Agrárinformatikai Nyári Egyetem és Fórum

Gödöllő, 2004. augusztus 25-27.

APPLIED REMOTE SENSING TECHNOLOGY TO ANALYZE THE LAND COVER/LAND USE CHANGE AT TISZA LAKE Yudhi Gunawan, [email protected] - Department of Land Use and Region Development, Debrecen University, H-4032 Debrecen, Böszörményi út 138, - ATT & IT Kft, Thököly út 58-60, Budapest Tamás János, [email protected] Department of Water and Environmental Management, Debrecen University, H-4032 Debrecen, Böszörményi út 138

Abstract The land cover/land use changes at Tisza Lake could be detected by satellite images. From the analysis the multi temporal images of Landsat TM and Landsat ETM within 13 years of the study area was obvious the transition of several land cover/land use categories. The changes were observed by using image differencing technique. Most of the categories converted into forest, wetland since the environmental issue in Hungary has been rising up.

INTRODUCTION Remote sensing technologies are becoming recognized as an important evaluation tool to conduct ecosystem evaluation and environmental management because the tools of remote sensing can be used to monitor land cover change and the environmental conditions of the Earth's surface. It is also possible to use these tools to investigate the interaction between land cover/land use change and environment impact of reservoir. The purpose of this research is to study the land cover change and fluctuation of the environmental conditions at the Tisza Lake and surrounding area using satellite imagery. Recently unplanned changes of land use have become a major problem. Large flooding, air pollution in large cities as well as deforestation, urban growth, soil erosion, desertification are all consequences of a mismanaged development planning without considering its environmental impacts. An attempt will be made to use the capabilities of remote sensing along with GIS tools to assess the status of the ecosystem and land cover/land use change in the surrounding of the lake and also water quality in the lake. STUDY AREA Tisza Lake is located in the northern part of Hungary. Based on UTM/WGS 84 coordinate system, Tisza Lake placed at 20º26’ to 20º52’ E and 47º43’ to 47º25’ N. Tisza Lake is an artificial lake that was formed by the construction of the Water Power Station of Kisköre in the middle of the steppe, the Great Hungarian Plain.

1

Agrárinformatikai Nyári Egyetem és Fórum

Gödöllő, 2004. augusztus 25-27.

[16] documented that by damming up the section between Kisköre and Tiszavalk (441.0403.2 river km) Tisza Lake has been made within the foreshore of the Tisza, which with its 127 km2 extension has become the second largest “stagnant” water of the Carpathian basin. Damming was enabled by putting the Kisköre hydrological power into operation in 1973. The aim of the investment was the complex ecogeographical reconstruction of the mid Tisza valley, the improvement of natural and social relationships.

Tisza Lake

Figure 1. The map of Tisza Lake ENVIRONMENTAL ISSUES IN TISZA LAKE The initial aims of damming a sort of lake in the Tisza River were to generate irrigation systems, energy power, industrial water, recreation area and natural protection. After the collapse of the command economy and industrial structures, because of the change of energy source to nuclear power, the rising need of natural protection and the increasing purposes of tourism, the objectives of this lake have grown and therefore there are urgent demands for appropriate strategies to avoid contradiction among purposes. For recreational and fishing purposes high water level is needed but it is not advisable for natural protection particularly not for breeding water-fowls. High velocity is required for obtaining energy but it is opposite of the need of tourism. The higher the storage level of a reservoir, the more electric power is generated. However, as storage level gets higher, the amount of water that can be stored for flood control decreases. Dam construction alters the natural balance of sediment flow in rivers by impounding sediment within and upstream of the reservoir and discharging clear water downstream [9]. This results in loss of storage capacity in the reservoir, and obstruction of intakes and abrasion of turbo machinery at the dam site. Upstream of the dam, it causes aggradations of tributaries, deposition at diversions, reduction in navigational clearance, and increase in flood frequency. Downstream, it leads to scouring of the river bed, degradation of tributaries, and undercutting at diversions [13]. Tisza Lake as a living river is vulnerable from water pollution as it happened in 2000, a cyanide pollution from mining problem in Romania [2]. Flushing method was applied to mitigate the water pollution but it is dissension with environmental protection, recreation and settlements along Tisza Lake.

