| Report Documents
|
|
|
| |
| Map Plotfiles
|
-
No files of this type available
|
| |
| Data Files
|
-
No files of this type available
|
| |
| Digital Map Files
|
-
No files of this type available
|
| |
| Image Document
|
-
No files of this type available
|
| |
| Video Files
|
-
No files of this type available
|
| |
|
All Documents
|
|
|
| Contact
|
|
-
If you have any questions on the information presented, or require additional report data or attachments, please contact the Report Contact
|
|
|
This project is phase II in a multi-phase project towards the development of simplified, unified and easy to understand surficial material mapping to support land use and water allocation decisions in the province. Phase II focused on the development of continuous spatial maps in places with thick surficial deposits and accumulations of unconsolidated sediments with high aquifer potential, such as in valley bottoms.
|
Author: Elyse Sandl, Deepa Filatow and Jennifer Todd
|
|
|
Date Published: Dec 2024
|
Report ID: 63123
|
Audience: Government and Public
|
During this second phase, a classification system was developed that includes two levels of Broad Landform classes. The first level is a broad landform class that includes two categories, Eroding Upland and Valley Fill. At the second level, the valley fill category is further divided into seven classes such as texture and aquifer type. The Southern Interior EcoProvince was selected as the study area for piloting a modelling approach to generate predictive maps using a random forest model to generate preliminary maps of each broad landform class. Training points were randomly generated within five mapsheet areas in the Clinton, Manning Park and Westwold areas. The training points were attributed by a terrain expert with a categorical value from each surficial material class, based on aerial imagery, elevation contours, information about the geological and glacial history of the area. The random forest models were used to identify the relationships between a surficial material class and 31 covariate raster layers. For each class, two models were developed, one with the original training points and a second where the number of training points within each category was synthetically balanced. Internal model metrics provide an estimate of model performance within the five mapsheets with training points. Overall, within the training point mapsheet areas, the models had high internal accuracy rates (98-99%) and balancing the training data consistently improved model performance. As the terrain and landscape within the Southern Interior EcoProvince is variable, training data from the five mapsheets did not sufficiently capture all of the landscape characteristics of the study area. Some of the predictive maps did not match professional knowledge in areas that had different characteristics than the training point mapsheets. This modelling pilot revealed that when generating predictive maps, it is important to be mindful of capturing the variability of the covariate distribution with your training data and of extending predictions beyond the rule shed or areas represented by the training data. Attribution of training points and random forest modelling is a cost-effective way to map simple discrete categories that are discernable from image interpretation. In future phases of the project the classification and predictive mapping approach can be rolled out across the province to create consistent surficial material mapping layers for B.C.
|
Report Type
Subject
| |
Region - Okanagan |
| |
Terrestrial Information - Terrain Mapping |
| |
Water Information - Groundwater |
| |
Water Information - Water Management |
| |
|
|
