Use of Remote Sensing in

Forestry Applications
Murat Tunç
110020209
Use of Remote Sensing in
Forestry Applications
There are many forestry applications that
remote sensing can be used for. Some of
these applications include terrain analysis,
forest management, recultivation, updating
of existing forest inventories, forest cover
type discrimination, the delineation of
burned areas, and mapping of cleared areas.
Unsupervised Classification
An unsupervised classification can be performed without any ground reference information.
This was evaluated, while the ground reference data was being collected, by having a general
sense of the area supplemented with DOQQs. Each unsupervised classification was created
through Erdas Imagine using the isodata algorithm. The unsupervised classification was first
seperated into 10 classes and an analyst would group the classes into hardwood, softwood,
mixed forest (when applicable), open field and water.
Examples of Unsupervised
Application
Image Type:Landsat ETM+
Path/Row:14/29 (subset)
Aquisition Date:July 21, 1999
Location:Adirondacks, New York
Unsupervised Classification defining
12 original clusters:
Overall accuracy is relatively low at 54%,
as mixed is often falsely classified in the
hardwood, softwood and open classes.
Hardwood is also often falsely classified in
as mixed further reducing overall accuracy
as there are limited blocks of solid
hardwood. This accuracy reflects relatively
simple analysis on the part of an analyst.
Accuracy can be increased with more effort
from a skilled analyst.
Unsupervised Classification defining 26
classes:
Increasing the number of classes defined
in the unsupervised classification decrease
the number of signatures that have to be
forced into an individual category as the
spectral range for each class increases. The
overall accuracy increase between the 12
and 26 class is due to the reduction in
discrepancies between mixed and softwood
and hardwood. In the unsupervised 26 class
classified image the salt and pepper effect
for mixed is greatly reduced, illustrating
larger blocks of solid hardwood.
Supervised Classification
Each supervised classification used the ground reference plots for training data for the
hardwood, softwood and mixed classes. For water and open field classes, DOQQ and other
higher resolution imagery was used for training. Erdas Imagine was used to create each
supervised classification using the maximum likelihood parametric rule.
Examples of Supervised Application
►Image Type:Landsat ETM+
►Path/Row:14/29 (subset)
►Aquisition Date:July 21, 1999
►Location:Adirondacks, New York
Supervised Classification with mixed
forest class:
Confusion occurs during classification
between the mixed class and the softwood
and hardwood classes as almost all
references have some degree of mixed
wood. Some of the open areas (fields and
shrub areas) carry the same spectral
response as hardwood, decreasing accuracy
Supervised Classification without mixed
forest class:
Overall accuracy increased by 11% with
the removal of mixed (varying
percentages of softwood and hardwood)
falsely being classified by the dominate
forest type
Buffering the ground reference data to a 45 meter radius, roughly equaling a
3X3 pixel group on a 30 meter resolution image, increasing the reference data
used for classification:
Supervised Classification with mixed forest class:
Buffering the ground reference samples to 45
meters increased each reference from 1 to 9
pixels increasing reference sample sized used
in classification. Overall accuracy increases
7% from the unbuffered mixed classification
as some of the single cells mixed classes were
removes from hardwood.
Supervised Classification without mixed forest class:
Overall accuracy was greatest in the
buffered no mixed wood classification.
Accuracy increased with the removal of
mixed, which was often confused with
hardwood and softwood.
Guided Clustering
Guided clustering is a hybrid of both unsupervised and supervised classification techniques. An
unsupervised classification was performed only on the ground reference plots that correspond
to each specific class. Therefore, for each class, an unsupervised classification was created. The
signature from the unsupervised classifications was used as training for the supervised
classification.
Examples of Guided
Clustering
Image Type:Landsat ETM+
Path/Row:15/30 (subset)
Aquisition Date:July 28, 1999
Location:Central, New York
Guided Clustering Classification with mixed forest class:
Some of the open class signatures used in
classification were similar to those of the
different forest classes, lowering the user's
accuracy for the open class. The mixed
forest type is spread throughout the hard
and softwood categories in a salt and
pepper effect, reducing the ability of the
classifier to delineate hard and softwood
classes. This is represented in the error
