canopy top height for the segments with lidar hits (r(2) = 0.61; SE = 3.15). This segment/height empirical relationship allowed us to extend the height estimates to polygons that have no lidar information using the image digital numbers. The segment/lidar estimates of height generally form a range centered on zero (no difference) to +/-6 m of the ground measured height for over 80% of the available validation plots, with a r(2) of 0.67 and a SE of 3.30 m.

The availability of high resolution (1 m or better) imagery from space opens up the possibility of automatic detection of coniferous trees. Our test site is located within the Greater Victoria Watershed (GVWD) on Vancouver Island, British Columbia, Canada. In previous research we have examined various filters for detecting trees over an area with mature and immature Douglas fir trees. We have obtained a 1 m spatial resolution digital orthophoto generated from aerial photography, MEIS 1 m multispectral imagery, and IKONOS panchromatic 1 m imagery over our test site. Within the test site, there are ground plots in which the location of each tree has been determined. These detailed plots are used to assess the accuracy of the methods used for tree detection for each of the high resolution image types. The characteristics of each tree are documented allowing for an assessment of the conditions leading to the identification, or lack of identification, of each tree. The comparison of three differing data sources, each with 1 m spatial resolution, indicates favorable results for the IKONOS satellite data. The highest proportion of the trees from the field stem plot data were identified with the IKONOS satellite panchromatic imagery. While the IKONOS results have a higher rate of false positives than the airborne multispectral data, a preference for the satellite data is due to characteristics such as ease of collection, large image extent, repeatability, and radiometric consistency over a larger area

RÉSUMÉ Dans cette communication, nous effectuons la fusion du système WRS de Landsat (Landsat Worldwide Reference System) avec les ensembles de données spatiales nationales canadiennes représentatives de l'utilisation du sol, de l'altitude et des caractéristiques de population. La fusion du système de la grille WRS avec d'autres données spatiales permet aux utilisateurs de faire des requêtes sur les contenus de chacune des scènes WRS et de résumer les résultats au plan national. Par exemple, en enlevant certaines scènes pour réduire le chevauchement, il y a 712 scènes qui se chevauchent et qui couvrent la masse continentale du Canada, dont 434 présentent un couvert forestier supérieur à 10%. Il y a actuellement 1200 scènes du Canada où le système de grille WRS n'est pas réduit pour minimiser le chevauchement. Les résultats des requêtes peuvent être considérés spatialement ou aspatialement. L'information générée par l'ajout de l'information spatiale à la grille WRS permet d'améliorer la sélection des images pour divers types d'échantillonnage pour obtenir une information nationale ou régionale. Dans cette communication, nous présentons un bref sommaire des données utilisées et du mode de création de ces ensembles de données de même que les résultats de certaines requêtes.

The feasibility of generating via Landsat TM data current estimates of cover type proportions for areas lacking this information in the national forest inventory was explored by a case study in New Brunswick. A recent forest management inventory covering 4196 km(2) in south-eastern New Brunswick (the test area) and a coregistered Landsat TM scene was used to develop predictive models of 12 cover type proportions in an adjacent 4525 km(2) region (the validation area). Four prediction models were considered, one using a maximum likelihood classifier (MLC), and three using the proportions of 30 TM clusters as predictors. The MLC was superior for non-vegetated cover types while a neural net or a prorating of cluster proportions was chosen for predicting vegetated cover types. Most predictions generated for national inventory photo-plots of 2 x 2 km were closer to the most recent inventory results than estimates extrapolated from the test area. Agreement between predictions and current inventory results varied considerably among cover types with model-based predictions outperforming, on average, the simple spatial extensions by about 14 %. In this region, an 11-year-old forest inventory for the validation area provided estimates that in half the cases were closer to current inventory estimates than predictions using the optimal Landsat TM model. A strong temporal correlation of photo-plot-level cover type proportions made old-values more consistent than predictions using the optimal Landsat TM model in all but three cases. Prorating of cluster proportions holds promise for large-scale multi-sensor predictions of forest inventory cover types.