Zaman Serileri İle Değişim Analizi: İstanbul, Sarıyer Örneği

Abstract

Sarıyer ilçesi, İstanbul’un Avrupa Yakası’nda ve İstanbul boğazın kuzey kesiminde yer almaktadır. İlçenin hem Karadeniz’e hem de İstanbul Boğazı’na kıyıları mevcuttur. Karadeniz kıyılarında dar bir sahil şeridi ve plaja uygun kumsalların hemen ardında dik yamaçlar yer alır. İlçe içerisindeki ormanlar İstanbul’un diğer ilçelerine oranla daha fazla alana sahiptir ancak bu alanlar geçmişten günümüze tahrip edilerek azalmaktadır. Kumsallar ve orman alanları haricinde ilçe içerisinde yerleşim ve tarım alanları da dikkate değer miktarda mevcudiyete sahiptir. Bu çalışmanın amacı; “Zaman Serisi Analizi” yöntemini kullanarak Sarıyer ilçesine ait arazi örtüsü ve arazi kullanım sınıflarının değişimini istatistik açıdan incelemektir. Yöntemin uygulaması iki aşamada gerçekleştirilmiştir. Bunlar uzaktan algılama verilerinin işlenmesi ve zaman serisi analizi aşamalarıdır. İlk aşamada, 1987, 1997, 2000, 2001, 2003, 2005, 2009 ve 2010 tarihli (LANDSAT TM + ETM) sekiz adet uydu görüntüsü ile çalışılmıştır. Bütün görüntülere sırasıyla kontrolsüz ve kontrollü sınıflandırma işlemi ve ardından doğruluk analizi uygulanmıştır. Kontrolsüz sınıflandırma işlemi ile görüntüden spektral sınıflar elde edilmiş, kontrollü sınıflandırma işlemi ile bu spektral sınıflardan CORINE veri tabanının birinci seviyesine ait dört adet bilgi sınıfına ulaşılmıştır. Bunlar, yapay yüzeyler, tarım alanları, su varlığı, orman ve yarı doğal alanlar sınıflarıdır. Sınıflandırma işleminin ardından her bir görüntünün sınıflandırma doğrulukları belirlenmiştir. Buna göre elde edilen en düşük doğruluk değeri, % 86.25 ile 2009 yılına, en yüksek doğruluk %91.40 ile 1987 yılına aittir. Bu değerlerin % 80 seviyesindeki doğruluk beklentisini karşıladığı görülmüş ve çalışmaya devam edilmiştir. Doğruluk analizi tamamlandıktan sonra uzaktan algılama verilerinin işlenmesine devam edilmiştir. Arazi kullanım sınıfları arasındaki değişimleri belirleyebilmek amacıyla değişim analizi yapılmıştır. Bu aşamada sınıflandırma sonucunda elde edilen görüntü matrisleri birbirleriyle karşılaştırılmıştır. Böylece hiç değişim göstermeyen alanların miktarı ve hangi sınıfın hangi sınıfa hangi miktarda dönüştüğü alansal olarak belirlenmiştir. Çalışmanın zaman serisi analizi bölümünde gelecek ile ilgili öngörü yapılması anaçlanmıştır. Bunun için değişim analizinde olduğu gibi sınıflandırma sonuçları kullanılmıştır. Bu çalışmaya başlamadan önce her bir görüntüde bilgi sınıflarının kapladıkları alanlar yüzde olarak ifade edilmiştir. Bunun sebebi geometrik dönüşüm sebebiyle ortaya çıkan alansal farklılıkların ortadan kaldırılmasıdır. Gelecek dönemlerde bilgi sınıflarında meydana gelmesi muhtemel değişimler doğrusal trend analizi ile belirlenmiştir. Bunun için zaman serilerinin özellikleri dikkate alınaraktoplam dokuz adet veri seti oluşturulmuş ve toplam dokuz adet analiz yapılmıştır. Bu analizlerden altı tanesinin sonucu kullanılabilir kabul edilmiş ve doğrulukları standart sapma ve güven aralığı hesaplanarak istatistik olarak incelenmiştir. Çalışma sonucu iki aşamada değerlendirilmeye müsaittir. İlk aşamada uzaktan algılama yöntemleri kullanılarak bilgi sınıfları arasındaki değişimleri belirlemek mümkün olmuştur. İkinci aşamada doğrusal trend analizi yöntemi istatistik analiz sonuçları birarada değerlendirilerek 2015, 2017, 2021, 2025, 2027 ve 2030 yıllarındaki orman ve yarı doğal alanlar, yapay yüzeyler ve tarım alanları sınıflarının nasıl değişim göstereceği ile ilgili olarak anlamlı sonuçlar elde edilmiştir. Buna göre 1987 yılından 2030 yılına kadar değerlendirildiğinde yapay yüzeylerde artış, tarım alanları ile orman ve yarı doğal alanlarda azalma söz konusu olacağı tespit edilmiştir.

