Ministry of Forests and Range Remote Sensing and Geo-Spatial Applications

Harvest Mapping Currency

The remote sensing group is also involved in assessing the currency of forest harvesting mapping in British Columbia based on change detection techniques from Landsat satellite imagery. The overall objective of this work is to provide strategic level data for;

1. Cut blocks missing from VRI, and
2. Gap analysis data for timber supply

Figure 1

Methodology

The workflow for the harvest mapping project is as follows:

1. Analysis by forest district - Multi-temporal Landsat images are subsetted for each area of interest (district,TSA).
   
2. Two-date landsat imagery - Multi-spectral imagery from year two is used. Also to provide change information the SWIR spectral band (band 5) is used to calculate a ratio image between old band 5 and new band 5.
   
3. Creation of 'harvestable' mask from Vegetation Resource Inventory (VRI) - A simple query is used to subset the harvestable forest stands.
   
4. Image pre-processing and analysis - In this step, all imagery is registered to a common database along with the ratio image and these data are loaded into eCognition.
   
5. eCognition processing, segmentation and classification - This software is used to segment and classify the objects in the image.
   
6. ArcGIS editing - This step is used to remove and edit polygons that were generated in error.
   
7. Summary report by forest district - This summary report describes such metadata as; input images, date and any extra comments relevant to the procedure and analysis.
   

Pixel versus Object-Based Analysis

Tradition change detection studies use pixel based analysis since these were usually less computationally demanding and could be performed relatively simply. As processing time and power increased, software was written to group pixels together that have similar spatial and spectral characteristics into objects (segments). Figure 2 compares pixel versus objects.

Figure 2

Once these segments are created, statistics and other contextual information become available at the segment level. For multi-scale studies, segmentation is particularly useful since the objects can be generated over all possible spatial resolutions presented in the input image allowing the analyst to develop hierarchical associations between segmentation levels. This idea is presented in Figure 3 where multiple segments are used to refine the object edges and improve the actual perimeter of the harvested area.

Figure 3

Summary

This method provides a fast and inexpensive way to map harvest areas in a transferable and repeatable process. The final cutblock shapes are very representative in terms of shape and size and can provide gap information in a timely fashion anywhere in the province.

Contact: Ann Morrison