2

Agrárinformatikai Nyári Egyetem és Fórum

Gödöllő, 2004. augusztus 25-27.

The geographical situation of the watershed surrounding the Tisza Lake as part of the Carpathian Basin influences the water level and it can cause high fluctuation of water level, also trigger flood risk, sedimentation and eutrophication. MATERIAL AND METHODS Data Description For the remotely sensed data, two date satellite images are acquired to perform the study analysis. All the satellite imageries were obtained from University of Maryland, Global Land Cover Facility as can be seen in table 1. Table 1. Satellite Imageries used for the study Type Landsat TM5 Landsat ETM7

Acquired 08-07-1987 20-08-2000

Pixel size 25 x 25 28 x 28

Path and Row 187/027 187/027

CORINE Land Cover scale 1 : 100 000 (CLC100) as reference data was used. CLC100 contains 44 classes within 3 levels. Topography map of Hungary in 1 : 10 000 scale and orthophoto which was made by Middle-Tisza District Environment and Water Authority with 0,5 meter resolution were applied as data complements. The photo was taken on 11 August 1997 in the morning with the flight height approximately 5000 m. The scale of the orthophoto is 1 : 30 000 and the accuracy of joining is approximately 1 m. Methodology 1. Proposed Approach The approach used in this study to detect the land cover/land use changes involves different preliminary processing steps before performing digital change detection. This objective could be achieved through: • Image rectification and co-registration of the 2 satellite images; • Unsupervised classification of the satellite scenes supported by some field knowledge in order to produce land cover/land use maps; • Performing a land cover comparison with CORINE Land Cover map of the study area (CLC100) using classification accuracy assessment method. • Performing a change detection; • Investigation of the relationship between land cover/ land use changes and the natural environment fluctuation. 2. Pre-processing After the remotely sensed data have been received, a geometric correction was carried out to change the map models and ensure all data have the same projection properties into Hungarian projection (EOV). A subset of the area of interest was then made to accelerate the processing time. The next step was to perform the co-registration of both satellite imagery using the topography map of Hungary at scale 1 : 10 000 scene as the reference or the master image. The co-registration was carried out using several ground control points and a first

3

Agrárinformatikai Nyári Egyetem és Fórum

Gödöllő, 2004. augusztus 25-27.

order polynomial rectification. Well-defined features in the master image such as road intersections, corners of large buildings, bends in rivers, industrial features were chosen as ground control points. 3. Unsupervised Classification Unsupervised classification is a numerical process that search for natural groupings of the spectral properties of pixels. This method usually used when less is known about the data before classification. It is then the analyst’s responsibility, after classification to attach meaning to the resulting classes [14]. The classification of the satellite images is done using a clustering algorithm called the Iterative self-Organization Data Analysis Technique (ISODATA). ISODATA method, available in ERDAS software, generally requires the analyst to specify some criteria such as the maximum number of clusters, the maximum percentage of pixels whose class values are allowed to be unchanged between iterations, and the minimum members in a cluster [5]. Unsupervised classification ran for both satellite imagery scene using 7 classes based on first and second level of CLC100, a number of iterations equal to 6, and a convergence threshold of 0.95. The 7 classes were then attributed using some field knowledge of the study region, partly some information from orthophoto covering a small area of the whole image, and the unclassified satellite image. 4. Classification Comparison The degree of agreement between the unsupervised classification of Landsat TM and ETM scene of the Tisza Lake area and the CLC100 classification as the reference of the same area was tentatively analyzed. The agreement analysis is done through the accuracy assessment technique between both classifications. The CLC100 was initiated for the EU countries in the 1980's to provide quantitative information on land cover, at a scale 1:100,000 [3]. CLC is mapped by interpreting satellite images, with the results stored as a database in a Geographic Information System [6]. 5. Change Detection There is a wide range of techniques used for land cover/land use change detection. These techniques can be subdivided into the following categories [1]: Composite Image; Image Comparison; Comparison of the Classified Images; Combination of the Classified Images; Radar Classification In this study multi date composite image method was applied. The modeling was constructed to carry out change detection using model maker tool of ERDAS 8.6 with image differencing technique (figure 2). Image Differencing involves subtraction of two images or two bands pixel by pixel [7]. A constant value of 128 was added to the difference file values to compensate for software limitations prohibiting negative values.

