اثر خطای زمین مرجع نمودن بر روی بازتابندگی در تصاویر با قدرت تفکیک مکانی پایین در مناطق شهری

نویسندگان

چکیده

زمین مرجع نمودن تصاویر ماهواره‌ای یکی از مراحل ضروری و اولیه در بسیاری از پردازش‌های سنجش از دوری است. این موضوع در تصاویر با قدرت تفکیک مکانی پایین، به دلیل دشواری در تعیین دقیق نقاط کنترل، نسبت به تصاویر با قدرت تفکیک مکانی بالا از دقت کمتری برخوردار است. هنگام استفاده از این تصاویر در مناطق شهری به دلیل پیچیدگی سطح شهر و وجود کلاس‌های مختلف در هر پیکسل (وجود پیکسل‌های مختلط) زمین مرجع نمودن کاری دشوار خواهد بود و دارای خطا است. وجود خطا در این حالت باعث جابجایی پیکسل‌ها نسبت به موقعیت واقعی شده، که این موضوع ایجاد خطا در مقادیر بازتابندگی پیکسل‌ها را در پی خواهد شد. در این پژوهش سعی در بررسی تاثیر این خطا بر روی بازتابندگی پیکسل‌ها نسبت به موقعیت دقیق آنها گردیده است. برای این منظور ابتدا با استفاده از تصاویر با قدرت تفکیک مکانی بالا و همچنین قدرت تفکیک طیفی بالا، نقشه کلاس‌های سطح شهر و طیف بازتابندگی آنها بدست آمد. سپس با شبیه سازی وجود خطا در زمین مرجع کردن تصاویر 500 متری سنجنده MODISمیزان تغییرات بازتابندگی و خطای نسبی آن در موقعیت‌های جدید و برای باندهای مختلف بدست آمد. از نتایج بدست آمده مشخص گردید اثر این خطا در مناطق شهری با بافت تکراری برای کلیه باندها و تا میزان 5/0 پیکسل جابجایی کمتر از 10 درصد و برای مناطق دارای تغییر بافت بین 10 تا 50 درصد است که میزان خطا با افزایش فاصله از موقعیت صحیح پیکسل افزایش می‌یابد.

کلیدواژه‌ها


عنوان مقاله [English]

Effect of georeferencing errors on reflectance of low spatial resolution images in urban areas

نویسندگان [English]

  • A. Ahmadian
  • M.R. Mobasheri
  • A.A. Matkan
چکیده [English]

1. Introduction
Georeferencing of the satellite images is of the initial steps in most remote sensing processing. Due to difficulties of control point’s selection in low spatial resolution images, georeferencing accuracy is lower. In urban areas, due to the complexity of the city (different classes in each pixel), georeferencing will be difficult and has more errors. This error makes changes on pixel location against the real that cause reflectance changes.
This study concerns the effect of this error on the reflectance of pixels. Using high spatial and high spectral resolution images, the urban areas were first classified and reflection spectrum of classes was obtained. Then, reflectance variations of pixels and their errors for new locations were defined by simulating the error in 500m images of MODIS for different bands.
In order to determine that point, the map of different classes of the city and their reflectance are needed. The usual method for description of the urban areas by remote sensing is the classification and also use of spatial unmixing model. The researches show that in urban areas, the measured reflectance by sensor is linear combination of reflectance of materials in instantaneous field of view. Therefore, for urban areas, it is possible to use a linear spectral mixture analysis.

2. Methodology
Since the investigation of georeferencing’s error of low spatial resolution images on the pixels reflectance in urban areas, is the main objective of this study, the city of Tehran was selected as a case study. The used data consist of the ground data and also satellite images of Geoeye,Hyperion and MODIS.
In the first step, classification map of the cityis produced by using the high spatial resolution images. Then, the contribution of each class in pixels is determinedby overlying a low spatial resolution image with classification map. In order to investigate the georeferencing’s error on the value of pixel’s reflectance, the location of pixel is changed and after calculation of new class contributions, the pixel’s reflectance is again calculated. By comparison of correct reflectance and reflectance of new locations, it is possible to determine the effect of displacement on the pixel’s reflectance.



