Description: The ABMIw2wLCV2010 was derived by applying a semantic and spatial generalization algorithm (Castilla et al., in preparation), and later updating to 2010 conditions, to a combination of two raster datasets: the Canadian Forest Service (CFS) Earth Observation for Sustainable Development (EOSD) LC dataset, and the Land Cover for Agricultural Regions of Canada, circa 2000 (LCARC, a.k.a. NLWIS) dataset of Agriculture and Agri-Food Canada (AAFC). Both datasets were created using digital classification of Landsat 5 and Landsat 7 ortho-images acquired around year 2000, with each Landsat scene classified individually and then mosaicked into the final raster product. NB. Both rasters use the same hierachical classification scheme. This feature class was obtained by merging 48 individual tiles whose frames are provided in the ABMIw2wLCV2000_48tiles feature class. Each tile is completely within a single Landsat scene, and its boundary roughly coincides with the seam lines used to stitch together the EOSD. Each map tile was created as follows:(0) The EOSD and NLWIS rasters were combined into a single raster according to a ruleset that selects a label for each pixel based on the values for that pixel in the two rasters. See the description of the SOURCE* fields in the ABMIw2wLCV2000_48tiles for further details. Lakes, rivers, pipelines, powerlines, railways and roads from the Government of Alberta (GoA) GIS layers were also 'burned' in this combined raster.(1) A morphological segmentation algorithm is applied to the gradient magnitude (a sort of 'edge' image) of the original Landsat image used to derive the EOSD, in order to break it down into a mosaic of tiny homogeneous regions (mean segment size around 10 pixels) separated by 1-pixel thick boundaries. NB. The edges of the features copied from the GoA's layers area are enhanced in the gradient magnitude image so as to preserve their shape in the output segmentation.(2)Regions having a clearly predominant landcover class ( more than 75% of its pixels belong to that class) are assigned to it, and adjacent regions having the same class are merged together. This results in a big portion of the image (usually more than 90%) being already labeled.(3)Endmembers, or typical spectral signatures (i.e., mean value in each band of the Landsat image) for each class are derived from these regions. (4)Unlabeled regions are assigned to the class of the most similar neighbour, providing the latter is already classified and bears enough similarity to it, ortherwise they are assigned to the class with the highest relative abundace within the region, where the latter is abundance weighted by distance in the feature space between the spectral signature of the region and the endmember of each class. Once again, adjacent regions having the same class are merged together after the assigment. NB. There are special rules on how to deal with semantically related classes, such as conifer, broadleaf and mixed forest. (5) Regions are reclassified into the final set of 11 landcover classes using a predefined crosswalk table, after which adjacent regions having the same class are merged together. NB. Treed wetlands (class 81 in EOSD) greater than 100 ha are assumed to be black spruce and thus conifer. Smaller treed wetlands are assigned to the class of the closest forest endmembers.(6) Regions smaller than the minimum mapping unit (MMU: 0.5 ha for water and 2 ha for the rest) are merged to their most similar adjacent neighbour. (7) Isolated regions smaller than 25 ha that show a low contrast with the encompassing region were eliminated, as they were found in a previous version to mainly correspond to spurious polygons.(8) Using the center of boundary pixels as initial vertices, the result is converted to an ESRI polygon shapefile using a line simplification algorithm. Once all tiles were ready, the final product was created by merging and dissolving the individual tiles into a seamless layer. After dissolving, a few seam lines were still visible due to differing classifications in the adjoining tiles. The labels of polygons affected by these inconsistencies were harmonized using semi-automated GIS procedures, after which a final 'dissolve' was performed.The circa 2000 map was updated to circa 2010 conditions using a hybrid procedure that included a combination of change detection and GIS techniques. The map was updated for new features appearing in the 2000-2010 period of the following types: forestry cutblocks, wildfires, surface mines, roads and urban