In 2014, the Alberta Energy Regulator (AER) initiated a Play-Based Regulation (PBR) pilot project as a step towards implementation of the Unconventional Regulatory Framework. One of the goals of the PBR pilot is to encourage companies in the unconventional play area to work together on plans for surface development to minimize the numbers of facilities and surface impacts. This dataset is one of a series created using earth observation imagery to assess surface change caused by energy exploration.
The PBR area extends from Twp. 52, Rge. 7, W 5th Mer. to Twp. 70, Rge. 5, W 6th Mer., covering the towns of Edson, Fox Creek, Mayerthorpe, Whitecourt, Swan Hills, and Valleyview.
For this digital data release, a land use and land cover classification dataset was derived from 2005 Landsat multispectral imagery for the PBR pilot area. The classification contains 13 classes: 0 - unclassified, 1 - exposed land/cut blocks/harvested areas, 2 - water bodies, 3 - transitional bare surfaces, 4 - mixed developed areas, 5 - developed areas, 6 - shoal, 7 - shrub land, 8 - grassland, 9 - agriculture areas, 10 - coniferous forest, 11 - broadleaf forest, and 12 - mixed forest. These categories can be used as baseline data for planning, managing and monitoring surface infrastructure needs and impacts.
The overall accuracy, defined as the total correct pixels (sum of the major diagonal) divided by the total number of pixels in the error matrix N, is 2132863/2806708 or 75.99% for this dataset.
The Producer accuracy (Table 1), or omission error, is an indication of the probability of a reference pixel being correctly classified. It is calculated as the total number of correct classified pixels in a class divided by total number of pixels in that class.
Table 1 - Producer accuracy results
Class | Producer Accuracy (%) | Pixel Ratio
----- --------------------- -----------
Developed | 95.15 | 58991/61995
Shrub/grass | 29.90 | 141155/472057
Agriculture | 69.78 | 123748/177344
Coniferous forest | 87.27 | 1027086/1176843
Broadleaf forest | 85.13 | 781883/918469
The User accuracy (Table 2), or commission error, is defined as the total number of correct classified pixels in a class divided by total number of pixels that were classified in that class. It indicates the probability that a pixel classified on the map represents that class on the ground.
Table 2 - User accuracy results
Class | User Accuracy (%) | Pixel Ratio
----- --------------------- -----------
Developed | 74.30 | 58991/79393
Shrub/grass | 51.76 | 141155/272687
Agriculture | 51.32 | 123748/241121
Coniferous forest | 91.92 | 1027086/1117376
Broadleaf forest | 71.35 | 781883/1095821
Accuracy assessment for the fused classification result was performed with validation data (derived from the ground-reference data in step 2) that include developed areas, mixed developed areas, coniferous forest, broadleaf forest, shrub/grass, and agriculture classes. Transient LULC types with mixed forest, exposed land/cutblocks/harvested areas, transitional bare surfaces, and shoal were not used for validation since ground reference data for these classes were not realistic.
developed - mixed developed area (class 5) and developed area ( class 6) were combined to one class
shrub/grass - shrubland (class 8) and grassland (class 9) were combined to one class
agriculture - agricultural area (class 10)
coniferous - coniferous forest (class 11)
broadleaf - broadleaf forest (class 12)
A major component of this dataset is derived from Ducks Unlimited Canada's drained and boreal Enhanced Wetland Classification inventories.
Http://www.capf.ca/pdfs/AWCS%20Draft%20requesting%20Public%20Feedback.pdf
Process steps performed in ENVI 5.1 to produce classification map:
1. Pre-release versions of annual Landsat Best Available Pixel Composite (LBAPC) datasets (1984 to 2012) were obtained from the Natural Resources Canada/Pacific Forestry Centre for testing purposes. Of these, the 2005 and 2006 LBAPC datasets were used to produce this vegetation recovery result. Publicly available annual Landsat Best Available Pixel Composite (LBAPC) datasets for 2012 and 2013 were obtained from the the Department of Geographical Sciences, University of Maryland. Of these, 2012 LBAPC data were used to produce Land Use and Land Cover (LULC) classification result.
2. The Alberta Biodiversity Monitoring Institute (ABMI) LULC data for 2010 and the Alberta Department of Energy (ADoE) oil and gas infrastructure data for 2012 were used to produce the 2012 ground-reference data. Normalized Difference Built-up Index (NDBI) was applied to 2012 and 2010 LBAPC data to produce LULC changes from 2012 to 2010. Ground reference data for 2012 was selected by subtracting these changes from the 2010 ABMI data. Ground-reference vegetation types for 2012 include coniferous forest, broadleaf forest, mixed forest, agriculture, grassland, and shrubland. Ground-reference developed areas for 2012 were queried from the ADoE oil and gas infrastructure data. Training data for vegetation types and developed areas were selected from the ground reference data. Training data for exposed land/cutblocks/harvested land and water were selected manually from the 2012 LBAPC data. Areas with cloud, cloud shadow, and noise were discarded from training and validation data. Maximum Likelihood (ML) classification algorithm, Spectral Angle Mapper (SAM) partial unmixing algorithm, and Constraint Energy Minimization (CEM) partial unmixing algorithm utilized the training data with 2012 LBAPC data to produce LULC classification results.
3. ML classification algorithm was applied to the 2012 LBAPC data to produce LULC classification result with coniferous forest, broadleaf forest, mixed forest, agriculture, grassland, shrubland, exposed land/cutblock/harvested land, and water classes.
4. LULC classification result with developed areas, mixed developed areas, and transitional bare surface classes were produced from the 2012 LBAPC data by dividing the CEM partial unmixing result by the SAM partial unmixing result, followed by the K-Means clustering. Post classification techniques (i.e., majority/minority filter, clump, and sieve) were applied on each of these classes to refine the result and to minimize false detections.
5. A fused classification result was produced by combining LULC classification results from step 3 and 4. A buffer with 90 m was applied on the stream channel datasets from Alberta Energy Regulator (AER) and Alberta Merged Wetland Inventory (AMWI) to track yearly changes associated with landslide activities and the appearance or disappearance of sandbars due to water level fluctuation. This buffer zone is assigned to a separate class called shoal.
6. Accuracy assessment for the fused classification result was performed with validation data (derived from the ground-reference data in step 2) that include developed areas, mixed developed areas, coniferous forest, broadleaf forest, shrub/grass, and agriculture classes. Transient LULC types with mixed forest, exposed land/cutblocks/harvested areas, transitional bare surfaces, and shoal were not used for validation since ground reference data for these classes were not realistic.