Study and Analysis of Worldview-2 Satellite Imagery for Evaluating the Energy Efficiency

South-EasternEuropean Journal
of EarthObservationandGeomatics / Issue
Vo1, No1, 2012

Land Use Determination of Kara Menderes Basin (ÇANAKKALE-

TURKEY) With Remote Sensing

Mülayim Güre a,*, Cengiz Akbulak b, Hasan Özcan c

aAssistant Professor, Faculty of Sciences and Arts,

Department Space Science and Technology,

bAssociate Professor, Faculty of Sciences and Arts,

Department Geography,

cProfessor, Faculty of Agriculture, Department of Soil Science,

CanakkaleOnsekiz Mart University, Turkey

*Corresponding author: , +902862171303

Abstract: Recently, there have been many studies on the detection and protection of natural sources. In this respect, especially the detection of land use deserves great importance. Remote sensing facilitates such detections. This is not limited to current land use, but also the change of the land use in time. Due to these facilitations, remote sensing data and methods are becoming more common in Turkey as it is in the world.

In this study, it is aimed to explore the land use utilizing the case of Kara Menderes Basin. As the data, the SPOT 5 satellite data images, which have high resolutions and belong to the date of May 10 2010, are used. The classification is achieved using maximum likelihood algorithms in the scope of supervised classification method. In this study, nine different land use types are detected. The distribution of these classifications are displayed in the headings of “settlements roads, stone mine”, “forest”, “scrub”, “Grassland and pasture”, “dry agriculture”, “irrigated agriculture”, “Fruit and vineyard”, “water surfaces ”, and the like. According to the results of the classification, the largest land use has been detected as forests with 55.7 percent.

Accuracy assessments of the classification results were done according to the “random sampling algorithms” and “the ground truth” methods that were highly used in the land studies literature. Accuracy assessment results are found to be 90.60 percent for random sampling algorithms and 90.34 percent fort he ground truth data.

Key words: Kara Menderes Basin (Çanakkale), remote sensing, image processing, land use classification.

1.  Introduction

Recently, information technologies are widely used for human purposes in everyday life. The use of satellites has helped to improve human life in various areas. Especially, the use of Global Positioning Systems (GPS) with the development of new satellites and the corresponding innovations, the systematic management of positioning and tracking technologies has gained increasing importance. Hence, Geographical Information Systems (GIS) has played an important role in tracking and positioning applications. GIS is “a system that captures, stores, analyzes, manages, and presents data with reference to geographic location data. In the simplest terms, GIS is the merging of cartography, statistical analysis, and database technology” (URL1).

Layers, the raw satellite images, are one of the most important elements in GIS applications. Remote sensing technologies have made possible the use of high resolution digital images in the exploration and investigation of wide land covers. When technologies developed by the use of high resolution digital imaging and integrated with GIS, the management and organization of natural and artificial resources have become more efficient (Yomralıoğlu and Çelik, 1994).

Observation and interpretation of various events that are taking place in the environment largely depends on the efficient use of these kinds of technologies. While land surveys could be seen sufficient for the observation of smaller areas, this method is not sufficient enough to use in the exploration and investigation of larger areas. Like other disciplines where these technologies are used, it is also possible to use satellite technologies for on-site monitoring of land cover in the process of monitoring and evaluation of agricultural fields in a quicker and continuous manner (Genç and Bostancı, 2007).

This paper is focused on the current land use in Kara Menders Basin. The types and the uses of lands that are located in the basin have been explored and explained with the help of satellite data. Plant calendar and seasonal condition are used in the classification system.

2.  Method

2.1.  Study Area and Data

The research team has purchased a digital image that covers the aforementioned study area. The image was taken by SPOT 5 on May 10, 2010. The image is presented in Figure 1 and Table 1. Because it was not possible to include the complete study area in one frame, the south and east parts of the Basin were purchased as individual frames. The images were then subjected to some preliminary applications, like; geometric correction, image mosaicking, intensity hue saturation (IHS) conversion, and extraction of the interested area from the whole image.

a. C Users Dimitris Desktop png
b. c.
Figure 1. a. Map of Turkey, b. 231 band combination SPOT 5 image, c. Study Area.
Table 1: Satellite Data, Acquisition date and Band Properties.
Satellite Data / Acquisition Date/time / spatial resolution (m) / Band1 µm / Band2 µm / Band3 µm / Band4 µm
SPOT 5 VNIR Level 2A / 12-05-2010/09:11 / 10 / 0.49-0.61 / 0.61-0.68 / 0.78-0.89 / 1.58-1.75
SPOT 5 PAN Level 2A / 12-05-2010/09:11 / 5 / 0.49-0.69

The GPS data (Fig. 2) was gathered and root mean square (RMS) was calculated to make sure that the geometric corrections on the image were precise. The researchers made sure that value of RMS was less than 4.07 meters.

