Monitoring Protected Areas Using Remote Sensing Technology

Zahra Ghofrani

, Kali Prasad Nepal

, and Adham Beykikhoshk

School of Engineering, Faculty of Science, Engineering and Built Environment, Deakin University, Geelong, Australia

Centre for Pattern Recognition and Data Analytics, Deakin University, Geelong, Australia

ghofrani@deakin.edu.au, kali.nepal@deakin.edu.au, abeyki@deakin.edu.au

Keywords: change detection, remote sensing, protected areas

Abstract: Due to irrational use of natural resources, human society is facing unprecedented threats. Remote sensing is

one of the essential tools to determine changes in various forms of biological diversity over time. There are

many methods to determine changes in protected areas, using satellite images. In this paper after introducing

different change detection methods and their advantages and disadvantages, a hybrid method is used to

analyse changes in forests and protected areas in a national park. Two Landsat images of Golestan National

Park in Iran (taken in 1998 and 2010) were used. This hybrid approach combines Change Vector Analysis

(CVA) for flagging the occurrence of changes, followed by signature extension to assign labels to changed

pixels. The main objective of this paper is to propose a method for discovering and assessing environmental

threats to natural treasures.

1 INTRODUCTION

Selecting the most appropriate change detection

method for a given application is difficult, and

requires consideration of the change type of interest

(Fraser, Olthof, and Pouliot, 2009). Wide range of

change detection algorithms are now available which

may be broadly grouped as classification methods

(Chen and Chen, 2012), (Hermitte, Verbesselt,

Verstraeten, and Coppin, 2011) and spectral

approaches (Fraser et al., 2009). If sources of image

noise are adequately controlled, spectral approaches

quantify the magnitude of reflectance changes

between different dates, which relate to a land surface

change.

One advantage is the potential to fine-tune change

detection sensitivity, while a limitation is the inability

to provide information on the nature of change e.g.

class label (Xiaolu and Bo, 2011). Examples of

spectral-based methods include: image differencing,

regression and change vector analysis (Fraser, Li, and

Cihlar, 2000), (Johnson and Kasischke, 1998),

(Prakash and Gupta, 1998), (Fraser, Olthof, and

Pouliot, 2009).

Classification approaches such as post-

classification comparison and two-date image

clustering, in contrast identify both the occurrence of

changed pixels and the type of change by directly

labelling land cover at two time periods. However,

they are susceptible to generating high levels of

commission error due to the multiplication of

individual errors (Yuan, Sawaya, Loeffelholz, and

Bauer, 2005), (Fraser, Olthof, and Pouliot, 2009).

There are also hybrid change detection procedures

that exploit the advantages of each approach, while

attempting to minimize their limitations (Luque,

2000), (Petit, Scudder, and Lambin, 2001),

(Silapaswan, Verbyla, and McGuire, 2001).

This paper presents a hybrid change detection

algorithm. In this approach, a mask of potential

changed pixels is first created by thresholding a two-

date change vector analysis (CVA) product. Land

cover class is then updated for changed areas only by

spectral signature extension, whereby changed pixels

are matched to the most similar labelled cluster from

a baseline land cover map.

This method exploits the benefits of both spectral

and classification type methods, and reduces their

weaknesses (Fraser et al., 2009). Thus, the accuracy

of this hybrid method is expected to be higher than

each method individually. It is also focused on

decreasing the role of human operators in the process.

This method extracts image data better than the others

and also enables labelling to be done automatically

using post classification comparison and pre-existing

knowledge of the land cover data.

This paper is organized as follows: Section 2

describes the case study area and the required data for

107

Ghofrani Z., Nepal K. and Beykikhoshk A.

Monitoring Protected Areas Using Remote Sensing Technology.

DOI: 10.5220/0005422001070113

In Proceedings of the Third International Conference on Telecommunications and Remote Sensing (ICTRS 2014), pages 107-113

ISBN: 978-989-758-033-8

analysis. Section 3 represents the proposed change

detection algorithm. Section 4 describes the

environmental analysis. Finally, in section 5, our

conclusions are drawn.