2.1. Generating the map of classes
The classification map of city was generated using the Geoeye images. This map was obtained via supervised classified methods, by use of large scale maps and some field data (determination of classesin some locations in city). In this study, four main classes of vegetation, soil, inscrutable surfaces and water were considered. Due to the variety of vegetation, soil and inscrutable surfaces, these classes were divided to the three sub-classes.

2.2. Calculation of classes’ reflectance
The selected method is based on the variations of pixels’ reflectance.The percentage of the classes along with their reflectance is necessary for calculation of the reflectance in a special pixel which was used by Hyperion images.In a location of the city where Hyperion images were available, the percentage of different classes was determined by overlying the image and map of classes. Then, the reflectance of each class was obtained in different bands by use of linear spectral mixture analysis and reflectance of each pixel.

2.3. Simulation of pixels’ reflectance
In this stage, an assumptive image of georeferencing with 500 m pixels was selected. By assuming that the pixel location of the mentioned image is accurate, the current location of the pixels was considered as correctone. By overlying the image and map of classes, the percentage of different classes in each pixel was determined. After that, by use of classes’ reflectance and linear spectral mixture analysis, the reflectance is simulated in each band. That reflectance is considered as correct one for each pixel. Now, if the georeferencing of the image has error and the location of the pixels varies, the reflectance of the pixels will change. So, it is necessary to calculate that error.

2.4. Simulation of georefrencing errors
In order to investigate the effect of georeferencing on the reflectance, the displacement of pixels will be considered. In the case of errors in georeferencing, the change of one pixel in different directions is possible. The displacement of the pixels in different directions is done with 20 m distances until one pixel. To investigate the changes of pixels’ reflectance, the values of relative error was calculated in different distances. In this regard, the relative error for all pixels was calculated respect to the real location of pixel. Since the assumption was on the use of MODIS 500 images, the mentioned process was carried out separately for 7 same bands and relative errors for each band was obtained.

3. Discussion
Generally, in urban areas, the available classes through the city havemajor variations which lead to the complexity of the image of city. For high spatial resolution images, due to the small dimensions of pixels, the variations of reflectance are noticeablefrom one pixel to the other. In contrast, for low spatial resolution images, due to the large dimensions of pixels,the variation is not major. That point is due to the uniform urban areas which hold the same percentage of each class in the pixels.
By changing pixel in different directions respect to the main location, the reflectance will change. For the uniform urban areas, the variations of reflectance are not major and the value of corresponding error is small. However, for the areas in which the classes suddenly change, the variations are high which increase the error. In this case, for the regions near to the border of variations, the error will increase by distancing from the correct location. Moreover, in some stages of the analysis, the small dimensions were considered for the pixels and the effect of georeferencing was again investigated.By reduction of pixels’ dimensions and due to the increase of variations in the classes, the changes of the reflectance increase, resulting in the increase of error. So, it can be concluded that the increase of spatial resolution increases the error on the reflectance of pixels.

4. Conclusion
Investigation of the results in 500 m pixels of MODIS in uniform urban areas showed that in the case of georeferencing’s error,the variation of reflectance respect to the real value is not major and can be neglected. That is due to the repeat of classes and approximate constancy of the percentage of each class in pixels. For example, the error value for a displacement less than 0.5 pixeldisplacement is near to 10%. But for the regions where the variations of the classes were high and the city texture suddenly changes, the variations of reflectance are high, resulting in the high relative error. For example, the error value for a displacement near to 0.5 pixeldisplacement is 30-40%. The atmospheric error is another error which can be important during the use of low spatial resolution imagesin urban areas. So, for the use of these images, the researchers recommend paying more attention and modifying the atmospheric images than the increase of georeferencing’s accuracy.

کلیدواژه‌ها [English]

  • Georeferencing
  • Reflectance
  • Spatial resolution
  • Classification