The Sarıyer township is located at north region of the Bosporus and in European Side. The township has shores to both the Black Sea and the Bosporus. It is existed that steep slopes behind narrow coastal strip and appropriate to the beach area on the Black Sea shores. Forest inside the township has larger area than the other townships of İstanbul. However, these forest areas are decreasing by destruction from past to present. Except of forests and beaches, there is a considerable amount of residential and agricultural areas inside the township. Before starting the study, a variety of other studies about time series analysis and change detection methods, and information about these studies were summarized in the literature review part. The basic principles of remote sensing (electromagnetic spectrum and the spectral reflectance properties of objects) are described in the chapter two. Digital image processing techniques (digital image, resolution, and the classification process) is represented in the chapter three. Change analysis, time series analysis and methods were defined in the chapter four. Geographic location, physical characteristics, meteorological characteristics, vegetation, information about land use and environmental issues of the study area were gived in the chapter five. Similarly, information about data were listed in the chapter five. Information about field work and application were given in the chapter six. Field work was carried out in order to obtain information about the study area. During the field work of the study area several photographs were taken and GPS measurement were done. The photographs and table of coordinates relevant to GPS measurements viewed in the appendix. In this study, both raster and vector data used. Sarıyer township boundary used in vector data format. With this data, the area of interest was obtained from the image frame and also, the subset area was cutted from the full image. All satellite images used in the study are raster data structure. Besides, all satellite images were georeferenced to UTM WGS84 datum and coordinate system. It is important to express the land use changes by numerical data for an objective point of view. In this study, it is aimed to determine land use condition by using satellite images and to make statistical analysis of changes with time series analysis. Application of the method was carried out in two stages; remote sensing and time series analysis. The first stage is to process the remote sensing data. At this stage, it is studied with 1987, 1997, 2 July 2000, 2001, 3 July 2003, 2005, 5 September 2009 and 8 September 2010 dated ( LANDSAT TM + ETM) eight satellite images. Unsupervised classification, supervised classification and then accuracy assessment processes were applied all the images, respectively. The ISODATA method were used in unsupervised classification process and the maximum likelihood method were used in supervised classification process. Spectral classes were obtained from the images by unsupervised classification process. Afterwards, four information classes based on the first level of CORINE database were determined by supervised classification process. These are artificial surfaces, forest and semi – natural areas, agricultural areas, water bodies. Images are relevant to the classification results can be viewed in the appendix. In this study, the water bodies class disregarded and, the other three classes (forests and semi – natural classes, agricultural areas, and artificial surfaces) were studied in order to random changes in the water bodies class. After the classification process, accuracy assessment were done for all images. With this operation, kappa statistical value and the overall accuracy were calculated. Tables containing these values can be examined in the chapter six. According to this assessment, the lower accuracy with 86.25 % belongs to the 2009 dated image, and the highest accuracy with 91.40 % belongs to the 1987 dated image. It was seen as these values meet the expectations of 80 % accuracy. It was appropriate to continue to the study with these accuracy results. Following the accuracy assessment operation, change detection process were done in order to determine changes between the land use classes. At this stage, image matrices obtained by classification process were compared with each other. Thus, it was determined that the amounts of change and no – change areas. The results of this process can be viewed at tables and figures in the appendix. Figures show the image of change detection process. Also, tables contain statistical information about the change. The second part of this study is to analyze time series. It is aimed to make prediction about the future, at the time series analysis part. For this purpose, the classification results were used, such as used in the change detection process. Before starting the second part of the study, the surface areas of information classes were expressed as