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

Satellite Mountain Pine Beetle Attack Mapping

British Columbia (B.C.) is currently experiencing the largest recorded mountain pine beetle infestation ever recorded in North America. Provincial projections indicate that 80 per cent of the merchantable pine in central and southern B.C. could be killed by 2013, and well over half that pine could be dead by the summer of 2007. In order to quantify past, present and future infestation levels, we have been working on two projects to develop a multi-temporal monitoring system using Landsat TM and ETM+ to determine;

1. The year of death and,
2. Current year attack mapping.

The Year of Death

Knowledge of the year of death has important implications for, i) quantifying the shelf-life of standing dead pine on the land base and ii) identifying the existing stocking levels of non-attacked timber.

To identify the year of death in the forests of British Columbia, we use a landsat based multi-temporal change detection procedure as shown in Figure 1.

Figure 1

In this figure, the landsat image sequence for the current analysis is shown on the left-hand side (1999-2006). The images are georeferenced together and a Tasselled Cap Transformation is applied to each image in the sequence, independently. This transformation is used to reduce the dimensionality of the original image data while at the same time producing components that represent ground characteristics namely; Brightness, Greenness and Wetness. To highlight mountain pine beetle attack, a wetness difference image (EWDI) is generated between the years under investigation and the difference map is threshold based on user-defined values developed during the algorithm calibration phase. The thresholding is used to mask out other areas which are characterized by a change in wetness such as those from harvesting and fires. Figure 2 shows a typical Enhanced Wetness Difference Image (EWDI) and the histograms of various disturbances.

Figure 2

In this work, we are currently testing this methodology on two landsat scenes located in central British Columbia. These scenes are illustrated in Figure 3 below.

Figure 3

The results of this study show the valuable potential of multi-temporal landsat imagery for determining the year of death of these forest stands.

Figure 4 illustrates the theory behind the year of death determination. In this figure, we see the affected pine stands showing a much larger change in wetness when compared to a non-infected spruce stand.

Figure 4

Figure 5

Figure 5 shows the results centered on one map sheet. In this figure, the 1999 image is presented in the top left, 2004 image in the top right, and 2006 image in the bottom left. Cumulative pine beetle damage over all available years is shown in the bottom-right for pure pine stands.

Figure 6 shows the year of death results for mapsheet 93f.017. Figure 6a shows the 50cm ortho-photo from 2005 and 6b shows the vector file on top of this ortho-photo colour-coded by the year of death. Figure 6c shows the legend displaying the various colours for each year from 2001 to 2006.

Figure 6a	Figure 6c
Figure 6b

Current year attack mapping

The purpose of the current year attack mapping project is to develop an operational approach for mapping pine beetle attack every year and to provide a finer-scale supplement to the overview surveys which are currently being used. To achieve this, we apply the same methodology as illustrated in Figure 1 but for Time 1 (pre-outbreak) and Time 2 (post-outbreak) images only. In this case, the landsat imagery must be analyzed in pairs to detect changes and in order to reduce change variability and non pine beetle related changes, certain image masks must be used to stratify the change detection procedure. These include, water and harvest masks, cloud masks and pine component masks. The methodology is shown in Figure 7 .

Figure 7

Similar to the year of death project described above, we are testing the current year of attack methods in several sites across British Columbia. The most recent test site is located in central BC near Prince George. As shown in Figure 7, this approach highlights mountain pine beetle red attack very well. On the left side of this figure, the VRI pine polygons are displayed in green over a 1:20k orthorectified aerial photograph - where the red attack is clearly visible. In the center of the figure, the binary attack map (attack and no attack), constrained by VRI polygon is illustrated in red. For comparison purposes, the aerial overview survey is shown in on the right side of Figure 8 and Figure 9 .

Figure 8

Figure 9

Once the change maps have been generated they are subject to an accuracy assessment to evaluate the usefulness of such a map product for end users. In this case, the accuracy assessment was derived from random point samples in the pine and pine-leading forest stands as classified from the VRI. Orthorectified aerial photographs at a scale of 1:20k were used as reference (ground truth) data for the accuracy assessment. The results of this assessment are presented in Figure 10 and 11

Figure 10

Figure 11

To date, the majority of Timber Supply Areas (TSAs) in B.C. have been completed for 2005. Figure 12 presents the progress of the 2005 red attack mapping from Landsat (in purple) together with two sample raster products (pine 100 and pine-leading) for the Quesnel TSA.

Figure 12

For more information, contact Xiaoping Yuan and Chris Butson