If a Tree Falls in a Forest and No one is Around to Map It…

I digitized downed trees in Northwoods using the orthorectified, georeferenced aerial photographs hosted by RIGIS. I used imagery collected in late 2018 and early 2019 because it was the latest leaf-off imagery available. Using leaf-off imagery was necessary in this project because downed trees lie on the canopy floor, and would be obstructed from sensor view if there were leaves present. The imagery has a 3 inch spatial resolution, which was adequate for the detection of most larger downed trees.

I defined the Northwoods as being bounded by Old North, Flagg, Plains, and Stony Fort Roads. I did not include the several houses/lots that are contained within this area.

I created my trails line shapefile using a georeferenced screenshot from  AllTrails.com , one of the only places I could find a complete Northwoods trail map. I intend to publish the trail map on ArcGIS online so that others can use the trail map line layer in future studies.

There is a significant amount of downed tree material in Northwoods. To maintain consistency in my digitizing, I created a 100 m fishnet grid over the woods to more easily track my progress and not miss any areas. I also set up the following parameters for what I considered to be a “downed tree”:

• Clearly not in the upright position (e.g. either held up at an angle by living trees or on the ground)
• Approximately 0.5 m in diameter
• At least 5 m in length

An example of trees that met these qualifications can be seen in the swipe adjacent.

In total, I mapped almost 1,800 downed trees in Northwoods. This was significantly more than expected. I estimate that about 10-15% of these marked trees were false postives (not actually a dead tree) or did not meet my criteria.

My resultant heat map shows no strong correlation with trails or roads, thought there is a significant hotspot near the drainage pond to the west. This area had nearly 80 downed trees in two of my sample grids (200 m^2 or 2.5 acres). I assume this disporportionately large number of downed trees to be the result of human activity, perhaps connected with the nearby solar farm. There is also a milder lengthwise hotspot along Flagg Rd and around the water tower.

I wanted to further explore the groupings in the downed tree distribution so I used the Point Clustering tool to identify major groups. The tool used the HDBSCAN (self-adjusting) method when assigning groups, which relies on group size (I used 20 points) instead of a defined distance.

The tool found eight groups that were significant given the parameters. The colours aren't significant, but are intended to separate groups visually. Grey points indicate ungrouped downed tree points (i.e. noise). Group extents may suggest areas that were selectively disturbed and may be important for future management decisions.

I selected for downed tree points within 50 m of trails and roads because these trees would likely be more important to identify for management decisions about Northwoods than downed trees in the interior forest.

I found there were 487 (of 1787 total) downed trees within 50 m of trails and roads in Northwoods. 27% is actually a high proportion when considering how much interior forest exists. From a visual inspection, there are more downed trees near northern trails than southern trails.

I also made a distance raster from the trails and roads layers I had created and extracted the distance values of all downed trees. I found that the distribution of distances was slightly left skewed (e.g. there were more downed trees near trails than far away) and had an average distance value of 101 m.

Some areas of the woods were ideal for manual digitization; these areas had neutral lighting and contained a moderate amount of deciduous leaf-off trees and few (if any) coniferous trees.

In some parts of the woods, long shadows led to a contrast-filled scene, which made it difficult to distinguish between shadows and downed (or standing) trees. This suggests that lighting (e.g. cloud cover) in imagery is very important in manual digitization.

Another difficult area for digitizing downed trees contained a relatively high proportion of coniferous trees, which obscured the ground even in Winter. In these areas, I had to make assumptions about a downed tree’s length, location, and orientation. Manual digitization carries inherent bias. I believe that my final dataset of downed tree points suffered primarily from false positives rather than false negatives.