Uydu verilerinden harita kapsamında görüntü üretmenin geometrik doğruluğu

Abstract

Uzaktan algılama gtinUmUzde birçok bilim dalları için önemli bir araç olmaktadır. Son yıllarda özellikle uydu verilerinden elde edilen bilgilerin görUntU işleme tekniklerine bağlı olarak kullanılması her alanda teknolojik bir yenilik getirmiştir. GünUmUzde çok sayıda uydu çeşitli amaçlarla faaliyetlerini sürdürmektedir. YerylizU kaynaklarını incelemek için fırlatılan LANDSAT ve SPOT uydularından veri toplanmaktadır. Bu veriler islenerek çeşitli amaçlar için kullanılmaktadır. LANDSAT-5 deki Tematik Haritalayıcı ile yüksek geometrik doğruluk elde edilmiştir, özellikle pankromatik modda SPOT' tan yüksek çözmesi nedeniyle ktiçllk ve orta Ölçekli topografik haritaların yapımında yararlanılabileceği ortaya çıkmıştır. Her iki uydudan toplanan verilerin 1:25000, 1:50000 ve 1:100000 ölçekli haritalara uygulanması veya özellikle bu haritaların güncelleştirilmesinde kullanılmasına gereksinim duyulmaktadır. Ancak bu verilerin söz konusu gereksinime cevap verebilmesi için su aşamada yetersiz olduğu savunulmaktadır [Bahr,1978; Borgeson, 1985]. Bu çalışmada uzaktan algılamanın etkin olduğu bir konu olan harita güncelleştirilmesi ve Üretimi uygulamalarında harita amaçlı uydulardan elde edilen verilerin islenerek harita kapsamında görüntü elde etme amacına yönelik geometrik ve kartografik doğruluğu araştırılmıştır. Bu amaçla öncelikle uzaktan algılamanın fiziksel prensipleri ele alınarak enerjinin atmosfer içerisinde yeryüzü objeleri ile etkileşimi konusunda özet bilgiler verilmiştir. Bunu takip eden bölümde haritalama amaçlı uydular tanıtılarak bunların algılama sistemleri ve özellikleri hakkında ayrıntılı bilgiler yer almaktadır. Uzaktan algılama verilerinin geometrik özellikleri vurgulanarak geometrik bozulmalar ve düzeltilmesi esaslarından bahsedilmiştir. Bu konuda, ilk jenerasyon uydu verilerinden başlanarak günümüz uydu verileri kullanılarak yapılan bir dizi uluslararası araştırma özetlenmiş ve bulunan sonuçlar sunulmuştur. Bu çalışmaların doğrultusunda, istanbul boğazı ve çevresini kapsayan LANDSAT-5 TM görüntü dataları kullanılarak düşeye çevirme doğruluğunun araştırılması amacıyla benzer bir uygulama yapılmış ve elde edilen bulgular sonuç bölümünde irdelenmiştir. Ayrıca İstanbul Adalar örneğinde uydu verileri kullanılarak alan karşılaştırması amacıyla bir uygulama yapılmış ve sonuçları verilmiştir.