2 CASE STUDY AND REQUIRED

DATA

Golestan National Park with 92,000 hectares area is

the biggest and oldest registered national park in Iran.

This forest was registered with the UNESCO World

Heritage List in 1976 as one of the 50 vital protected

areas on earth. The geographical area of Golestan

National Park is between 55° 43’ 16” to 56° 15’ 31”

longitudes and 37° 16’ 51” to 37° 32’ 27” latitudes.

The average elevation of this park is 1378 meters.

Different data types used in this study are

introduced below.

2.1 Topographic Map

The only map available for the study area is a

1:250,000 topographic map produced in the spring of

1998 by a group of forestry research organizations

using Landsat ETM+ images (Figure 1).

Figure 1: The topographic map of the study area in 1998

2.2 Landsat Images

For this research, two Landsat images taken in August

1998 and 2010 (a period of 12 years, which is an

appropriate period for assessing environmental

changes) were used. The radiometric and geometric

calibration parameters of these images are available

and cloud cover over the area in the images is

negligible. Image dimensions are 8091 × 7231 pixels

and the field of view is about 185 × 175 km (Figure

2). For Landsat TM images, the UTM system and

WGS84 ellipsoid were used for geo-referencing.

2.3 High Resolution Images

Since there was no updated map for the case study

area, in order to evaluate the accuracy of the method,

we used Geoeye high-resolution images for 2010.

The mosaic Geoeye images have been cut to the

thresholds of Landsat images’ latitude and longitude

(Figure 3).

Third International Conference on Telecommunications and Remote Sensing

108

(a) (b)

Figure 2: Landsat images (false colour composites), of the study area (a) August 1998 and (b) August 2010.

3 CHANGE DETECTION

ALGORITHM

This method consists of four main steps. Pre-

processing level as the first step, includes obtaining

images and reference maps, image registration and

normalization. The purpose of this step is to prepare

the images for the next step.

In the second step, thematic map is produced from

satellite images and available maps of the area, by

clustering an image as the baseline (master) image

and labelling the clusters based on the reference

maps. Then based on the post- classification

comparison method, the changes’ nature are labelled.

Post-classification comparison applies a comparison

between the feature vector of each changed pixel and

the centre of the labelled clusters. Based on this

comparison the changed pixels will be labelled

(change map).

Finally in the last step, a reference map is

produced using the high-resolution Geoeye image

which is needed for accuracy assessment.

Figure 3: Mosaic Geoeye image of the study area, 2010.

3.1 Pre-processing

Image pre-processing stage, includes both the

geometric correction and normalization of images

that have been taken at different times from the same

area. Pre-processing consists of the following

operations.

3.1.1 Geometric Correction

In this study, the geo-referencing of the 1998 image

is performed using the 1:250,000 topographic maps.

After geo-referencing of 1998 image, the 2010 image

is registered to 1998 image. Sub-pixel accuracy for

image registration is obtained.

3.1.2 Image Normalization

After a careful image-to-image spatial registration the

images must be radiometrically normalized. Accurate

normalization is essential for the combined CVA and

post classification comparison change detection

approach, since both methods assume that a pixel's

reflectance is stable through time unless a land cover

change occurs (Fraser, Olthof and Pouliot, 2009).

Histogram Matching is a common technique for this

reason which uses the histograms for image

processing and colour adjustment between images.

3.2 Thematic Map Generation

The change detection procedure requires a baseline

land cover classification from which changes are

detected at nominal 12-year intervals. The thematic

map is produced using an unsupervised clustering

approach that combines features of the Enhancement

Classification (ECM) and Classification by

Progressive Generalization (CPG) methods.

Monitoring Protected Areas using Remote Sensing Technology

109

The enhanced imagery is clustered to a number of

spectral clusters. Visual quality checking is an

important part of this and each subsequent

generalization step, and is performed by comparing

the previous generalization with the current one to

ensure that no significant land cover information is

lost. The overall coverage of the Earth's classes such

as soil, water, plant and etc., are determined.

Generalization proceeds by progressively merging

spectrally similar and spatially adjacent clusters to

generate conceptual classes. Final cluster merging

and labeling to a land cover classification is based on

expert image interpretation and available reference

data (Fraser, Olthof, and Pouliot, 2009).