matrix as the fact that mixed is often
misclassified in both the hard and
softwood categories.
Guided Clustering Classification without mixed forest class:
Removing the mixed class
increased overall accuracy 14%
as the hard and softwood
categories were no longer being
classified in the mixed class.
There is still a discrepancy
between hardwood and softwood
which may be caused from
reference samples near the edge
of two forest types.
Guided Clustering Classification using a majority filter with mixed forest class:
Using a majority filter in the post
processing procedures greatly reduced
the salt and pepper effect of the mixed
class while smoothing the remaining
classes. Softwood was still often
missclassed in the mixed category as
the majority of the mixed surrounds
the softwood. Many of the softwood
reference plots are near the edges of
the softwood stands which may have
caused them to be misclassified under
mixed.
Guided Clustering Classification using a majority filter without mixed forest class:
Removing the mixed class and
performing a majority filter in the
post processing procedures resulted
in an overall accuracy of 88% the
best accuracy achieved using the
guided clustering method. The
removal of the mixed class and
smoothing of each class greatly
reduced the false classification that
may have been due to reference
samples near the edge of landcover
types or references in areas of
uncertainty (mixed).
Some Examples of Remote Sensing
in Forestry Applications
This is a color composite image of Queen
Charlotte Islands, British Columbia using
Landsat TM bands 5,4 and 3. (A) appears dark
green because this is a mature area of forest
with no cut, (B) appears bright pink because
this is area that has been clear cut or recently
logged and (C) appears bright green because
these are areas that have been clear cut but
have started regrowth.
Applications of Remote Sensing for
Forest Fires in Poland
► For several years the number of fires in European forests has increased. In the case of
Polish forests, a new category of forest fire risk has appeared in some regions. Forest
fires are a result of the simultaneous existence of at least three unfavourable
phenomena: long- term drought, the effect of air pollution (decline and decay of trees,
the formation of loose canopy and lush growth of grasses - all resulting in large amounts
of inflammable material) and high tourist presence in forests. Because of this situation,
the Polish Forest Service has started to apply new techniques and technologies for forest
protection: satellite data and Geographic Information Systems (GIS).
► High resolution satellite data such as Landsat Thematic Mapper, SPOT and ERS-SAR
combined with low resolution satellite images such as NOAA-AVHRR offer new
possibilities to monitor forest fires. They have a number of advantages over conventional
means of observation: Using satellite images, it is possible to view vast expanses of land
(thousands to tens of thousands of km2 on one image). This can be performed regularly
for the same area and recorded in different wavelengths, thus providing information on
the state of forest resources. Satellite data can be acquired without encountering
administrative restrictions
► GIS may be useful in immediately finding the most direct access to endangered areas
and in locating water sources
After locating the fire on the digital forest, the parameters and description of
endangered forest stands are displayed, which facilitates fire fighting
interventions
Burnt area seen on SPOT (black) and ERS-SAR (light green) images. Multitemporal
images (in this case ERS-SAR) allow monitoring of recultivation
Changes in Forest Land after
Hurricane Hugo
After classifying the before and after Landsat TM images into
thematic land cover consisting of Water, Forest, and Bare Soil,
differences between the classified images are calculated to arrive
at an assessment of change.

Red depicts areas that were forested and are bare soil; Blue
depicts water and wetlands; Green depicts forests not
significantly affected by the hurricane.

Conclusion:
► We can see that remote sensing in forestry is very
important. Because we can see the places that are
damaged by people, or the places that are cut from the
space; and you can get these images when the satellite is
passing above your region.
► You can understand where there is a fire, how big the fire
is or where the fire is going to without going to that place,
and you can take precautions immediately.
► For example after a hurricane or disasters, the damages on
forests can be understood in a very short time, and the
places where regrowth must be made can be seen from
above.
References:
► http://www.fao.org/sd/EIdirect/EIre0074.htm
► http://www.geog.ucl.ac.uk/%7Esmerino/
► http://forest.esf.edu/technicalAnalysis.html
► http://www.emporia.edu/earthsci/student/talk2/forestry1text.htm