development. Cutblocks were extracted from the cumulative ABMI cutblock layer. A difference image (NDVI2000 – NDVI2010) was computed, and statistics for each cutblock polygon were derived. Cutblocks harvested between 2000 and 2010 were preselected by thresholding and those close to the threshold were visually inspected before the final selection. The wildfires were extracted from GoA’s Historical Wildfire Perimeter Dataset using a similar procedure than for the cutblocks. Mines, roads and urban were extracted from the 2010 ABMI Human Footprint dataset using GIS overlay operations that compared these with preexisting features of this type in the 2000 map. The different types of new features were integrated into a single disturbance layer, the outlines of this layer were generalized to give them a similar appearance than those of the 2000 map, the disturbance layer was stamped on top of the 2000 map, and slivers (small polygons resulting from boundary mistmatches) were removed. The Landcover_Polygons_2010 feature class has a new field called ‘MOD_TY’ (modifier type) that records the type of disturbance a new polygon not belonging to the developed class is. Possible values are ‘Cutblock’, ‘Burnt’, ‘Mining’ and ‘Tailing pond’ (the latter belongs to water class). NB. Cutblocks, mines and other disturbance features that already existed in 2000 have their MOD_TY field blank in this version.Landcover class definitions (field LC_class):20 - Water: lakes, lagoons, rivers, canals, and artificial water bodies. Shallow open water is included in this category, unless there is more than 20% vegetation cover, in which case it belongs to the relevant vegetated class.31 - Snow/Ice: areas permanently covered by snow or ice, including glaciers. 32 - Rock/Rubble: bedrock, rubble, talus, blockfield, lava beds, or other natural impervious surfaces. 33 - Exposed Land: bare soil (barren, non-agricultural), river sediments and cutbanks, pond or lake sediments, reservoir margins, beaches, landings, fresh (less than 1yr) cutblocks, recently burned areas, mudflat sediments, surface mining, or other non-vegetated (less than 10% trees, or less than 20% shrub/herb) surfaces. 34 - Developed: urban and built-up areas (including industrial sites), impervious artificial surfaces (e.g. airport runaways), railways and roads. Acreages and farmsteads are included in this class. Oil and gas well pads are included in this class if connected to a road and not abandoned or under reclamation. Urban terrain under development is included in this class, even if the land is exposed. Urban green areas are excluded of this class if larger than 2 ha and if they have less than 2 buildings per hectare. 50 - Shrubland: At least 20% ground cover which is at least one-third shrub (shrub: a woody plant not considered a tree), with no or little presence of trees (less than10% crown closure). Examples of plants belonging to this class in Alberta are alder, willow, juniper, and sagebrush. Shrubby fens and other non-treed woody wetlands, usually associated with floodplains and the shores of lakes and streams, belong to this class. Recent clearcuts and forest fires are initially labeled as shrubland to differntiate them from the surrounding forest.110 - Grassland: Predominantly native grasses and other non-woody vegetation (e.g., forbs) with a minimum of 20% ground cover. May include some shrub cover (but less than a third of the vegetated area) or a few trees (but the tree cover cannot exceed 10%). Land used for range with native unimproved grasses (a.k.a. rough pasture) is included in this class. Alpine meadows fall into this class. Marshes and other non-woody wetlands with at least 20% vegetation cover (sedges, cattails, or moss) belong to this class. NB. A forestry cutblock harvested more than year ago that contains seedlings, but where the latter cover less than 10% of the area of the cutblock, belongs to this class. If the cutblock had no successful regeneration and was covered by more than 20% shrubs, then it would belong to the 'shrubland' class.120 - Agriculture: annually cultivated crops, tame pastures (fields planted or sown with non-native grasses/legumes where livestock is directly grazing on them in the summer), forage crops (same as the previous, but instead of grazed, cut for hay) and woody perennial crops (fruit orchards and vineyards). Bare agricultural (i.e., tilled) soil belongs to this class and not to 'exposed land'.210 - Coniferous Forest: Treed areas with at least a 10% ground cover of trees (a.k.a. crown closure), where