Predictive Ecosystem Mapping of the Chinchaga Watershed Management Basin, North Eastern British Columbia

Input data for this project consisted of Sentinel 2 spectral imagery, the provincial Digital Elevation Model, ClimateBC data, and Predictive Ecosystem Mapping provincial products. Both spectral and topographic indices were created to assist in model performance. A geographically stratified sampling approach was taken in order to generate 1012 points.

These training points were attributed to allow for the prediction of three simplified land cover realms, as well as seven identified provincial Eco-Groups across the Chinchaga. Together, the explanatory variables and training points were fed into a Random Forest model, and the classes were predicted across the study area with 84.9% (Three Class) and 78.2% (Seven Class) overall accuracy, respectively. Included in this report are a brief discussion on explanatory variable contribution to model performance, as well as an analysis of sample design and its implications on class accuracies in the 7 Class prediction.