Current Research
Article on the Automated Landform Mapping Project (2019) (click to zoom)
For the past two years, I have been working with Dr. Jerome Lesemann and graduate students on the Automated Landform Mapping (ALM) project. This project is funded by Natural Resources Canada. The basic idea behind this project is to automatically identify landforms in the Arctic. We are researching methods to automatically identify landforms from digital elevation models (DEMs), so that a national inventory can be created.
Canada is a vast, sparsely occupied country. The majority of the population live within 100 km of our southern border and there is virtually no agriculture north of 56ยบ N (the Peace River region of Alberta). The area to the north of this, including the 3 territories (Yukon, Northwest, and Nunavut) and the northern parts of all provinces except for Ontario and the maritime provinces are virtually unoccupied except for concentrations around the cities and towns.
Although First Nations have occupied this country for millennia, their populations were also quite low and relatively concentrated in areas of resource abundance. Although there have been resource surveys and aerial mapping performed, there are large areas where very few people have ever set foot.
Many people from more populated areas of the planet would be surprised to find that not only are many features are unnamed, but that we have limited knowledge of the resources of the North. There simply isn’t the long-term knowledge of local areas around settlements that local residents have acquired in other parts of the world, because there aren’t many settlements.
If you Google the phrase “how many mountains are there in Canada,” you will see the apparently authoritative answer that there are 21,324. If you look closer, you will see that this number represents the number of named mountains, but ignores thousands of unnamed ones. Mountains receive a lot of attention because they are prominent and represent a barrier to transportation, so they are relatively well known. Googling the count for less important landforms such as lakes, marshes, or tombolos provides no convenient answers; we simply haven’t compiled a comprehensive inventory of the landforms of Canada.
One of the issues is that of ontology. What exactly is a mountain? How high must it be? How distant must one peak be from another before it is considered to be a separate mountain? How much must the elevation decrease before nearby peaks are considered to be separate? For all of human history, we have applied informal criteria to identify landforms. This has resulted in a hodgepodge of named and unnamed peaks which are highly dependent on local context.
Canada has a strong tradition in Geoscience and Geomatics. In fact, the first vector Geographic Information System (GIS) was developed in Canada. Developments in these technologies have been spurred because of the huge size of the country and the difficulty in managing it. Whereas the topographic mapping of Canada identified the major features and the lay of the land, relatively little information was provided about smaller features and what those features were.
We stand at the cusp of a new era in the mapping. The combination of Convolutional Neural Networks (CNNs) and extensive, detailed Digital Elevation Models (DEMs) of large parts of the country now mean that Remote Predictive Mapping (RPM) can soon be used to identify different landforms and to provide some basic characteristics. We can sidestep issues of ontology and use qualified geomorphologists to train computers to identify landforms. Because CNNs are flexible enough to tolerate ambiguous definitions, they can handle the fact that Mount Carleton, the highest point in New Brunswick (817 m), is less than the 1000 m high unnamed peak only 9 km west of my house in Nanaimo.
Our research is currently centred on eskers, which are sinuous ridges of sand and gravel which are formed within glaciers as meltwater makes its way out of the glacier. These are similar to riverbeds, however, since the water and the glacier are now gone, they are exposed for our inspection, and this makes their use for valuable for mineral exploration.
Draped Sentinel 2A image of preliminary esker classification results showing some areas were classified successfully and others nearby that were missed (red=areas classified).
As we develop and refine our procedures, we hope to be able to add other types of landforms and build upon this framework. One day, we will be able to know not only what the mountain names are, but also their height, outline, volume and basic composition are. Once we have this information, we will be able to set our criteria, and then come up with a better answer about how many mountains there really are in Canada.