Thompson-Okanagan Ecosystem Explorer

Technical Documentation

Technical documentation related to the Thompson-Okanagan Ecosystem Explorer, co-developed by Okanagan Collaborative Conservation Program and Thompson-Nicola Conservation Collaborative, is found below.

Return to Ecosystem Explorer

Supplementary Documentation for the Thompson-Okanagan Wetland Explorer Tool, Predictive Model, and Data Layers

Written by Kristina Deenik, MSc, RPBio & Evan Lavine, BES, ADGIS

2025-04-15

1. Wetlands Explorer - layers explained

Study Areas: The study area is based on select watershed boundaries with an additional area included for the proposed National Park Reserve in the South Okanagan-Similkameen. The Okanagan study area is differentiated because LiDAR was available for a higher resolution analysis.

  • Okanagan
  • Thompson Nicola and Okanagan

Species at Risk: Species at Risk (SAR) and Critical Habitat (CH) data are shown here from the BC Data Catalogue. The “Occurrence - Various Sources” dataset was created by combining WHF_Survey, WHF_Incidental, WSI_Survey and WSI_Incidental dataset (WHF = wildlife habitat feature, WSI = wildlife species inventory as per the BC Ministry of Environment). The datasets were filtered based on the following criteria: SAR points must either be provincially listed as either a RED or BLUE, or either a 1 or 3 on the SARA schedule, or be considered extinct, extirpated, endangered, threatened or special concerns in the COSEWIC status. The Conservation Data Centre (CDC) SAR polygons were also used and filtered using the same approach; however, due to the size and span of the American Badger and Caribou polygons, these two were excluded from the analysis. The CDC polygons were then converted to multiple points and combined into a single point dataset along with the WSI and WHF points. This combined point layer was used to identify the closest SAR to a wetland.

Administrative Boundaries: Federally and provincially available datasets.

Protected Areas: Federally and provincially available datasets.

Freshwater Atlas: Provincially available datasets.

Global Surface Water Dynamics 1999-2021: Annual Water Coverage from 1999-2023 produced by Global Land Analysis and Discovery (GLAD) laboratory in the Department of Geographical Sciences at the University of Maryland (Pickens et al. 2020). They classified land and water in all 3.4 million Landsat 5, 7, and 8 scenes (30-m spatial resolution) from 1999 to 2018 and performed a time-series analysis to produce maps that characterize inter-annual and intra-annual open surface water dynamics. We are displaying here the Permanent, Seasonal or Ephemeral and the Loss classes from their Interannual Dynamics Classes.

Wetland Model - Thompson Nicola: The “High Probability Wetlands 18m Resolution” layer shows wetlands predicted by the model with >90% probability in the Thompson, Nicola and South Okanagan areas where LiDAR was not available. This spatial file has been attributed with information relating to proximity and diversity of SAR, critical habitat, streams and other wetlands. The “Thompson Nicola Model Probability 0-100” is a raster layer that displays the model-assigned probability (0 to 100%) that each pixel in the study area represents a wetland. It provides a continuous surface of likelihood values, allowing users to visualize areas with lower confidence that may be missed by a strict ≥90% threshold. This layer is especially useful for identifying potential wetland margins, areas of hydrologic connectivity, and subtle linkages between wetlands and nearby water bodies.

  • High Probability Wetlands 18m Resolution
  • Thompson Nicola Model Probability 0-100

Wetland Model - Okanagan: The “High Probability Wetlands 3m Resolution” layer shows wetlands predicted by the model with >90% probability in the Okanagan where LiDAR was available. This spatial file has been attributed with information relating to proximity and diversity of SAR, critical habitat, streams and other wetlands. The “Okanagan Model Probability 0-100” is a raster layer that displays the model-assigned probability (0 to 2
100%) that each pixel in the study area represents a wetland. It provides a continuous surface of likelihood values, allowing users to visualize areas with lower confidence that may be missed by a strict ≥90% threshold. This layer is especially useful for identifying potential wetland margins, areas of hydrologic connectivity, and subtle linkages between wetlands and nearby water bodies.

  • High Probability Wetlands 3m Resolution
  • Okanagan Model Probability 0-100
2. Case Study - layers explained

Kamloops Fringe Area and Municipal Boundary: The study area for this case study.

Historic Loss and Drought Intolerant Wetlands: This layer identifies wetlands that show signs of both long-term water loss and sensitivity to drought. Specifically, it identifies wetland areas where open water has decreased over time (since 1984/1999) based on global surface water layers, and where water extent diminished during drought years compared to normal years. Three Landsat-derived global datasets at 30 m spatial resolution were used to identify historic wetland water loss: the Global Surface Water Explorer, the Dynamic Surface Water Maps of Canada, and the Global Surface Water Dynamics dataset. Each provides a different perspective on long-term surface water trends—ranging from summarized occurrence and transitions, to a Canada-optimized annual water map, to a full time series of water presence percentages. Wetlands showing water loss in any of these datasets were flagged as having experienced historic decline. To enhance detection of small or variable wetlands not well captured by global datasets, a Random Forest model was applied to Sentinel-2 imagery at higher spatial resolution or 10-m. Wetlands assessed had a modeled probability of 50% or greater of being wetlands and contained visible open water during at least one observation period. Water extent was mapped for drought vs. non-drought years to assess the hydrologic characteristic of wetlands. Two drought periods were analyzed: 2020 (normal) vs. 2021 (drought) and 2022 (normal) vs. 2023 (drought). By comparing these time periods alongside historical water trends, the map identifies wetlands that experienced open water loss during droughts and those that remained wetted—offering insight into their function and ecosystem services.