coniferous trees (spruce, pine, fir, larch) are 75% or more of the crown closure. Young plantations or regenerating cutblocks of conifer trees belong to this class no matter that the trees are less than 5 m tall, providing crown closure has reached 10%. Treed wetlands (e.g., black spruce bogs and fens) are included in this class providing they are conifer dominated and crown closure exceeds10%.220 - Broadleaf Forest: Treed areas with at least a 10% ground cover of trees, where broadleaf trees (trembling aspen, balsam poplar, white birch) are 75% or more of the crown closure. Young plantations or regenerating cutblocks of broadleaf trees belong to this class no matter that the trees are less than 5 m tall, providing crown closure has reached 10%. Treed swamps along river floodplains and other treed wetlands are included in this class providing they are broadleaf dominated and crown closure exceeds 10%.230 - Mixed Forest: Treed areas with at least a 10% ground cover of trees, where neither coniferous nor broadleaf trees account for 75% or more of crown closure.The overall thematic accuracy of the map, as estimated by an extensive validation dataset (see description of the relevant feature class) is 75% with 11 classes, and 88% if these classes are grouped into 5 general classes. User and Producer accuracies for specific classes in each level appear in tables appended to this geodatabase. Caveats:The width of roads has been systematically exaggerated to a minimum of 60 m (two Landsat pixels); therefore the real extent of the Developed class is greatly overestimated.Forest areas harvested or burned between 2000 and 2010 were assigned to the ‘Shrub’ LC class to differentiate them from their surroundings. They can be identified using the ‘MOD_TY’ field, which will respectively contain the value ‘Cutblock’ or ‘Burnt’.The accuracy of the shrub class is low (30%). Many shrub polygons are in reality forest (beyond new cutblocks and burnt areas), especially in the North. This was not possible to correct with the input data at hand.
Copyright Text: ABMI Remote Sensing Group 2013, based on the EOSD and NLWIS 2000 raster datasets and on hydrography and acces GIS layers from the Government of Alberta. Update to 2010 based on ABMI Human Footprint dataset.
Description: The ABMIw2wLCV2010 was derived by applying a semantic and spatial generalization algorithm (Castilla et al., in preparation), and later updating to 2010 conditions, to a combination of two raster datasets: the Canadian Forest Service (CFS) Earth Observation for Sustainable Development (EOSD) LC dataset, and the Land Cover for Agricultural Regions of Canada, circa 2000 (LCARC, a.k.a. NLWIS) dataset of Agriculture and Agri-Food Canada (AAFC). Both datasets were created using digital classification of Landsat 5 and Landsat 7 ortho-images acquired around year 2000, with each Landsat scene classified individually and then mosaicked into the final raster product. NB. Both rasters use the same hierachical classification scheme. This feature class was obtained by merging 48 individual tiles whose frames are provided in the ABMIw2wLCV2000_48tiles feature class. Each tile is completely within a single Landsat scene, and its boundary roughly coincides with the seam lines used to stitch together the EOSD. Each map tile was created as follows:(0) The EOSD and NLWIS rasters were combined into a single raster according to a ruleset that selects a label for each pixel based on the values for that pixel in the two rasters. See the description of the SOURCE* fields in the ABMIw2wLCV2000_48tiles for further details. Lakes, rivers, pipelines, powerlines, railways and roads from the Government of Alberta (GoA) GIS layers were also 'burned' in this combined raster.(1) A morphological segmentation algorithm is applied to the gradient magnitude (a sort of 'edge' image) of the original Landsat image used to derive the EOSD, in order to break it down into a mosaic of tiny homogeneous regions (mean segment size around 10 pixels) separated by 1-pixel thick boundaries. NB. The edges of the features copied from the GoA's layers area are enhanced in the gradient magnitude image so as to preserve their shape in the output segmentation.(2)Regions having a clearly predominant landcover class ( more than 75% of its pixels belong to that class) are assigned to it, and adjacent regions having the same class are merged together. This results in a big portion of the image (usually more than 90%) being already labeled.