The space era began three decades ago with the launch of Sputnik I, the first artificial Earth satellite, by the Soviet Union in October 1957. Within a short time span history unfolded numerous exciting events, including the first Earth orbital flight of man. Then in July 1969, under the insightful eye of the television camera, the world viewed man landing on the surface of another planetary body-the moon. By 1990, the international space community had launched more than 3000 satellites from different sites around the Earth. NASA's LANDSAT remote sensing satellites began in 1972 followed by SPOT (France, 1986), MOS-1 (Japan, 1987), and others. Each of these has contributed toward enhancing the value of remote sensing data for medium and small scale topographic mapping and for solving Earth's resource problems [Barrett, 19821. Cartography is generally seen as the science, technology and the art of making maps and map related representations, In cartography, remotely sensed images are being increasingly used, in two main ways : a) as data sources for interpretation and measurement, b) in photo and image mapping [MUller, 1991]. In the mapping sciences, the term 'remote sensing' normally refers to the detection of distant objects due to the electromagnetic radiation emitted or reflected by these objects and recorded in some way. In other words, remote sensing is the measurement or acquisition of data about an object or scene by a satellite or other instruments above or far from the object without coming into direct contact with them. Remote sensing applications in generally consists of two steps; firstly that data acquisition which use of different remote sensors and secondly data analysis as computer image processing, classifications and interpret the data [Brown, 1987]. In 1972, NASA (National Aeronautics and Space Administration) initiated the first program specializing in the acquisition of remotely sensed digital satellite data. The previous LANDSAT satellite system is known as the ERTS (Earth Resources Technology Satellite) and later renamed LANDSAT-1. It was equipped with a RBV (Return Beam Midicon) television camera system and a four waveband Multispectral Scanning System (MSS). xi LANDSAT-1,2,3 are called first generation and LANDSAT-4,5 are called second generation LANDSAT series. LANDSAT-5 was launched in March, 1984 with the TM (Thematic Mapper) sensor on board. In common with the previous satellites, LANDSAT-5 also has a four channel MSS. The MSS is a line array sensor device which uses an ascillating mirror to scan at right angles to the satellite flight directions. LANDSAT MSS scene covers an area 185 km x 185 km, consisting of 2340 scan lines with about 3240 pixel per line. The spatial resolution of MSS data is 57m x 79m with a 79m x 79m IFOV (Instantaneous-Field-of-View). The TM scanner is a multispectral scanning system much like the MSS, but it hass higher spatial, spectral and radiometric resolution than MSS. The TM acquires data in seven spectral bands. Four of these bands are located in portions of the spectrum not sensed by the MSS. The TM instantaneous-f ield-of view (IFOV) is 30m x 30m for the six reflective spectral bands (bands TM1 to TM5 and TM7) and 120m x 120m for the thermal band (band TM6). The band designations, spectral ranges and principal applications of TM are as follows : Band Spectral Number Range \xrn Applications 1 0.45 to 0.52 Sensitive to chlorophyll and cortonoid concentrations for soil/vegetation differentiation, deciduous/coniferous differentiation. Coastal water mapping. 2 0.52 to 0.60 Sensitive to green reflectance by healthy vegetation. 3 0.63 to 0.69 Sensitive to chlorophyll absorbtion for plant species differentiation. 4 0.76 to 0.90 Sensitive to near infra-red reflectance of healthy vegetation for biomass surveys. 5 1.55 to 1.75 Sensitive to vegetation moisture and snow/cloud reflectance differences. 6 10.40 to 12.50 Thermal mapping. 7 2.08 to 2.35 Sensitive to vegetation moisture to hydroxyl ions in minerals for geological mapping. xii The French satellite system SPOT (Systeme Probatoire de 1 'Observation de la Terre), which translated literally means the Earth Observation Test System is the first Earth resources satellite launched from Europe. The SPOT system was developed by the French Centre National d'Etudes Spatials (CNES) with participation from both Belgium and Sweden [Doyle, 1982; Gugan.1987]. SPOT carries two identical pushbroom scanners, these are called HRV (High Resolution Visible) scanners and used to record in either panchromatic or multispectral mode (XS). When in panchromatic mode all of the detectors are sampled, giving a spatial resolution of around 10m and when in multispectral mode only half of the detectors are sampled, giving a spatial resolution of around 20m [Curran, 1989]. The properties of this two operational modes compared below : Panchromatic mode Multispectral mode Waveband name Waveband width Waveband name Waveband width visible 0.51 - 0.73 \xm green 0.50 - 0.59 urn red 0.61 - 0.69 p./7? near infrared 0.79 - 0.89 \im The terms accuracy, precision, resolution and scale are used almost interchangeably in reference to spatial data. Accuracy refers to the relationship between a measurement and the reality which it purports to represent. Precision refers to the degree of detail in the reporting of a measurement or in the manipulation of a measurement in arithmetic calculations. Finally, the resolution of a data set defines the smallest object or feature which is included or is discernable in the data. Scale and resolution are intimately related because there is a lower limit to the size of an object which can be usefully shown on a paper map. This limit is often assumed to be 0.5 mm as a rule of thumb, so the effective resolution of a 1:1000 map is about 1000 times 0.5 mm, or 50 cm, although the standards of most mapping agencies are subtantially better. The effective resolutions of some common map scales, assuming 0.5 mm resolution are as follows ; xm Scale Effective resolution 62.5 cm 5 m 12.5 m 25 m 50 m 125 m Images such as those derived from remote sensing are composed of pixels of uniform size and shape, with associated measures of reflected or emitted radiation. After classification, each pixel is assigned to one of a number of classes, often of land use or land cover. Unlike the cartographic case, there are no objects or features to be located, and accuracy is simply a function of the errors in the assignment of classes to each pixel. The standard method of measuring accuracy in a classified image is to compare the classes assigned to a sample of pixels to the true classes on the ground ('ground truth') and express the result in the form of a table, the rows representing the assigned class and the columns the ground truth. Pixels which fall on the diagonal of the table are correctly classified; pixels which fall off the diagonal in a column are termed 'errors of omission' since the true value is omitted from the assigned class; and pixels which fall off the diagonal in a row are termed 'errors of commission' since they appear as false occurenHxs of a given class in the data. The contents of the table can be summarized in statistics such as the percentage of cells correctly classified [Bahr,1991; MUller,1991]. XIV The quantitative use of remote sensing satellite images in many applications requires that the geometric distorsion inherent in these images be corrected, or rectified, to a desired map projection, the most widely used technique relies on ground control points to empricially determine a mathematical coordinate transformation to correct the geometry [Ford, 1985]. In this study, using the method of least squares, accuracy of the geometric transformation and of the rectification of the satellite image to a map projection are investigated. Explicit relations between the global accuracy of the transformation and the number, location, and local accuracy of the ground control points are obtained. The results are applied to the correction of a LANDSAT TM image including the Bosphorus region, Istanbul. Image is rectified to the Universal Transverse Mercator Projection. In the least squares transformation approach, the image distorsion is modeled empirically as a mapping transformation from the map projection coordinates to the acquired image coordinates. The mapping function is generally chosen to be a polynomial, first employed by Markarian [Markarian, 1973]. The transformation coefficients depending on the degree of the polynomial are determined from a set of GCPs, which are physical features that can be correctly located in the image and on the 1:25000 scale map. Typical GCPs are highway intersections, land-water interfaces, or field boundries. The geodetic rectification of TM image data involves the following steps : - Location of GCPs in the image (pixel and line coordinates) and on 1:25000 scale map (UTM coordinates) of the study area; - Checks to eliminate points of questionable reliability; - Least square solution of polynomial rectification equations using GCPs scattered throughout the data set to determine the coefficients which must be applied to the image coordinates in order to derive UTM map coordinates. Coordinates of the GCPs in image and map space have been checked to identify suspect points. The procedure for this is referred to as a point-pair distance check between all possible combinations of point-pairs. Then, it has been investigated, the optimum number of GCPs required and the minimum degree of the polynomials which give the minimum RMSxy value, used in the rectification process.