4

Figure 2. Model maker to perform change detection

Agrárinformatikai Nyári Egyetem és Fórum

Gödöllő, 2004. augusztus 25-27.

RESULT AND DISCUSSION Image Rectification and Restoration Raw digital images usually contain geometric distortions. The intent of geometric correction is to compensate for those distortions such as systematic distortion and random distortions by analyzing well distributed ground control points (GCPs) occurring in an image. The overall Root Mean Square Error (RMSE) of the co-registration process was around 13 m for the Landsat TM satellite scene and 14 m for the Landsat ETM scene (see figure3).

Figure 3 : Control point errors for Landsat TM and ETM co-registration

Classification

The result of unsupervised classification for Landsat TM and ETM along with a histogram values and class names are shown in figure 4 below.

Figure 4: Unsupervised classification of Landsat TM and ETM Classification Comparison

5

Agrárinformatikai Nyári Egyetem és Fórum

Gödöllő, 2004. augusztus 25-27.

The classification comparison between the Landsat imageries and the reference data was done by accuracy assessment. An important factor in determining the accuracy of a classification is the number of reference pixels used. [4] stated that it has been shown that more than 250 reference pixels are needed to estimate the mean accuracy of a class to within plus or minus five percent. The accuracy assessment process done by stratified random method in ERDAS, with 256 points selected. This sampling method used to provide statistical validity and minimize spatial correlation.

Figure 5: Location of the 256 random sampling points shown in the Landsat ETM viewer (left) and CLC100 viewer (right) The accuracy assessment presented in a classification error matrix or confusion matrix or so called a contingency table respectively, as reported below. Table 2. Error Matrix Resulting from Classifying Training Set Pixels at Landsat TM & ETM

Overall accuracy is computed by dividing the total correct (sum of the major diagonal) by the total number of pixels in the error matrix; the producer’s accuracy indicates the probability of a reference pixel being correctly classified and is a measure of omission error. Whereas the user’s accuracy is the probability that a pixel classified on the map actually represents that category on the ground, and is a measure of commission [15].

6

Agrárinformatikai Nyári Egyetem és Fórum

Gödöllő, 2004. augusztus 25-27.

The overall accuracy of 26% and 38% indicates a pretty low agreement between the Landsat TM and ETM unsupervised classification with the reference CLC100 classification. This low accuracy number may be explained by the fact that CLC100 was carried out in Hungary in 1990 by applying paper based interpretation and then transforming it to digital form using scanning method [11]. The assess land cover/land use up to 2 level of CLC100 were not enough to represent the land cover/land use classes. And the gap of over 3 or 10 years between the data can carry a lot of land cover/land use change especially if considered the high growth rate of urbanization and tourism in the last decades. It also indicates that unsupervised classification alone is not sufficient enough to depict the land cover/land use in the study area. Further classification method has to be applied such as supervised classification together with field inspection for the uncertain spots. Computation of Kappa statistics was made in an attempt to get as much information from the error matrix as possible. Table 4 contains the report of the Kappa statistics of Landsat ETM and TM for each category of classes as can be seen below. Table 4. Kappa Index of Landsat ETM (left viewer) and TM (right viewer) unsupervised classification

Kappa Index or k (KHAT) statistic is a measure of the difference between the actual agreement between reference data and an automated classifier and the chance agreement between the reference data and a random classifier [16]. Conceptually, k can be defined as Observed accuracy – chance agreement K=

(1)

1 – chance agreement

Water-body was the most accurate class in Landsat TM (0.90) and ETM (0.66) respectively. It has to be taken into consideration that the comparison between CLC100 which was made in 1990 in Hungary and TM 1987 contains a lot of error in the class of wetlands, forests and arable land. Comparison between CLC100 (1990) and Landsat ETM (2000) consists of mistakes among wetlands, forests and settlements. It might be because of the big change in land management and land reform in agricultural area ([10] reported that in the 1990s, parallel with the political changes, the transformation of Hungarian agriculture also began), the need of conservation area and the expansion of urbanization. In Landsat TM, the settlement (0.35) and in Landsat ETM, grassland (0.47) however was reasonably accurate compared to other classes.