(3)Endmembers, or typical spectral signatures (i.e., mean value in each band of the Landsat image) for each class are derived from these regions. (4)Unlabeled regions are assigned to the class of the most similar neighbour, providing the latter is already classified and bears enough similarity to it, ortherwise they are assigned to the class with the highest relative abundace within the region, where the latter is abundance weighted by distance in the feature space between the spectral signature of the region and the endmember of each class. Once again, adjacent regions having the same class are merged together after the assigment. NB. There are special rules on how to deal with semantically related classes, such as conifer, broadleaf and mixed forest. (5) Regions are reclassified into the final set of 11 landcover classes using a predefined crosswalk table, after which adjacent regions having the same class are merged together. NB. Treed wetlands (class 81 in EOSD) greater than 100 ha are assumed to be black spruce and thus conifer. Smaller treed wetlands are assigned to the class of the closest forest endmembers.(6) Regions smaller than the minimum mapping unit (MMU: 0.5 ha for water and 2 ha for the rest) are merged to their most similar adjacent neighbour. (7) Isolated regions smaller than 25 ha that show a low contrast with the encompassing region were eliminated, as they were found in a previous version to mainly correspond to spurious polygons.(8) Using the center of boundary pixels as initial vertices, the result is converted to an ESRI polygon shapefile using a line simplification algorithm. Once all tiles were ready, the final product was created by merging and dissolving the individual tiles into a seamless layer. After dissolving, a few seam lines were still visible due to differing classifications in the adjoining tiles. The labels of polygons affected by these inconsistencies were harmonized using semi-automated GIS procedures, after which a final 'dissolve' was performed.The circa 2000 map was updated to circa 2010 conditions using a hybrid procedure that included a combination of change detection and GIS techniques. The map was updated for new features appearing in the 2000-2010 period of the following types: forestry cutblocks, wildfires, surface mines, roads and urban development. Cutblocks were extracted from the cumulative ABMI cutblock layer. A difference image (NDVI2000 – NDVI2010) was computed, and statistics for each cutblock polygon were derived. Cutblocks harvested between 2000 and 2010 were preselected by thresholding and those close to the threshold were visually inspected before the final selection. The wildfires were extracted from GoA’s Historical Wildfire Perimeter Dataset using a similar procedure than for the cutblocks. Mines, roads and urban were extracted from the 2010 ABMI Human Footprint dataset using GIS overlay operations that compared these with preexisting features of this type in the 2000 map. The different types of new features were integrated into a single disturbance layer, the outlines of this layer were generalized to give them a similar appearance than those of the 2000 map, the disturbance layer was stamped on top of the 2000 map, and slivers (small polygons resulting from boundary mistmatches) were removed. The Landcover_Polygons_2010 feature class has a new field called ‘MOD_TY’ (modifier type) that records the type of disturbance a new polygon not belonging to the developed class is. Possible values are ‘Cutblock’, ‘Burnt’, ‘Mining’ and ‘Tailing pond’ (the latter belongs to water class). NB. Cutblocks, mines and other disturbance features that already existed in 2000 have their MOD_TY field blank in this version.Landcover class definitions (field LC_class):20 - Water: lakes, lagoons, rivers, canals, and artificial water bodies. Shallow open water is included in this category, unless there is more than 20% vegetation cover, in which case it belongs to the relevant vegetated class.31 - Snow/Ice: areas permanently covered by snow or ice, including glaciers. 32 - Rock/Rubble: bedrock, rubble, talus, blockfield, lava beds, or other natural impervious surfaces. 33 - Exposed Land: bare soil (barren, non-agricultural), river sediments and cutbanks, pond or lake sediments, reservoir margins, beaches, landings, fresh (less than 1yr) cutblocks, recently burned areas, mudflat sediments, surface mining, or other non-vegetated (less than 10% trees, or less than 20% shrub/herb) surfaces. 