Wetlands - Broad Scale - Project Extents

The "WETLAND_BROAD_POLY" dataset is a composite of several individual wetland mapping projects, collectively referred to as "Wetlands - Broad Scale". The methodology used, map accuracy, and credits vary by project; these project details are outlined in Table 1 (see below). The spatial extent of individual project areas are delineated in this feature class.

The "Wetlands - Broad Scale" dataset is intended to be used as a broad scale planning and management tool to identify potential distribution and abundance of wetlands. Wetlands were mapped to wetland class (shallow water, marsh, swamp, fen, and bog), following the Canadian Wetland Classification System using a predictive model. This dataset is intended to support land management and regional land use planning processes. Local scale (10k) manual wetland mapping, and additional physical assessments (i.e. ground inspections) may be required to undertake habitat enhancement, environmental assessment, reclamation planning, or environmental mitigation over small to moderate areas.

[

[Map development and attributes:

The "Wetlands - Broad Scale" dataset was developed using a random forest machine learning model to predict wetland classes. Various satellite imagery sources and landscape variables derived from a digital elevation model (DEM) were used as primary inputs to predict wetlands. The source dataset has a resolution of 10 x 10 m. Training and validation data are a mix of ground plots (site visit and ecosystem plots), aerial survey plots, and interpreted polygons. Each predictive wetland map within the composite has met the minimum criteria of a map accuracy greater than or equal to 70 % and a Kappa coefficient greater than 0.60.

The map (or producer's) accuracy measures the percentage of wetland features that are correctly classified to one of the five wetland classes.

The Kappa coefficient statistic is used to measure the extent to which the model has correctly predicted, given the set of validation data. A value of 0 indicates predicted values are entirely random. A value of 1 indicates a perfect model. As a general rule, Kappa coefficients less than 0.60 indicate a poorly performing model, values of 0.61 to 0.80 indicate substantial agreement between predicted and validation data, and values of 0.81 to 1.00 indicate almost perfect agreement.

The size of the smallest wetland that can be reliably mapped, the Target Mapping Unit (TMU), was not established for this dataset at the time of publication. Wetland classes smaller than a TMU of 2.0 hectares in this dataset should be used with caution. Wetlands below the TMU have a higher potential to be associated with classification error. The reported map accuracy is adequate for the intended purpose, and assumes that training data has adequately captured variation in landscape and vegetation structure between and within wetland classes.

[

[Table 1: Unique project details for each mapping area within the "Wetlands - Broad Scale" dataset.

]{style="font-weight:bold;"}]{style="font-weight:bold;"}]{style="font-weight:bold;"}]{style="font-weight:bold;"}

