Académique Documents
Professionnel Documents
Culture Documents
Page | 1
Methods
Data was collected from Landsat Thematic Mapper (TM) imagery from the years 1988,
1992, 1995, 2000, 2005, and 2011. Images were taken in the summer months, and the images
covered a total area of 2,378 KM2 in the area of northwestern Washington State.
David Wallins (2016) created Tasseled Cap brightness and greenness index values using
the work of Cohen et al. (1998). The image file created by Wallin (2016) accounted for ten
bands, with two indexes for each of the 5 time period images.
Using directions provided by Wallin (2016) I performed an ISODATA unsupervised
classification of the northwestern Washington scene. I classified clearcut areas by year for each
of the five time periods provided by the imagery. I classified the image into 6 different
Cole
Page | 2
information classes, using the file created by Wallin (2016) in color infrared. I then created an
elevation
vegetation
these results,
clearcuts less
to clump
Informatio
n Class
No Change
1988-1992
1992-1995
1995-2000
2000-2005
2005-2011
Colo
r
used a 6 pixel by 6 pixel smoothing window for this technique. I then compared this final land
classification image to the areas of existing land ownership in the same area. The resulting
Results
My final classification image after using the clumping and sieving techniques resulted in
a total harvested land area of 28,858 hectares. Judging by the results of the classification, the two
Cole
Page | 3
top time periods of timber harvest occurred between the years of 1988-1992 and 2005-2011
where 28% and 27% of the forested area, respectively, were harvested. From 1992-1995, there
was a drastic decrease in hectares of timber harvest of over 400%, compared to the rates of the
previous years from 1988-1992. 1992-1995 was the down year, however, as the timber harvest
amount steadily rose in the subsequent years. The trends in timber harvesting can be seen in
Table 1. Over the course of the 23 years analyzed by the study, a total of 28,858.06 hectares of
timber had been harvested from the total forest land cover, or a total of 28.98% of the total forest
land cover of the study area.
TABLE 1. TIMBER
Time
Period
No
Hectares
208,966.
# of Years
Harvest
Rate
(ha/yr)
Total
Harvest
(% forest)
Harvest
Rate (%
forest/yr)
Cole
Page | 4
Change
88-'92
92-'95
95-'00
00-'05
05-'11
Total
94
7,849.13
1,882.88
5,399.13
6,036.56
7,690.38
28,858.
06
4
3
5
5
6
1,962.28
627.63
1,079.83
1,207.31
1,281.73
7.88%
1.89%
5.42%
6.06%
7.72%
1.97%
0.63%
1.08%
1.21%
1.29%
28.98%
After finding the overall harvesting statistics for the study area, I then cross-referenced
the found data with the existing land ownership of the study area within 4 different categories:
private, National Forest, DNR, and wilderness. The results showed that private lands were
harvested almost twice as much as National Forest and DNR lands, and over 4 times as much as
Wilderness lands, with almost 900 thousand hectares harvested over the 23 year span. National
Forest and DNR lands both had very similar overall harvest rates, with a total harvest rate of
almost 600 thousand hectares harvested over the 23 year span. The 4 years between 1988 and
1992 saw the highest timber harvest rates for all ownership areas. The rates are kept relatively
low the subsequent years until the time span from 2005-2011 where there was a prominent
increase in timber harvest rates. See Table 2.
TABLE 2. TIMBER HARVEST RATES IN HECTARES PER YEAR FOR OWNERSHIP AREA AND
PERCENT OF TOTAL OWNERSHIP AREA PER YEAR.
Time
Perio
d
Yea
rs
88-'92
92-'95
95-'00
00-'05
05-'11
Discussion
Wilderness
%/yr
ha/yr.
.
0.24
104.59
%
0.00
0.00
%
0.04
13.13
%
0.02
7.84
%
0.96
284.59
%
National
Forest
%/yr
ha/yr.
.
0.22
331.11
%
0.03
59.00
%
0.01
11.08
%
0.03
37.16
%
0.37
364.90
%
Private
%/yr
ha/yr.
.
1,019.9 0.47
4
%
0.14
404.25
%
0.40
689.88
%
0.37
648.03
%
0.27
395.53
%
DNR
ha/yr.
506.39
164.38
365.74
514.29
236.71
%/yr
.
0.35
%
0.09
%
0.32
%
0.45
%
0.25
%
Cole
Page | 5
The majority of harvest in the study area is from privately owned lands. This comes as no
surprise as these lands are mostly controlled and maintained by private logging companies for
timber production. Again unsurprisingly, DNR lands constitute the second most harvested lands
as the timber in DNR controlled areas are harvested to fund public services. Wilderness areas are
the least harvested most likely due to the fact that it constitutes the least amount of land, and it is
less harvested due to these areas being highly restricted by government. The most interesting
result found was the high volume of National Forest land shown to be logged over the 23 year
time span. I would have expected to see the amount of National Forest land to be much less
impacted due to the high government regulation of National Forest areas. However, these lands
were the second most logged areas in terms of overall timber harvest. The most likely reason for
this finding is due to error in the classification portion of the study. Areas could have been
misclassified or incorrectly classified as a different time period clearcut which would then lead to
over- or underrepresented harvest areas in ownership lands.
This particular scenario could be the case for most of the inaccuracies of this study.
Although unsupervised classification is in many ways an objective classification method, it is
still open to subjectivity in that the conductor of the classification has to make the decision on
which areas of the image to classify into a certain information class. There are many different
methods to remedy this subjectivity, but the chances of having more than 90 or 95 percent
accuracy when originally classifying the image is slim. Especially with having to classify up to
50 different classes, there is quite a bit of room for misinterpretation.
Another potential area of misinterpretation is in the unclassified images. With the
brightness and greenness indexes on which I based my classifications on, it is difficult to
Cole
Page | 6
interpret which areas are harvested using burning techniques and which areas were disturbed
from natural causes, as these areas show up with the same brightness values.
Another problem that could have affected the overall accuracy of the study was the
masking area applied to the ownership image. This mask was applied while having an overall
classification accuracy of 61.5%. This mask application is acceptable, although it was used with
a misclassification of almost 40%. This misclassification in turn could have affected the
classification of the resulting ownership areas, resulting in lower overall classification accuracy.
The same problems could be found in the sieve and clumping techniques used, as there is some
subjectivity in which thresholds for the techniques should be used.
Overall I believe this study was at least marginally successful at quantifying the timber
harvest rates in the 23 year time span of the imagery of the study area. The trends in ownership
areas seem to show a trend, but the sudden spike in timber harvest leads me to believe there
could be some refining that needs to be done in the initial information class classification and
masking. This study could also benefit greatly from the application of better and more thorough
ground truth data.
Bibliography
Cohen, W. B., M. Florella, J. Gray, E. Helmer, & K. Anderson. 1998. An Efficient and Accurate
Method for Mapping Forest Clearcuts in the Pacific Northwest Using Landsat Imagery.
Photogrammetric Engineering & Remote Sensing, 64(4), 293-300.
Quinn, D. 2015. Unsupervised Classification of Whatcom County Using the ISODATA
Clustering Algorithm.
Cole
Page | 7
Wallin, D. 2015. Lab V: Change Detection Using Unsupervised Classification with ENVI.
Retrieved from http://faculty.wwu.edu/wallin/envr442/ENVI/442_change_lab_ENVI.htm