The satellite images that belong to the area of interest were first subjected to the mosaicking procedure, then they were combined using ERDAS© Imagine Software (Version 9.0). Later, using IHS conversion, images were matched to produce high resolution areal images. The result of IHS conversion provided a new tree-band image. This final image was used in the definition of ground control points.

The area of the Basin was extracted with careful attention to the borders of the Basin via the use of subset function in ERDAS© Imagine Software.

2.2.  Analysis

Supervised and unsupervised classification methods are the two commonly used classification methods in the land use classification procedures that use remote sensing data. In supervised classification method, the training stage data has to be defined in the analysis software. Only after this definition, the classification is possible. For this procedure, some preliminary data is needed (Matkav and Sunar, 1991).

The supervised classification method is used in this study. Because in this method, it is possible to gather control points using ground work and related to these control points. It is possible to reach more trustable results which can easily be subjected to various verification procedures.

Although maximum likelihood and minimal distance algorithms are used during classification procedures, maximum likelihood algorithm is the one used the most in the related literature (Matkav and Sunar, 1991). Hence, this study uses maximum likelihood algorithms. In the categorizations where maximum likelihood algorithm is used, “the probability density functions are used to classify an unidentified pixel by computing the probability of the pixel value belonging to each category” (Richards and Jia, 2006). With this the probability of the pixel value occurring in the distribution of a given class would be defined by the computer.

When classifying an unknown pixel, both variance and covariance of the category spectral response patterns are quantitatively evaluated by the maximum likelihood classifier. To do this, there needs to be an assumption that the distribution of points forming the category training data is normally distributed. For common spectral response distributions, this assumption of normality is generally reasonable. According to this assumption, it is possible to completely describe the distribution of a category response pattern with the use of the mean vector and the covariance matrix. Given these parameters, it becomes possible to compute the statistical probability of a given pixel value being a member of a particular land cover class. When the probability values are plotted in a tree-dimensional graph, the vertical axis is associated with the probability of a pixel value being a attributed to one the class. The resulting bell-shaped surfaces are called probability density function and there is one such function for each spectral category (Lillesand, et al., 2004).

Figure 2. The use of different types of land taken by GPS control points.

In this study, nine different control types belonging to land use classification were used. First, the classification process is done using ERDAS© Imagine 9.0 software. In this process a set of controls were prepared via the use of Area Interesting Object (ΑΟΙ) tool kit. Signature Editor of the data set was created with the use of the control areas. The data set generated in this stage was used in the supervised classification process. Thus, the classification process was completed using supervised classification method. Because there were many classes as a result of the classification, the similar classes were combined and renamed using the “recode” function in ERDAS© Imagine software version 9.0. Following this transaction, to eliminate the noises in the classified images, image filtering was utilized. The filtering process was conducted using the statistical filtering (majority 3x3) function in ERDAS©. Thus, the current land use status was defined using the supervised classification method which was obtained by classifying the satellite images (Fig. 3).

Figure 3. The final image as a result of supervised classification.

3.  Results

The classification accuracy analysis of the application is the most important step in land use classification to reach reliable outcomes of remote sensing imaging processes (Jensen, 1996). In this study, random sampling and ground truth data analysis were used as analysis procedures. For this purpose, "Overall Accuracy” and "Error Matrix" which are widely used in the literature as accuracy analysis techniques were utilized. Analysis was done with the use of, "Accuracy Assessment" function in the imaging software.

In the process of the accuracy analyses of the image, the determination of the number of sample points should be done according to the size and number of classifications in the image (Jensen, 1996). In this study, taking into account of the limits provided in the literature, a total of 675 points were controlled using 75 points for each class. Researchers selected simple random sampling (SRS) method to ensure the selected control points were representative and were chosen without user bias. Ground truth studies were conducted after the acquisition of remote sensing data. During the control point determination process, GPS data was gathered and transferred to the system. Relying on the assumption that classification is accurate and reliable when the final result of the classification is 80 percent and above; we conclude that the classified images were accurate and reliable (Koç, and Yener, 2001).

Table 2. "Kara Menderes" Basin land-use status according to satellite image for the period of May 2010.
Land Use Type / May 2010
Ha / %
Irrigated agriculture / 9,142 / 4.6
Dry agriculture / 42,118 / 21.1
Grassland and pasture / 10,611 / 5.3
Fruit and vineyard / 14,735 / 7.4
Forest / 111,249 / 55.7
Scrub / 8,066 / 4.0
Settlementroads, stone mine / 2,664 / 1.3
Water surfaces / 913 / 0.5
Other / 72 / 0.0
Total / 199,569 / 100

The classification results, which were done by the use of supervised classification method, were subjected to the classification accuracy analysis to portray the land use with the utilization of remote sensing images.

The results of classification accuracy analysis for May 2010 are shown in Table 3 and 4. For the aforementioned date, while the classification accuracy analysis done using SRS method resulted with a mean accuracy level of 90.60 %; when ground truth method was used, the same analysis resulted with a mean accuracy level of 90.34 %. These results revealed that the results reached with the use of both SRS and ground truth methods were comparable to each other. Thus, as we mentioned before, the classification done with both methods were at an acceptable level.