34 - Developed: urban and built-up areas (including industrial sites), impervious artificial surfaces (e.g. airport runaways), railways and roads. Acreages and farmsteads are included in this class. Oil and gas well pads are included in this class if connected to a road and not abandoned or under reclamation. Urban terrain under development is included in this class, even if the land is exposed. Urban green areas are excluded of this class if larger than 2 ha and if they have less than 2 buildings per hectare. 50 - Shrubland: At least 20% ground cover which is at least one-third shrub (shrub: a woody plant not considered a tree), with no or little presence of trees (less than10% crown closure). Examples of plants belonging to this class in Alberta are alder, willow, juniper, and sagebrush. Shrubby fens and other non-treed woody wetlands, usually associated with floodplains and the shores of lakes and streams, belong to this class. Recent clearcuts and forest fires are initially labeled as shrubland to differntiate them from the surrounding forest.110 - Grassland: Predominantly native grasses and other non-woody vegetation (e.g., forbs) with a minimum of 20% ground cover. May include some shrub cover (but less than a third of the vegetated area) or a few trees (but the tree cover cannot exceed 10%). Land used for range with native unimproved grasses (a.k.a. rough pasture) is included in this class. Alpine meadows fall into this class. Marshes and other non-woody wetlands with at least 20% vegetation cover (sedges, cattails, or moss) belong to this class. NB. A forestry cutblock harvested more than year ago that contains seedlings, but where the latter cover less than 10% of the area of the cutblock, belongs to this class. If the cutblock had no successful regeneration and was covered by more than 20% shrubs, then it would belong to the 'shrubland' class.120 - Agriculture: annually cultivated crops, tame pastures (fields planted or sown with non-native grasses/legumes where livestock is directly grazing on them in the summer), forage crops (same as the previous, but instead of grazed, cut for hay) and woody perennial crops (fruit orchards and vineyards). Bare agricultural (i.e., tilled) soil belongs to this class and not to 'exposed land'.210 - Coniferous Forest: Treed areas with at least a 10% ground cover of trees (a.k.a. crown closure), where coniferous trees (spruce, pine, fir, larch) are 75% or more of the crown closure. Young plantations or regenerating cutblocks of conifer trees belong to this class no matter that the trees are less than 5 m tall, providing crown closure has reached 10%. Treed wetlands (e.g., black spruce bogs and fens) are included in this class providing they are conifer dominated and crown closure exceeds10%.220 - Broadleaf Forest: Treed areas with at least a 10% ground cover of trees, where broadleaf trees (trembling aspen, balsam poplar, white birch) are 75% or more of the crown closure. Young plantations or regenerating cutblocks of broadleaf trees belong to this class no matter that the trees are less than 5 m tall, providing crown closure has reached 10%. Treed swamps along river floodplains and other treed wetlands are included in this class providing they are broadleaf dominated and crown closure exceeds 10%.230 - Mixed Forest: Treed areas with at least a 10% ground cover of trees, where neither coniferous nor broadleaf trees account for 75% or more of crown closure.The overall thematic accuracy of the map, as estimated by an extensive validation dataset (see description of the relevant feature class) is 75% with 11 classes, and 88% if these classes are grouped into 5 general classes. User and Producer accuracies for specific classes in each level appear in tables appended to this geodatabase. Caveats:The width of roads has been systematically exaggerated to a minimum of 60 m (two Landsat pixels); therefore the real extent of the Developed class is greatly overestimated.Forest areas harvested or burned between 2000 and 2010 were assigned to the ‘Shrub’ LC class to differentiate them from their surroundings. They can be identified using the ‘MOD_TY’ field, which will respectively contain the value ‘Cutblock’ or ‘Burnt’.The accuracy of the shrub class is low (30%). Many shrub polygons are in reality forest (beyond new cutblocks and burnt areas), especially in the North. This was not possible to correct with the input data at hand.
Copyright Text: ABMI Remote Sensing Group 2013, based on the EOSD and NLWIS 2000 raster datasets and on hydrography and acces GIS layers from the Government of Alberta. Update to 2010 based on ABMI Human Footprint dataset.