Leveraging New Geospatial Tools and Technology for Mapping Wetland across Alberta’s Boreal Forests

Wetlands are a valuable part of our environment. They help clean our water by filtering out pollutants and trapping sediments, they mitigate floods by slowing down and storing fast-moving run-off, and they are biodiversity hotspots for all sorts of plant and animal species. If we don’t know where these wetlands are, however, it makes it very hard to manage, conserve, and protect them. This is why mapping wetlands is important. It is also important to keep mapping them, so that we can monitor them and notice when they change, particularly if this change goes beyond expected seasonal or year-to-year dynamics.

Mapping wetlands in Alberta can be downright tricky. Not only is Alberta a very large place (roughly 661,800 km^2, or the size of Germany and Italy combined), but the size and type of wetland environments vary considerably across the province. Peatlands, in particular, cover large portions of Alberta’s remote boreal forests, while small but important prairie potholes, which are often mineral wetlands, are scattered throughout the province’s southern agricultural regions. Previous efforts undertaken by various organizations – e.g., Government of Alberta and Ducks Unlimited – have generally focused on one particular portion of Alberta, and have each employed a unique set of data sources and mapping techniques. Thirty-three of these individual inventories make up the Alberta Merged Wetland Inventory – an agglomeration of data sets brought to a common thematic standard (following the Canadian Wetland Classification System). The inventory covers the majority of the province with the exception of areas in the National Parks. While this is certainly the most comprehensive wetland map of Alberta available, the inventory’s data gaps and inconsistencies limit its use.

In response to the need for a consistent, comprehensive, and up-to-date map of Alberta’s wetlands, we, and our collaborators at the Geospatial Centre of the Alberta Biodiversity Monitoring Institute (ABMI) decided to leverage some recent technological advances and trends in the geospatial sciences to tackle this challenge. First, we took advantage of the high volumes of open-access satellite data currently available. We employed Sentinel-1 (radar) and Sentinel-2 (optical) imagery, produced as part of the European Space Agency’s Copernicus mission. These data are produced at a high frequency (every 5 days), at a spatial resolution of 10 m, and are freely available. They provide an ongoing, reliable source of information on Earth’s land surface conditions, and therefore support ongoing, future updating of map products that incorporate them.

Second, we capitalized on the use of a freely-available, cloud-computing platform (Google Earth Engine) in order to both access and process these large amounts of satellite data without having to rely on costly local data storage and processing power, or the long time periods such work would require. This means that not only can we access and process data in a speedy fashion, but we can do so with a lot of data. Plus, when we’d like to update our map, we can do so in a timely, and cost-efficient manner. As I mentioned, continuing to map and monitor wetlands and any changes we see in them over the long term is important. As our climate changes, and as Alberta’s landscape undergoes further human development and resource extraction, wetlands can be affected. We need to be aware of this, and be able to keep ourselves up to date on current wetland conditions.

Finally, we employed an increasingly popular type of modeling algorithm – machine learning – to predict wetland location. These empirical algorithms are flexible, scalable, and can handle large volumes of data of different types. This means that if when we want to expand on the small area we are working on for testing and validation purposes, we can reproduce our approach and scale it up to a larger area without too much difficulty.

Flowchart of methods used to model wetland occurrence by the AGRG.

Bringing all these tools and technology together, we modeled probability-of-wetland-occurrence (i.e., the probability that wetland exists at any given location) across an initial study area covering 13,700 km^2 of northwestern Alberta, with good success. Our best model included satellite optical, radar, and topographic inputs, with the latter being a primary driver of predicted wetland location. When converted to wetland vs. non-wetland maps, all of our models produced high classification accuracies.

Our collaborators, the ABMI, then took our successful approach and have expanded it. They now offer, for any member of the public for free, a map of wetland occurrence across Alberta’s boreal region through their website (www.abmi.ca). You’ll notice, too, that along with this they also offer other wetland products that have since been built upon this classification, including maps of boreal peatland and fen. Pretty cool, huh?

Map of boreal wetland probability of occurrence in northern Alberta produced by the ABMI.

To get the scientific version of what we did, you can find our publication using the following citation. To access the ABMI’s wetland information products, visit: http://abmi.ca/home/data-analytics/da-top/da-product-overview/GIS-Land-Surface.html

Hird, J., DeLancey, E., McDermid, G., & Kariyeva, J. (2017). Google Earth Engine, Open-Access Satellite Data, and Machine Learning in Support of Large-Area Probabilistic Wetland Mapping. Remote Sensing, 9(12), 1315. https://doi.org/10.3390/rs9121315

Recent Posts
Featured Posts
Follow Us
  • Facebook Basic Square
  • Twitter Basic Square
  • Google+ Basic Square

Applied Geospatial Research Group