7

Agrárinformatikai Nyári Egyetem és Fórum

Gödöllő, 2004. augusztus 25-27.

Many considerations should be taken into account when assessing this comparison analysis, among others, how appropriate is the sample size of control points, and how well the understandings of the classification system have lead to a good attribution.

Change Detection The Landsat TM 1987 image has a cell size of 25 m whereas Landsat ETM 2000 has a cell size of 28 m, therefore a re-projection to change the image resolution of the TM 1987 was done to change it to a 28 m cell size image. The water-body is mostly lost in favor of wetlands (2%) as can be looked at table 5. The changes of water bodies to wetlands seem due to natural process trigger by human activity for instance the sedimentation process and eutrophication. Table 5. The summary matrix of the change detection between Landsat TM 1987 and Landsat ETM 2000

Most of the wetlands were converted to water-body (15%). This could happen because of the relatively big drought during the 1990’s and then when the normal precipitation returned, most of the wetlands were covered back by water. The data which obtained from VITUKI concerning discharge and velocity of the water flow, can be seen the dryer period reflected by the lower water level. Another high transformation of wetlands into forest class (13%) might be due to the new environmental protection policy in Hungary. Arable lands were distributed to all land cover/land use categories except into water-body. The forest has the biggest (26%) chance to be converted from arable land. The increase of urbanization (19%) is raising the need of settlements (recreational areas, closed gardens) that could be trigger for converting arable land. The arable lands also have high transformation into wetlands (14%). Since the Hungarian agricultural production exceeds the limits of the EU regulations, the less fertile soils are reannexed to natural conservation or the process of forestation has started [8] There was misinterpretation at the settlement category. The settlement transformed into wetland, forest, arable land, grassland and industrial area. This error could be occurred due to the mixed pixels. These mixed pixels present a difficult problem for image classification, since their spectral characteristics are not representative of any single land cover type. Further effort has to be done to overcome of the classification of mixed pixels by applying Spectral Mixture Analysis and Fuzzy classification [15].

8

Agrárinformatikai Nyári Egyetem és Fórum

Gödöllő, 2004. augusztus 25-27.

Environmental analysis Some of the environmental issues that can be investigated in this study include forest and wetland loss; dredging activities, sedimentation, turbidity and levels of suspended solids along Tisza Lake and surrounding. One of the pressures from human activities on a water body is the continuous discharge of sediments that worsen its turbidity. This condition appears due to the presence of high level of suspended solids in the waters and at certain times of the year the presence of large phytoplankton population. The spatial and temporal variability of some water quality variables such total suspended solids or nutrient levels can be examined using historical water quality monitoring data and GIS tools to display a map of monitoring stations locations along with their measurements [12]. Raster images of average values of water quality data can be produced using GIS and then overlaid on the satellite images to study any correlation between the reflectance in the Landsat image, land cover change and water quality. Suspended solids in water produce visible changes in the surface of the waters and in the reflected solar radiation. Such changes in the spectral signal from surface water can be captured by the satellite [15]. The other purposes of this research was to study the fluctuation of some surface water quality parameters such as total suspended solids, heavy metals, and sediment quality in response to land cover change. The search for water quality data on the study area was not very fruitful since most of the data was either unavailable or missing in the period of study. Unfortunately, due to time limitation and difficulty to find appropriate water quality data, the environmental analysis has not been carried out as planned but further research is still going on. CONCLUSSION AND SUGGESTION The research was done to study the land cover classification, accuracy, and change during a 13 year period of the Tisza Lake area. The accuracy assessment of this study was pretty low indicating that further classification method has to be applied. Results of the change detection showed an increase of natural protection area such as forest and wetland categories due to the EU regulation on conservation aspect and could be the raising of awareness to environmental concern. The environmental analysis of the study area that would investigate the impact of land cover fluctuation on water quality and environmental issues was not conducted due to the complexity of the problem, the availability of the data and time limitations. Further research has to be done to work out the problem of mixed pixel classification and the environmental issues in the study area. Updating the reference land cover map data of CLC50 instead of CLC100 could be increase the accuracy of the land cover/land use classification in the study area. The CLC50 at the scale of 1 : 50 000 which has extended nomenclature until level 5 classes representing the landscape condition and the 4 hectare minimum mapping unit gives better differentiation of forests and semi-natural vegetation and wetlands, which is important for conservation and biotope mapping and also has been adapted to the geographic condition of Hungary and the land use or land cover that are not present in Hungary were omitted, for instance 1.2.3.1 Military and naval ports [3].