+-----------------------+-----------------------+-----------------------+ | Project Name\ | Information\ | Description\ | | \ | \ | \ | +-----------------------+-----------------------+-----------------------+ | Beaver River\ | Last Update\ | October 2019\ | | \ | \ | \ | +-----------------------+-----------------------+-----------------------+ | | Project Area\ | The Beaver River | | | \ | Watershed wetland map | | | | is located in east | | | | central Yukon and has | | | | a total area of 6,146 | | | | km2. The wetland map | | | | consists of the | | | | Beaver River | | | | watershed, including | | | | the Rackla and East | | | | Rackla rivers, and a | | | | portion of the Keno | | | | Ladue watershed.\ | | | | \ | +-----------------------+-----------------------+-----------------------+ | | Methods\ | Sentinel-1, | | | \ | Sentinel-2 and | | | | landscape variables | | | | derived from the | | | | ArcticDEM, version 3, | | | | were used as primary | | | | inputs to predict | | | | wetlands. Sentinel | | | | imagery was from | | | | 2018. Training and | | | | validation data was | | | | comprised of 250 | | | | ground plots (site | | | | visit and ecosystem | | | | plots), 264 aerial | | | | survey plots, and | | | | 1,621 interpreted | | | | polygons. Polygons | | | | were interpreted from | | | | a combination of | | | | SPOT-6, Pleiades-1, | | | | ESRI World Imagery, | | | | and Sentinel-2 | | | | imagery. Training | | | | polygons reflected | | | | the extent and | | | | variability of nine | | | | land cover classes | | | | within the planning | | | | area and are in | | | | proportion to the | | | | aerial extent of land | | | | cover class | | | | (including wetland | | | | classes). Interpreted | | | | polygons were used to | | | | train the model. The | | | | model was validated | | | | using point location | | | | of aerial field calls | | | | and ground plots. The | | | | ratio of training to | | | | validation data was | | | | 3:1. In this dataset, | | | | the smallest mapped | | | | wetland, or minimum | | | | mapping unit (MMU), | | | | is 2 pixels or 200 | | | | square metres. The | | | | pixel resolution of | | | | the wetland map is 10 | | | | m, however all single | | | | pixels were merged | | | | into to their | | | | neighbouring pixel | | | | value.\ | | | | \ | +-----------------------+-----------------------+-----------------------+ | | Accuracy\ | The final | | | \ | classification map | | | | accuracy was 81 % | | | | with a Kappa of 0.77 | | | | across all wetland | | | | and land cover | | | | classes. Isolating | | | | specifically the | | | | wetland classes, the | | | | map accuracy was 78 % | | | | (Kappa 0.69). The | | | | resulting map | | | | accuracy meets the | | | | project goal of | | | | greater than 70 % | | | | accuracy for a | | | | predictive map | | | | produced at a survey | | | | level intensity 4 to | | | | 5 as per the ELC | | | | guidelines for | | | | mapping.\ | | | | \ | +-----------------------+-----------------------+-----------------------+ | | Credits\ | Preliminary wetland | | | \ | classification and | | | | final predictive map | | | | was completed by the | | | | Ecosystem and | | | | Landscape | | | | Classification (ELC) | | | | Program, Fish and | | | | Wildlife (F&W) | | | | Branch, Department of | | | | Environment, | | | | Government of Yukon, | | | | the Government of | | | | Yukon with input from | | | | Palmer Environmental | | | | Consulting Group, | | | | Drosera Ecological | | | | Consulting, and | | | | CryoGeographic | | | | Consulting. Training | | | | and assessment data | | | | was collected by | | | | Drosera Ecological | | | | Consulting, Lori | | | | Schroeder Consulting, | | | | CryoGeographic | | | | Consulting, and F&W | | | | staff. Classification | | | | of wetlands was | | | | completed by | | | | CryoGeographic | | | | Consulting with input | | | | from Drosera | | | | Ecological Consulting | | | | and ELC program | | | | staff.\ | | | | \ | +-----------------------+-----------------------+-----------------------+ | Peel\ | Last Update\ | March 2022\ | | \ | \ | \ | +-----------------------+-----------------------+-----------------------+ | | Project Area\ | The Peel Watershed | | | \ | wetland map is | | | | located in northern | | | | Yukon and has a total | | | | area of 67,366 km2. | | | | The watershed is | | | | drained by six major | | | | tributaries\u2014the | | | | Snake, Wind, Bonnet | | | | Plume, Hart, Ogilvie, | | | | and Blackstone.\ | | | | \ | +-----------------------+-----------------------+-----------------------+ | | Methods\ | Sentinel-1, | | | \ | Sentinel-2, ALOS | | | | PALSAR (HH and HV | | | | polarizations), and | | | | landscape variables | | | | derived from the | | | | ArcticDEM, version 3, | | | | were used as primary | | | | inputs to predict | | | | wetlands. Sentinel | | | | imagery was from | | | | 2018. Segmented | | | | objects were used to | | | | assign a wetland | | | | class* and can be | | | | considered the | | | | minimum map unit | | | | (MMU) (as opposed to | | | | a single pixel). | | | | Segments were created | | | | using eCognition | | | | software with | | | | Sentinel-2 bands | | | | (red, green, blue, | | | | NIR, and SWIR) at 10 | | | | m spatial resolution. | | | | Segmentation scale | | | | parameters were 175 | | | | and 100. Shape and | | | | compactness | | | | parameters were 0.1 | | | | and 0.9, | | | | respectively. | | | | Training and | | | | validation data was | | | | comprised of 1,122 | | | | interpreted polygons | | | | trained on 514 ground | | | | plots (site visit and | | | | ecosystem plots). | | | | Polygons were | | | | interpreted from a | | | | combination of SPOT-6 | | | | and ESRI World | | | | Imagery. Interpreted | | | | polygons were | | | | randomly assigned by | | | | class to training and | | | | validation datasets | | | | (748 and 374, | | | | respectively). The | | | | ratio of training to | | | | validation data was | | | | 3:1.