In this research the 1998 Landsat image is divided

into 22 clusters using Iso-data clustering method. The

number of clusters is chosen to be twice as the number

of conceptual classes which are 11 in the case of this

study area. The output for this level is a clustered

image (22 clusters), which is labelled based on

1:250,000 topographic map and converted to 11

conceptual classes.

3.3 Change Detection Process

In this section, the sub-steps of change detection

process will be explained in the following stages.

3.3.1 Tasseled Cap Transformation

Tasseled Cap transformation is a well-known

methods of enhancing spectral information content

for Landsat TM data. Tasseled Cap transformation

especially optimizes data viewing for vegetation

studies. Tasseled Cap index was calculated from data

of the related six TM bands (King and O'Hara, 2001).

Three of the six tasseled cap transform bands are

often used:

 Band 1, brightness as a measure of soil

 Band 2, greenness as a measure of vegetation

 Band 3, wetness as interrelationship of soil and

canopy moisture

This transformation is used to calculate brightness

and greenness of both images (1998 and 2010), which

are the input for CVA analysis.

3.3.2 Applying Change Vector Analysis

A change vector can be described by an angle of

change (vector direction) and a magnitude of change

from date 1 to date 2 (Fraser et al., 2009), (Chen,

Gong, He, and Shi, 2003). We used brightness and

greenness as inputs of CVA to measure and monitor

reforestation and deforestation of the region of study.

The bands are observed in measurement space with

brightness placed along the X-axis and greenness

placed along the Y-axis.

Change direction is achieved by measuring the

angle between corresponding pixels in different times

(1998 - 2010) and the magnitude of change is

achieved using Euclidean distance between vectors.

Magnitude of change vector and its direction are

described by Eq. (1) and Eq. (2) respectively.

 









 















 









(1)

 

















(2)

G1, G2, B1, and B2 are values of greenness and

brightness in two images, which are obtained from

Tasseled cap transformation. To specify the

reforestation and deforestation of the jungles,

greenness and brightness values should be compared

(Kuzera, 2005). Angles measured between 90 and

180 degrees, show reduction in brightness and

increase in greenness, this change is considered as

reforestation. Angles measured between 270 and 360

degrees, show reduction in greenness and increase in

brightness, this change is considered as deforestation

(Kuzera, 2005). Angles measured from 0 to 90 and 90

to 180 degrees, show reduction or increase for both

greenness and brightness, respectively. This is known

as a stable condition, indicating no change in the

vegetation of the area (Kuzera, 2005).

According to the magnitude of change vectors,

damaged pixels are categorized into 4 levels of low,

moderate, severe and very severe deforestation. For

this reason four equal intervals are applied as below:

 Interval [1-100]: Low change

 Interval [100-200]: Moderate change

 Interval [200-300]: Severe change

 Interval [300-400]: Very severe

Values less than 1 are considered as noise and values

higher than 400 as outlier. The thresholds defined are

quite tentative. In Figure 4, the various degrees of

grayscale represent different degrees of degradation,

the darker shades show more severe deforestation and

vice versa.

3.3.3 Post Classification Comparison

Change labeling is accomplished by iteratively

updating land cover starting from the baseline

classification for only those pixels identified as

Third International Conference on Telecommunications and Remote Sensing

110

changed in the CVA change mask. Post-classification

comparison method involves comparison of the new

feature vector of each changed pixel with the major

feature vector of the cluster centres (determined in the

first image). New classes of land cover pixels are

determined by assigning a pixel to the most similar

cluster and corresponding existing land cover maps,

so the new cluster of changed pixels are achieved.

Figure 4: Result of applying the CVA method to two

Landsat images of the study area, for 1998 and 2010.

Figure 4: Result of applying the CVA method to two

Landsat images of the study area, for 1998 and 2010

To understand the change trends, the feature vector of

each changed pixels of 2010 image, is compared with

feature vector of classification cluster centres of 1998

image. The changed pixel is assigned to the

cluster with the most similarity with cluster centre.