9

Agrárinformatikai Nyári Egyetem és Fórum

Gödöllő, 2004. augusztus 25-27.

ACKNOWLEDGMENT The comments and suggestion of the following experts were highly appreciated in making of this paper: Márta Gunawan - FÖMI, Lelkes Miklós and Ernő Wittman – ATT & IT Kft., Gábor Bálint – VITUKI, Lovás Áttila & Vajk Ödön – KÖTIVIZIG, Gross Miklós – Eurosense. This paper also supported by OTKA T043179; T047366

REFERENCE 1. Belaid, M.A., 2003. Urban-Rural Land Use Change Detection and Analysis Using GIS & RS Technologies. Geographic Information for Planning. Morocco 2. Biró, E., Bodó, P., Demeter, I., Janiak., P., Dias, L.M., & Viss, V., 2001. EPCEM Project Report: Improving the Water Quality of the River Tisza. Center for Environmental Management & Policy. University of Debrecen. Hungary 15 pp 3. Büttner, G. 1998. The Phare CORINE Inventories in CEE: A Support System for Environmental Decisionmakers. Phare CORINE Land Cover Project. FÖMI Remote Sensing Centre. Budapest. Hungary 4. Congalton, R.G. 1991. A Review of Assessing the Accuracy of Classifications of Remotely Sensed Data. Remote Sensing of Environment 37(1): pp. 35-46 5. ERDAS. 2002. ERDAS Field Guild, 6th Edition. Atlanta. Georgia 6. European Environment Agency. 1999. CORINE Land Cover, Technical Guide. JRC. Italy. 7. Fung, T. and Zhang, Q., 1989, Land Use Change Detection with Landsat Digital Data in the Kitchener-Waterloo Area, Remote Sensing and Methodologies of Land Use Change Analysis, Waterloo, Department of Geography Publications Series, Occasional Paper No. 6, University of Waterloo, pp. 135-153. 8. Gunawan, Y. 2004. The Evaluation of Landuse Changes in 2002 compared to 1982 in Hungary and in the Northern Great Plain Region. Department of Landuse and Region Development. University of Debrecen. 9. Hotchkiss, R. H., and X. Huang, 1995. Hydrosuction Sediment-Removal Systems (HSRS): Principles and Field Test. Journal of Hydraulic Engineering 121:479-489. 10. Márkus, B., Szabo, Gy., Nyiri, J., & Udvardy P. 2003. Land Use Planning for Rural Development in Hungary.West Hungary University, Geoinformatics College. Székesféhervár 11. Maucha, G. (2004). Personal Communication 12. Mrini, M. 2002. Land Cover Change and Fluctuation of Water Quality in Corpus Cristi/Nueces Bay. Advanced Remote Sensing. Texas. 13. Morris, G. L. & J. Fan, 1998. Reservoir Sedimentation Handbook, McGraw-Hill, NY. 14. Jensen. 1996, Thematic Information Extraction Image classification. NY. 15. Lillesand, T. M. & Kiefer, R. W. 2000. Remote Sensing and Image Interpretation, Forth Edition. John Wiley & Sons Inc. New York. 724 pp 16. TERRA Foundation for Environment Protection and Education. 2000. Cyanide and heavy metal pollutions on the Tisza in spring 2000. Debrecen, http://www.terra.hu/cian/vizgyujto-en.html

10