Client: Western University, Ute Mountain Ute Tribe, University of Toronto Scarborough, Haliburton Highlands Trust

Year: 2022 - Present

Focus Area: Landscape change (Image classification)

Tools: SyncroSim, ecoClassify, ST-Sim

 

Project Overview

Ecologists are increasingly challenged by the overwhelming volume of remote sensing data available, making it difficult to extract the information needed for conservation and management decisions. Machine learning techniques for image classification offer a powerful solution, helping to automate manual tasks, improve accuracy, and unlock novel ecological insights.

We used this approach in a partnership with Western University to classify alpine snow cover, a notoriously difficult task due to the complex terrain and atmospheric conditions typical of mountain environments. Yet, understanding patterns of snow cover is critical in alpine ecosystems, where it strongly influences biodiversity. Using the ecoClassify package within the SyncroSim modeling framework, we developed an image classifier for snowpack across 23 alpine meadows on Jumpingpound Mountain in Kananaskis, Alberta (Fig. 1). The model incorporated high-resolution Sentinel-2 imagery and a Digital Elevation Model (DEM) to generate a decadal history (2014–2024) of snow cover patterns.

The outputs of this image classification were then directly used to inform state and transition models with ST-Sim. Fine-scale estimates of historical snow cover provided by the classifier were used to set baseline conditions for simulation. These forecasts included overwinter drivers of snowpack including sublimation, wind ablation, snow fall, and snow accumulation, under different scenarios of climate change. We then linked these environmental changes to models of population dynamics and gene flow for the Rocky Mountain Apollo butterfly (Parnassius smintheus) and its host plant (Sedum lanceolatum), enabling us to evaluate future population viability and habitat connectivity.

More applications of ecoClassify

In addition to this alpine snowpack project, ecoClassify has supported several other applications across North America:

1. Mapping Invasive Tamarisk in Mancos Canyon, Colorado (2024)
   To support the Ute Mountain Ute Tribe’s restoration efforts, we used ecoClassify to map tamarisk (Tamarix spp.), an invasive species of concern in Mancos Canyon. We used machine learning methods on Sentinel-2 imagery and LANDFIRE vegetation data to produce:

• A tamarisk distribution map for establishing a baseline and monitoring remediation efforts
• A SyncroSim library with a trained model for future classification

2. Inventorying Urban and Peri-Urban Agriculture (UPA) in Toronto, Ontario (2024-Present)
   As part of the “Toward Sustainable Urban and Peri-urban Agriculture for Net-zero Food Systems (TOSustain)” research program, we used ecoClassify to create an inventory of existing and potential UPA land across Toronto. These classifications support efforts to expand low-emission food production systems in urban Canada. More information is available at TOSustain.

3. Wetland Mapping in Haliburton County, Ontario (2024–2025)
   In partnership with the Haliburton Highlands Land Trust, we developed a machine learning pipeline using ecoClassify to generate reliable, scalable wetland maps (Fig. 2). The project combined high-resolution remote sensing data with a Digital Elevation Model (DEM) to produce a wetland inventory for different major classes of wetlands (e.g., bog, fen, marsh, swamp) and establish a framework for future mapping.