Since in 1998 image, each cluster has a distinct

relation with an information class, finding the most

similar cluster is the same as labelling changed pixels

in 2010 image with a new information class.

This model is used for cost-effective classification in

large and remote areas and regions where it is difficult

to collect data. The main benefit of this approach is

that by using post-classification comparison method

and a strong knowledge of land cover data the

labelling process will be done automatically.

3.4 Accuracy Evaluation

Since there is no updated reference map available for

the area, in order to evaluate the accuracy of the

obtained change detection results, a reference map is

produced by using both the Landsat (2010) and a

high-resolution Geoeye image. Geoeye image is

taken at the same time as the Landsat image (2010)

and covers the whole Golestan national park. It is

later cropped so it covers the same latitudes and

longitudes that Landsat image covers.

To produce the reference map, Landsat 2010

image is first clustered into 33 clusters. The obtained

clusters are compared to the information classes

recognized from the high-resolution Geoeye image.

In this way the correspondence between clusters and

conceptual classes are determined which leads to the

classification of 2010 Landsat image. This image is

used as the reference map to evaluate the change

detection results. We gained 85% accuracy for the

proposed change detection method.

Figure 5: Changes of classes in terms of pixels

50 000

100 000

150 000

200 000

250 000

300 000

350 000

400 000

450 000

1→8 2→8 2→9 2→10 3→4 3→7 6→10 8→10 10→8 10→9

Number of pixels

Changes of classes in terms of pixels

1: Dense grassland

2: Planted forest

3: Dense forest

4: Road

5: Non- dense forest

6: Semi-dense forest

7: Shrubbery

8: Irrigated agriculture and plantations

9: Residential

10: Non-dense grassland

11: Water

Monitoring Protected Areas using Remote Sensing Technology

111

4 ENVIRONMENTAL ANALYSIS

Figure 5 shows the classes which have the most

change rate and magnitude of their change in terms of

pixels. As the statistics show from 1998 to 2010 the

number of pixels which converted from planted

forest, semi-dense forest, dense forest, and dense

grassland classes to road, residential, low density

grassland, irrigated agriculture, plantations,

shrubbery and non-dense forest classes is very high

and this represents a serious degradation in this area.

Road construction in forests regardless of its negative

effects on the forest, inappropriate urban

development, human progression in nature, cutting

trees for fuel, human farming in the forests to provide

food supply, and etc. are some main reasons for

degradation in this area.

A similar research was developed and

demonstrated by Fraser using six national parks in

Canada. It covered a range of geographical and

ecological conditions and was subject to a variety of

change agents including forest harvesting, wildfire,

land use development, and climate/weather (Fraser,

Olthof and Pouliot, 2009). In contrast to Golestan

National Park area that is located on one Landsat

scene and there is no need to mosaic Landsat images,

the area of Fraser’s study was vast and required more

than one Landsat frame to provide complete

coverage. They used 30m resolution Landsat EM and

ETM+, from 1990 to 2005 to generate baseline land

cover classification at five years intervals. Due to

huge height difference, removing haze and

topographic effects for Canada’s national parks was

necessary. However in Golestan National Park,

topographical elevations are fairly smooth and there

was no need to apply topological corrections in pre-

processing. Moreover, radiometric normalization in

Canada national parks was done by using filtering,

while for Golestan National Park it was done by

histogram matching. In both methods, identifying the

changed pixels and labelling them, were determined

using CVA and signature extension. Finally in

Canada baseline land cover and changes were

validated by updated available maps and in Golestan

National Park by high-resolution Geoeye images (due

to the lack of updated maps). Fraser reported 92%

correctly identified changed pixels and 8% omission

error rate in Canada’s parks.

5 CONCLUSION

Timely and accurate change detection of Earth’s

surface features is extremely important for

understanding relationships and interactions between

human and natural phenomena in order to promote

better decision making. Remote sensing data are

primary sources extensively used for change

detection in recent decades and many change

detection techniques have been developed based on

them. The common goal of all these algorithms is to

improve the accuracy of the information extracted

from remote sensing images. In this paper, a change

detection method was proposed to determine changes

in the forests of Northern Iran (Golestan National

Park). Using the combination of spectral and

classification methods lead to an acceptable accuracy.