The study proved that the 55.7 percent of the basin area was covered with forests which corresponded to the largest class found in the basin. Dry agricultural areas were the second largest class with 21.1 %. This was followed by the areas dedicated to fruit and vineyard, grassland and pasture, irrigated agriculture, and scrub, with 7.4 %, 5.3%, 4.6%, and 4% respectively. These results were shown in Table 2. In total, 1.8 % of the basin was dedicated to residential use, roads, stone mine, and water surfaces.

Table 3. Classification results via the simple random sampling method with the
error matrix and accuracy levels.
Reference Data
Settlement roads, stone mine / Other / Irrigated agriculture / Dry agriculture / Grassland and pasture / Fruit and vineyard / Forest / Scrub / Water surfaces / Total / Reference Number / Total number / Correct Number / Produced accuracy / User accuracy
Classification Data / Settlement roads, stone mine / 73 / 0 / 4 / 3 / 0 / 0 / 0 / 0 / 0 / 80 / 76 / 80 / 73 / %96.05 / %91.25
Other / 0 / 51 / 1 / 0 / 0 / 0 / 0 / 0 / 0 / 52 / 51 / 52 / 51 / %100.00 / %98.08
Irrigated agriculture / 0 / 0 / 70 / 0 / 0 / 4 / 0 / 1 / 2 / 77 / 77 / 77 / 70 / % 90.91 / %90.91
Dry agriculture / 1 / 0 / 0 / 65 / 0 / 2 / 1 / 0 / 0 / 69 / 82 / 69 / 65 / % 79.27 / %94.20
Grassland and pasture / 0 / 0 / 1 / 5 / 59 / 2 / 1 / 4 / 0 / 72 / 65 / 72 / 59 / % 90.77 / %81.94
Fruit and vineyard / 1 / 0 / 0 / 2 / 1 / 48 / 1 / 2 / 0 / 55 / 61 / 55 / 48 / %78.69 / %87.27
Forest / 1 / 0 / 1 / 5 / 1 / 2 / 72 / 1 / 0 / 83 / 76 / 83 / 72 / %94.74 / %86.75
Scrub / 0 / 0 / 0 / 2 / 3 / 3 / 1 / 69 / 0 / 78 / 77 / 78 / 69 / % 89.61 / %88.46
Water surfaces / 0 / 0 / 0 / 0 / 1 / 0 / 0 / 0 / 71 / 72 / 73 / 72 / 71 / %97.26 / %98.61
Total / 76 / 51 / 77 / 82 / 65 / 61 / 76 / 77 / 73 / 638 / 638 / 638 / 578
Overall classification accuracy = % 90.60
Tablo 4. Classification results via the ground truth method and the error matrix and accuracy level.
Reference Data
Settlement roads, stone mine / Other / Irrigated agriculture / Dry agriculture / Grassland and pasture / Fruit and vineyard / Forest / Scrub / Water surfaces / Total / Reference Number / Total number / Correct Number / Produced accuracy / User accuracy
Classification Data / Settlement roads, stone mine / 29 / 0 / 0 / 0 / 0 / 0 / 0 / 0 / 0 / 29 / 29 / 29 / 29 / %90.63 / %100.00
Other / 0 / 17 / 0 / 0 / 0 / 0 / 0 / 0 / 0 / 17 / 17 / 17 / 17 / %100.00 / %100.00
Irrigated agriculture / 0 / 0 / 37 / 0 / 0 / 3 / 0 / 0 / 2 / 42 / 42 / 42 / 37 / %94.87 / %88.10
Dry agriculture / 1 / 0 / 0 / 55 / 0 / 2 / 1 / 0 / 0 / 59 / 59 / 59 / 55 / %90.16 / %93.22
Grassland and pasture / 0 / 0 / 1 / 2 / 17 / 0 / 0 / 3 / 0 / 23 / 23 / 23 / 17 / %85.00 / %73.91
Fruit and vineyard / 1 / 0 / 0 / 1 / 1 / 22 / 0 / 0 / 0 / 25 / 25 / 25 / 22 / %73.33 / %88.00
Forest / 1 / 0 / 1 / 3 / 1 / 2 / 40 / 0 / 0 / 48 / 48 / 48 / 40 / %97.56 / %83.33
Scrub / 0 / 0 / 0 / 0 / 1 / 1 / 0 / 23 / 0 / 25 / 25 / 25 / 23 / %88.46 / %92.00
Water surfaces / 0 / 0 / 0 / 0 / 0 / 0 / 0 / 0 / 22 / 22 / 22 / 22 / 22 / %91.67 / %100.00
Total / 32 / 17 / 39 / 61 / 20 / 30 / 41 / 26 / 24 / 290 / 290 / 290 / 262
Overall classification accuracy = % 90.34

4.  Conclusions