Drought Tolerant and Historically Stable Wetlands: are those that remained wetted during both normal and drought years and have not experienced long-term loss of wetted area over time.

3. High-Level User Tips

Exploring Wetland Probability and Extent:

  • Helpful for understanding regional patterns and observing watershed-scale dynamics, wetland clustering, or connectivity trends across landscapes.
  • Go beyond the 90% “High Probability” wetland layer and explore the full 0–100% wetland probability raster. This can reveal transitional zones, potential wetland margins, and landscape connections between wetlands.
  • In the Okanagan, use the 3 m LiDAR-based wetland layers to examine small or complex wetland features not well captured at 18 m resolution.
  • Clicking on individual wetland polygons opens attribute tables with valuable information such as species at risk, proximity to streams, and other nearby wetlands—useful for assessing conservation potential or ecological importance.
  • Use the map to explore whether nearby wetlands fall within protected areas, face development pressure, or have experienced historical hydrologic changes.

Interpreting Water Loss and Drought Sensitivity:

  • Use the Historic Loss and Drought-Intolerant Wetlands layer to assess vulnerability: Wetlands flagged in this layer have shown both long-term water loss and drought-related drying. These sites may be less resilient to climate stress and warrant further attention.
  • Use the Drought-Tolerant Wetlands layer to identify resilient sites: These wetlands retained water during drought and show no historic loss, suggesting strong natural buffering (e.g., groundwater input) and potential value as climate refugia.
  • Compare across years: Visual comparisons of drought years (2021, 2023) with normal years (2020, 2022) can reveal wetlands that lose open water during dry conditions—indicating drought sensitivity. Download these layers to explore them on your own.

Applying the Layers for Planning and Assessment:

  • Overlay wetland layers with species at risk, stream networks, and disturbance indicators to assess both the likelihood of wetland presence and its ecological or functional significance.
  • Consider the distribution of wetlands in relation to one another—such as drought-intolerant clusters or high-probability corridors—when assessing ecological connectivity or vulnerability.
  • These layers are useful for identifying candidate sites for restoration, protection, or further study—especially within municipal planning, watershed management, or climate adaptation initiatives.
  • Incorporate into field studies and environmental assessments: Use model outputs to prioritize field validation, locate headwater wetlands, assess nearby stressors, or explore potential carbon storage areas.

Caveats and Model Limitations:

  • Not a regulatory tool: These layers are designed for planning, prioritization, and exploration—not for regulatory or site-level decision-making. Field verification is essential to confirm wetland presence, extent, and function.
  • Field validation remains essential: Ground-truthing and local knowledge are particularly important in complex or marginal areas where remote sensing may struggle to distinguish wetland features.
  • Screening, not certainty: Treat model outputs as a guide to support further investigation, not as definitive representations of wetland status.
  • Resolution limits: With a spatial resolution of 18 m (or 3 m in LiDAR-optimized areas), small or narrow wetlands may be underrepresented or missed—especially in areas with complex microtopography or fine-scale hydrologic features.
  • Training data bias: The model was trained using existing wetland inventories (e.g., FWA), which may be biased toward non-treed wetlands. Some wetland types may be underrepresented or misclassified, particularly in forested or transitional environments.
  • Global open water datasets limitations: These datasets may miss small water features due to a spatial resolution of 30m. These datasets have their own lists of caveats from the original data sources and reports, which should be reviewed prior to use or interpretation.
  • Probabilistic output interpretation: The Random Forest model outputs a probability for each pixel based on the proportion of decision trees that classify it as a wetland. A 90% probability means that 90% of the trees voted for the wetland class—but this should not be interpreted as a 90% statistical confidence. These scores reflect model agreement, not statistical confidence. While high-probability areas (e.g., ≥90%) are more likely to be true wetlands, lower-probability areas can still contain wetlands—especially in regions with ambiguous input signals or limited training representation. ○ Threshold caution: A strict ≥90% cutoff for “high probability” wetlands may exclude some valid wetlands with lower scores. Use the full 0–100% probability surface for a more nuanced view, especially around wetland margins or in uncertain landscapes. ○ Regional performance variation: Model performance varies based on regional data availability and quality. For instance, the Okanagan region benefits from LiDAR data, improving the accuracy of elevation, topographic characteristics and flow modeling, which may not be available in other areas.
  • Snapshot in time: Models are based on data from limited time windows (e.g., summer 2019). Wetland conditions vary seasonally and interannually, so the outputs represent a temporal snapshot—not a permanent state.
  • Classification uncertainty: Despite high overall accuracy (e.g., 94% in Thompson-Nicola), misclassifications can occur—especially in transitional zones (e.g., riparian areas, moist forests) and because wetlands are dynamic.
  • Climate change considerations: Wetland extent, water levels, and hydrological function are sensitive to both short-term weather events and long-term climate trends. Changes in temperature and precipitation patterns due to climate change—such as increased drought frequency, altered snowmelt timing, or extreme rainfall events—can significantly influence wetland dynamics. For example, wetlands that currently appear stable may become more prone to drying, while others may expand or shift spatially in response to changing water availability. Because the models and datasets in the Wetland Explorer are based on historical and recent satellite imagery, they provide a snapshot of wetland conditions under past and present climate regimes. They do not incorporate projections or predictive modeling of future change. Therefore, users should interpret results in light of possible future hydrologic shifts, especially when planning for long-term conservation, restoration, or land use decisions.