Atticus Environmental Services Ltd.



For more detailed information refer to:

Atticus Environmental Services Ltd. (March 2021). FSJ Wetland Mapping Pilot Project. Technical Report prepared for the Province of BC and Fort St. John Land and Resource Management Planning.

Project Lead / Senior Ecologist: Terry Conville

Technical Analyst: Ramon Melser



Project Description:

This FSJ Wetland Mapping Pilot Project mapped/identified wetlands within the Chinchaga and Chowade watersheds (within the Fort St. John (FSJ) Timber Supply Area (TSA)), following the machine learning (ML) and data mining approach outlined in the published report Predictive Wetland Mapping of the Williston Drainage Basin (Filatow, Harvey, Carswell, and Cameron, 2020).

This current project mapped the location of wetlands and other simplified land cover realms (i.e. water, wetland, upland, and alpine/subalpine) utilizing photo-interpreted training point data and Random Forest (RF) decision tool software. The pilot project also tested the prediction model’s ability to identify and map more detailed ecosystem groups (wetland classes, alpine realms, and upland land classes) within each of the study areas.

Overall, the modeling appears to perform very well when identifying the abundance and distribution of simplified land cover realms (i.e. water, wetland, upland, and alpine), with a combined score between 85-90% for all classes (in each study area). Wetlands themselves were predicted 90% of the time in the Chinchaga but only 60-68% of the time in the Chowade. The Chinchaga 7-Class model achieved an overall validation score of 78%, and the Chowade 12-Class model achieved an overall validation score of 77%. However, a wide range of scores were obtained (from poor for some to substantial agreement for others) for individual group/class predictions when compared with the training point dataset.

This pilot project (and approach) is useful for identifying and mapping wetlands at the realm class level in both low elevation gentle terrain as well as in sloping and rugged terrain at moderate to high elevations. The results show that when compared to existing wetland spatial data, the accuracy and spatial distribution of wetlands appears to be improved and quite reasonable using this modeling approach and does not appear to under or over represent the amount of wetlands in either study area. This is important in the FSJ TSA, as wetlands overall are known to be underrepresented and there is a lack of consistent wetland data in the FSJ TSA (Atticus, 2021).

The scores represent how well the prediction model agrees with the photo interpreted data. These results need to be tested against field information before finalized validations can be obtained. Further, the results and scores provided, although objectively obtained, are internally generated, and do not replace the need for an independent third party accuracy assessment.

Chowade Watershed - Random Forest Classification Results

The map below showcases the predicted raster outputs of a Random Forest modelling approach taken to identify and predict both Land Cover Realms (3 Class) and Ecological Groups (12 Class) within the Chowade Watershed Basin.
Note: Anthropogenic disturbance (i.e. roads as well as oil and gas development and associated infrastructure) is not modelled but overlaid separately from the prediction process.

CLICK BUTTON IN BOTTOM LEFT TO TOGGLE LAYERS



Chinchaga Watershed - Random Forest Classification Results

The map below showcases the predicted raster outputs of a Random Forest modelling approach taken to identify and predict both Land Cover Realms (3 Class) and Ecological Groups (7 Class) within the Chinchaga Watershed Basin.
Note: Anthropogenic disturbance (i.e. roads as well as oil and gas development and associated infrastructure) is not modelled but overlaid separately from the prediction process.

CLICK BUTTON IN BOTTOM LEFT TO TOGGLE LAYERS