In comparison with the conducted research on

national parks of Canada, lack of updated reference

maps, has a direct impact on the final accuracy. The

results of the assessment indicated that change

detection method should be developed based on local

knowledge. While this method provides a set of

generic procedures and tools for change detection, its

successful application requires an analyst

experienced in land cover interpretation and image

processing. In particular, the baseline land cover

labeling, assessing results from the image correction

methods, determining a CVA change threshold, and

development of signature extension rules, are

subjective and will determine the final accuracy of the

land cover change products. This algorithm is a cost-

effective change detection method in large areas and

tries to minimize the role of the human operator. It

can be implemented for most forests regardless of

their vegetation. This study is intended to explorer use

of high resolution images in the future in order to

investigate its capabilities to determine the change of

plant species. In future this method also can be

elevated using optimisation methods to find the best

values for CVA thresholds, number of clusters, and

similarity measure and result in an extended

intelligent version of current change detection

method.

REFERENCES

Chen, J., Gong, P., He, C., Shi, R. & P. (2003). Land-

Use/Land-Cover Change Detection Using Improved

Change-Vector Analysis.

Chen, X., Chen, J. (2012). An automated approach for

updating land cover maps based on integrated change

Third International Conference on Telecommunications and Remote Sensing

112

detection and classification methods. ISPRS Journal

of Photogrammetry and Remote Sensing, 71, 86-95.

Fraser, R. H., Li, Z., Cihlar, J. (2000). Hotspot and NDVI

differencing synergy (HANDS): A new technique for

burned area mapping over boreal forest. Remote

Sensing of Environment, 74, 362-376.

Fraser, R. H., Olthof, I., Pouliot, D. (2009). Monitoring land

cover change and ecological integrity in Canada’s

national parks. Remote Sensing of Environment, 113,

1397-1409.

Hermitte, S. L., Verbesselt, J., Verstraeten, W. W., Coppin,

P. (2011). A comparison of time series similarity

measures for classification and change detection of

ecosystem dynamics. Remote Sensing of Environment,

115 (12), 3129-3152.

Johnson, R. D., Kasischke, E. S. (1998). Change vector

analysis: A technique for the multi spectral monitoring

of land cover and condition. International Journal of

Remote Sensing, 19, 411-426.

King, R., O'Hara, C. (2001). Tasseled Cap Transformation.

Computational Geospatial Technologies Center,

Mississippi State University.

Kuzera, K. (2005). Monitoring vegetation regeneration and

deforestation using change vector analysis: Mount St.

Helen's study area.

Luque, S. S. (2000). Evaluating temporal changes using

Multi-Spectral Scanner and Thematic Mapper data on

the landscape of a natural reserve: The New Jersey Pine

Barrens, a case study. International Journal of Remote

Sensing, 21, 2589-2611.

Petit, C., Scudder, T., Lambin, E. (2001). Quantifying

processes of land-cover change by remote sensing:

Resettlement and rapid land-cover changes in south-

eastern Zambia. International Journal of Remote

Sensing, 22, 3435-3456.

Prakash, A., Gupta, R. P. (1998). Land-use mapping and

change detection in a coal mining area—a case study in

the Jharia coalﬁeld, India. International Journal of

Remote Sensing, 19, 391-410.

Silapaswan, C., Verbyla, D., McGuire, A. (2001). Land

cover change on the Seward Peninsula: The use of

remote sensing to evaluate the potential influences of

climate warming on historical vegetation dynamics.

Canadian Journal of Remote Sensing, 27, 542-554.

Xiaolu, S., Bo, C. (2011). Change Detection Using Change

Vector Analysis from Landsat TM Images in Wuhan.

Procedia Environmental Sciences, 11(A), 238-244.

Yuan, F., Sawaya, K. E., Loeffelholz, B. C., Bauer, M. E.

(2005). Land cover classification and change analysis

of the Twin Cities (Minnesota) Metropolitan Area by

multitemporal Landsat remote sensing. Remote

Sensing of Environment, 98, 317-328.

Monitoring Protected Areas using Remote Sensing Technology

113