Thematic accuracy analysis of the CORINE classification based on Sazlidere basin

The Sazlıdere Basin is located on the European side of Istanbul within the borders of Arnavutkoy and Basaksehir districts. The total area of the basin, which is largely located within the province of Arnavutkoy, is approximately 177 km2. The basin is one of seven important basins that provide drinking water to Istanbul with an annual average of 55 million m3 of water. The construction of Sazlıdere Dam was completed between 1991-1996 and Sazlıdere Dam Lake started to hold water from 1996. The Sazlıdere Basin is faced with intense urbanization pressures due to the Northern Marmara Motorway, 3rd airport and Channel Istanbul Projects, which are planned to be realized in the Arnavutkoy region. A part of the Arnavutkoy district center is developing in the eastern borders of the Sazlıdere Basin. The area is located inside the long range protection zone and outermost region of the basin. Due to the mentioned projects, intense land cover and land use changes occur in the Sazlıdere Basin. In this study, land use and land cover classes of the basin were determined by remote sensing methods. The changes that took place over the years in these classes were statistically analyzed. The thematic accuracy assessment of the processes was investigated by carrying out field studies. 1975, 1997, 2000, 2011, 2012, 2015 and 2016 dated LANDSAT MSS, ETM and OLI images were used in this this study. 1975, 2000, 2012 and 2016 dated images were supervised classified based on CORINE Land Cover first level to determine the land use and land cover classes and their changes inside the Sazlıdere Basin area. As a result, four information classes were identified. These classes are water bodies, forest and semi-natural areas, agricultural areas and artificial surfaces. Accuracy analysis of all the images were performed following the classification process. By generating error matrix the kappa statistical value and general accuracy of each image were determined. It has been understood that the values expressing the accuracy of the images were above the acceptable limits. According to the classification results, from 1975 to 2012, there was an increase in agricultural areas, water bodies and artificial surfaces classes. It has been determined that the class of forests and semi-natural areas decreased between the same dates. On the other hand, forest and semi-natural areas increased while the agricultural areas decreased from 2012 until 2016. The reason for this situation may be the real estate market that is affected by the projects that are planned around the study area and the agricultural areas that are changing owners with different income expectations. The transformation of unused agricultural areas into bare soil, shrubbery and grassland has led to an increase in the class of forests and semi-natural areas. After the accuracy assessment, the change detection process has been performed so that it can be determined which class changes to which class in which direction. It was determined at this stage that during the period from 1975 until 2012, the forests and semi-natural areas tended to become agricultural land. However, from 2012 until the year 2016, the opposite was true, with the appearance of unused agricultural areas, this regions were turned into the forests and semi-natural areas. In addition, some agricultural and forest and semi-natural areas on the eastern side of the basin have been transformed into artificial surfaces in 1975 until 2016. In addition, in 1975, 2000, 2012 and 2016 dated classified images, the status of land use and land cover classes within each protection zone and reservoir area of the basin has been examined separately. The supervised classified images that have the smallest mapping units 0.09 ha and 0.64 ha were generalized to be compatible with the CORINE Land Cover data. The image pixels have been rearranged by using the thematic pixel aggregation method as the smallest mapping unit is 25ha. These results were compared with CLC 2000 and CLC 2012, which were obtained by digitizing land cover and land use classes on satellite images. It has been determined that the compared results are compatible with each other in terms of quality and quantity. However, in the case where the smallest mapping unit was accepted as 25 ha, it was seen that there was a great loss in land cover and land use information. In order to investigate the easiness and quick determination of the current land cover and use classes of Sazlıdere Basin, various arithmetic operations were applied to the image dated to 2016. MNDWI was used to determine the water bodies class, NDVI was used to determine the vegetation thematic class, BAEI was used to determine the artificial surfaces thematic class and SI was used to determine the bare soil thematic class. In order to obtain information classes compatible with thematic classes, the pixel-based classification process was applied again in 2016 dated image. When the results are compared with each other and with high spatial resolution images, index results have been found to be successful in achieving a general knowledge of the terrain cover and use situation of the basin. However, according to the classification results, it has been understood that there is a certain amount of data loss in the thematic classes studied. Sazlıdere Basin has been researched by using various remote sensing methods to determine land use and land cover classes based on the CORINE classification system. Then it was decided to investigate the thematic accuracy with the field studies. For this purpose, the LUCAS methodology, which is used to control the land cover and land use databases generated from satellite images, has been utilized. Starting from the point selection in this methodology, each stage of fieldwork is organized with certain standards and rules. In Sazlıdere Basin, 81 points were determined in a 9 x 9 grid structure with 1 km x 1 km intervals in the area where the land use and cover varied the most. Eight separate field works were carried out and 71 points were visited. GPS measurements, photographs, land cover and usage were recorded and cross-sections were taken in accordance with the LUCAS methodology guidelines. High spatial resolution images were used to determine land cover and land use of 10 points that could not be reached. The region chosen for the field works is approximately 90 km2. This area is about 13% of the total area of Arnavutkoy. This region was separated from 2016 dated satellite image and classified based on the first levels of the CORINE and LUCAS classification systems and the accuracy analysis was performed. MNDWI, NDVI, BAEI and SI methods were used to obtain information on the thematic classes of the chosen region in a short time and more easily. Then, results were compared with the classification results for the same thematic classes. While significant results are obtained in artificial surfaces, water and plant thematic classes, information losses in the soil thematic class have been mentioned. There are six information classes in the image of the pilot region that are classified according to the first level of the LUCAS classification system. These classes are water bodies, artificial surfaces, forest areas, bare soil areas and shrubland and grassland areas. Although the shrublands and grasslands in the LUCAS classification system fist level are specified as separate classes, but they are not considered to be able to distinguish this classes in 30 m spatial resolution image, and they are considered as a single class. The images classified according to the first level of the LUCAS classification system were compared with the 81 points' data and 82% match was determined. At the next stage, the image was rearranged and generalized to be 25ha so that the pixels were compatible with the smallest mapping unit of CLC. As a result of this generalization, the match ratio has declined to 71%. In order to prevent the loss of land cover and land use from the square, the classified image pixels have been rearranged so that the smallest mapping unit is 5 ha. In comparison with fieldwork data, the match rate increased to 77%.