The conceptual model from Figure 1 within Objective 2 is expanded upon in order to accomodate the format in which the LiDAR data was delivered. The steps below outline the order in which the model was applied, as well as the purpose of each step.

1. Ground-point Filtering

The LiDAR data received from OMAFRA was not classified so that vegetation and buildings could be removed easily. Therefore, it was necessary for the LiDAR data to be filtered so that only ground-points were considered. The LidarGroundpointFilter tool within WhiteboxTools was used to remove off-terrain objects. The tool works by removing data points that exceed user defined thresholds relating to the slope between points. The appropriate thresholds are typically dependent upon the topography of the study area. For the portion of the Peterborough LiDAR dataset being considered, it was found that a 4.0-meter radius with a threshold of 15.0° removed the off-terrain points satisfactorily.

2. Nearest Neighbour Interpolation

The LiDAR dataset is interpolated using the nearest neighbour method in order to create a DEM with a rough surface, which should preserve the furrows and ridges within the DEM. WhiteboxTools’ LidarNearestNeighbourGridding is used to interpolate the LiDAR data which is stored in the LAS format. Since furrows range in width, a series of DEMs with decreasing grid cell sizes was interpolated. DEMs were interpolated at 1.0 metre, 0.75 metre, 0.50 metre, and 0.25 metre solutions. Since the LiDAR points being interpolated have already been filtered, all points were used during interpolation.

3. Mosaic DEMs

The Peterborough LiDAR dataset was received as 1 square kilometre tiles. Once the data were interpolated, the selected tiles were then mosaiced into a singular raster image that covers the extent of the study area. The images were mosaiced together using the Mosaic tool within WhiteboxTools. NearestNeighbour resampling was selected, although it was likely not needed given that tile images do not overlap in anyway.

4. Fill Missing Data Holes

Since the pointcloud that was interpolated was filtered, there were ‘holes’ of NoData within the DEM in places where no ground points exist, such as in rivers, ponds, and very dense forest canopies. Small data holes, such as those with dense forest cover, were filled using the FillMissingData tool from WhiteboxTools. Larger holes caused by ponds and building footprints remained in the image as it is not possible for them to contain tilled soil. Once the small NoData holes had been filled, the base DEM was complete.

5. DEM Smoothing

In order to isolate the surface roughness of the DEM, it was necessary to smooth the surface of the DEM. This was done using the FeaturePreservingDenoise tool within WhiteboxTools. The smoothing tool is based on a modified algorithm developed by Sun et al. (2007), and its benefit over other low-pass filters (such as a mean filter) is that feature edges are preserved, which increases the effectiveness of the model in the next step. Since multiple sets of data were being considered, it was important that the filter size maintain the same physical size throughout, as the surface features being smoothed were not changing in elevation. The filter size was set to 11.0 metres in diameter, which was adjusted accordingly for each dataset with a varying resolution.

6. Subtract

Once the smoothed raster had been created, the difference between it and the original DEM was taken using the Subtract tool within WhiteboxTools. The surface texture (containing the furrows) was now isolated and was the only data within the raster.

7. Anisotropic Identification

With the ridges and furrows successfully separated from the rest of the elevation data, the surface roughness raster image was processed through the MaxAnisotropyDev tool within WhiteboxTools. Similar to the conditions set by the FeaturePreservingDenoise tool, a radius in physical meters was be used throughout. The minimum radius used was be 3.0 metres (which is the smallest value that the tool can accept), while the maximum was be 9.0 metres, which works well with how the tool seperates data.

The furrows and ridges should have then been successfully identified.