percentage for each image in order to eliminate the differences in fields based on geometric correction. Time series were created with the results of the classification process. There are four components that affect the time series. These components may affect time series in together or one by one. It is appears that the trend component is effective for the time series in this study. Changes in information classes likely to occur in future periods were determined by linear trend analysis. The linear model is tt = b₀ + b₁.t for the trend component estimation. The b₀ and b₁ are inferring the estimations of 0 and 1 and they were calculated by solving the least squares normal equations (Σyt = nb₀ + b₁ Σt and Σyt.t = b₀ Σt + b₁ Σt²). For this operation, nine data sets were formed in total with classification results. While data sets were being created, the properties of time series were taken into consideration. In this process, nine trend analysis were performed by using nine time series. Six of the analysis results were considered available to be used for the statistical analysis. For statistical analysis, standard deviation and confidence range were calculated. Standart deviation and confidence interval calculations were performed individually for the years 2009, 2015, 2017, 2021, 2025, 2027, and 2030. Based on the results of the analysis, it is understood that prediction can be made for these dates. Standart deviation and confidence interval calculations were performed separately for the classes of agricultural areas, artificial surfaces, forests and semi – natural areas. Standart deviations were calculated with s2 = [vv] / n equation. In this equation, s represents the standart deviation, and also, n represents the number of samples. The confidence intervals were calculated with A = µ ± Zs x S equation. In this equation, A (A1,A2) represents the symmetrical two sided confidence interval value, µ represents the predictive value, and also S represents the probability value. Zs is a coefficient corresponding to the probability value from the standart normal distribution table. Confidence interval calculations were made for both the probability value of 90 % and % 50. Statistical analysis was performed for the results by the standart deviation and confidence interval calculations. The result of the study can be commented in two stages. Changes between information classes were designated at the first stage, by using remote sensing methods. In the second stage, the results of the linear trend analysis and statistical analysis were construed in common. Thereby, significant results were obtained about the changes in area of the artificial surfaces class, forest and semi – natural areas class, agricultural areas class for 2015, 2017, 2021, 2025, 2027 and 2030 years. Time series analysis method is applied to determine land use changes along with many other scientific fields. In this study, land use change predictions were made by using time series analysis for feature in Sarıyer township. The results of the time series analysis section of the study were evaluated in three categories. The first of these are for between 1987 and 2030 years, the second is for between 1987 and 2010 years, and the third is for between 2010 and 2030 years. It is determined that, the extent of forest and semi – natural areas have a decrease of 30.65 %, artificial surfaces have an increase of 212.46 % and agricultural areas have a decrease of 93.97 % for between 1987 and 2030 years. It is determined that, the extent of forest and semi – natural areas have a decrease of 15.64 %, artificial surfaces have an increase of 32.80 %, and agricultural areas have a decrease of 86,39% for between 2010 and 2030 years. It is determined that, the extent of forest and semi – natural areas have a decrease of 17.79 %, artificial surfaces have an increase of 135.29 %, and agricultural areas have a decrease of 55.69 % for between 1987 and 2010. While the study was generally considered, It was predicted that the extent of forest and semi – natural areas, and agricultural areas were decreased, also, the extent of artificial surfaces were increased in Istanbul, Sarıyer. This study is suitable for development and elaboration. Population data can be obtained for the elaboration of the study. Determining the correlation between the population data and class areas change over years will be useful. For this purpose, it is possible to take advantage of various mathematical or statistical models. Also, one recommendation for improving the study is to examine the effects of possible third bridge to the township. For this idea, to examine the changes in the township caused by the Fatih Sultan Mehmet Bridge will be useful. Similarly, while examining the changes, social and economical evaluations must be considered. Thus, it is posible to gain beneficial results about observing the construction and protection of current state of forest areas for authorized and concerned persons.