\ | | | | \ | +-----------------------+-----------------------+-----------------------+ | | Accuracy\ | The final | | | \ | classification map | | | | accuracy was 75 % | | | | with a Kappa of 0.70 | | | | across all wetland | | | | and land cover | | | | classes. Isolating | | | | specifically the | | | | wetland classes, the | | | | average map accuracy | | | | was 79 % with a Kappa | | | | of 0.69. Swamp and | | | | marsh had the lowest | | | | map accuracy with 70 | | | | % and 69 % | | | | respectively. The | | | | resulting map | | | | accuracy meets the | | | | project goal of | | | | greater than 70 % | | | | accuracy for a | | | | predictive map | | | | produced at a | | | | reconnaissance survey | | | | level as per the ELC | | | | guidelines for | | | | mapping.\ | | | | \ | +-----------------------+-----------------------+-----------------------+ | | Credits\ | Preliminary wetland | | | \ | classification and | | | | final predictive map | | | | was completed by Wood | | | | Environment & | | | | Infrastructure | | | | Solutions under | | | | contract to Ecosystem | | | | and Landscape | | | | Classification (ELC) | | | | Program, Fish and | | | | Wildlife (F&W) | | | | Branch, Department of | | | | Environment, | | | | Government of Yukon. | | | | Ground plot data used | | | | for training and | | | | validation data was | | | | collected by various | | | | Government of Yukon | | | | staff and contractors | | | | between 1975 and | | | | 2020.\ | | | | \ | +-----------------------+-----------------------+-----------------------+ | Mayo McQuesten\ | Last Update\ | March 2022\ | | \ | \ | \ | +-----------------------+-----------------------+-----------------------+ | | Project Area\ | The Mayo and | | | \ | McQuesten Watersheds | | | | wetland map is | | | | located in central | | | | Yukon and has a total | | | | area of 7,514 km2. | | | | This project focused | | | | on enhancing map | | | | accuracy within the | | | | Mayo and McQuesten | | | | watershed's 2,495 | | | | km2 sub-basins of | | | | Haggard Creek, Mayo | | | | Lake, Sprague Creek, | | | | and the Lower-South | | | | and Mid-South | | | | McQuesten (hereafter | | | | sub-basins).\ | | | | \ | +-----------------------+-----------------------+-----------------------+ | | Methods\ | Se Sentinel-1, | | | \ | Sentinel-2, ALOS | | | | PALSAR (HH and HV | | | | polarizations), and | | | | landscape variables | | | | derived from the | | | | ArcticDEM, version 3, | | | | were used as primary | | | | inputs to predict | | | | wetlands. Sentinel | | | | imagery was from | | | | 2018. Segmented | | | | objects were used to | | | | assign a wetland | | | | class* and can be | | | | considered the | | | | minimum map unit | | | | (MMU) (as opposed to | | | | a single pixel). | | | | Segments were created | | | | using eCognition | | | | software with | | | | Sentinel-2 bands | | | | (red, green, blue, | | | | NIR, and SWIR) at 10 | | | | m spatial resolution. | | | | Segmentation scale | | | | parameters were 200 | | | | and 50. Shape and | | | | compactness | | | | parameters were 0.1 | | | | and 0.9, | | | | respectively. | | | | Training and | | | | validation data were | | | | comprised of 482 | | | | interpreted polygons | | | | trained on 98 ground | | | | plots (site visit and | | | | ecosystem plots), and | | | | drone imagery. | | | | Polygons were | | | | interpreted from a | | | | combination of | | | | SPOT-6/7, ESRI World | | | | Imagery, Sentinel-2 | | | | imagery, and 2021 | | | | drone imagery taken | | | | over 5 different | | | | locations, totaling | | | | 9.3 km2. Interpreted | | | | polygons were | | | | randomly assigned by | | | | class to training and | | | | validation datasets | | | | (232 and 250, | | | | respectively). The | | | | ratio of training to | | | | validation data was | | | | approximately 1:1.\ | | | | \ | +-----------------------+-----------------------+-----------------------+ | | Accuracy\ | Over the entire study | | | \ | area, the final | | | | classification map | | | | accuracy was 97 % | | | | with a Kappa of 0.95 | | | | across all wetland | | | | and land cover | | | | classes. Isolating | | | | specifically the | | | | wetland classes | | | | within the study | | | | area, the average map | | | | accuracy was 89 %. | | | | Within the | | | | sub-basins, the final | | | | classification map | | | | accuracy was 97 % | | | | with a Kappa of 0.93 | | | | across all wetland | | | | and land cover | | | | classes. Isolating | | | | specifically the | | | | wetland classes | | | | within the | | | | sub-basins, the | | | | average map accuracy | | | | was 88 %. The | | | | resulting map | | | | accuracy meets the | | | | project goal of | | | | greater than 70 % | | | | accuracy for a | | | | predictive map | | | | produced at a level 4 | | | | to 5 survey level as | | | | per the ELC | | | | guidelines for | | | | mapping.\ | | | | \ | +-----------------------+-----------------------+-----------------------+ | | Credits\ | Preliminary wetland | | | \ | classification and | | | | final predictive map | | | | was completed by Wood | | | | Environment & | | | | Infrastructure | | | | Solutions under | | | | contract to the | | | | Ecosystem and | | | | Landscape | | | | Classification (ELC) | | | | Program, Fish and | | | | Wildlife (F&W) | | | | Branch, Department of | | | | Environment, | | | | Government of Yukon. | | | | Ground plot data used | | | | for training and | | | | validation data was | | | | collected by F&W | | | | staff, CryoGeographic | | | | Consulting, and Wood | | | | Environment & | | | | Infrastructure | | | | Solutions in summer | | | | 2021.\ | | | | \ | +-----------------------+-----------------------+-----------------------+

*Mahdavi, S., B. Salehi, M. Amani, J.E. Granger, B. Brisco, W. Huang, and A. Hanson. 2017. Object-based Classification of Wetlands in Newfoundland and Labrador using Multi-temporal PolSAR Data. Canadian Journal of Remote Sensing, 43(5), 432-450.

Distributed from GeoYukon by the Government of Yukon. Discover more digital map data and interactive maps from Yukon's digital map data collection.

For more information: geomatics.help@yukon.ca