Laserfiche WebLink
MEMO 89.2 <br />Table 1. RMS Errors of SPDSS Landsat Imagery <br />Landsat <br />Frame Number of <br />GCPs Horizontal RMS <br />(Pixels) Vertical RMS <br />(Pixels) Total RMS <br />(Pixels) <br />032/032 17 0.2085 0.2155 0.2996 <br />033/032 17 0.2131 0.2151 0.3044 <br />033/033 19 0.2203 0.2207 0.3119 <br />034/032 22 0.3070 0.3678 0.4791 <br />034/033 22 0.3755 0.2917 0.4858 <br />2.2.3 Radiometric Correction <br />Pixel values in commercially available satellites are calibrated to fit a certain range of radiance values of <br />the earth surface in the form of Digital Numbers (DN). Since each sensor has its own calibration <br />parameters used in recording the DN values, the same DN values in two images taken by two different <br />sensors may represent two different radiance values. In order to compare and analyze images taken by <br />different sensors (e.g., Landsat 5 TM versus Landsat 7 ETM) it was necessary to convert the DN values to <br />absolute radiance values. <br />In addition to the instrument calibration, it was necessary to normalize image pixel values for differences <br />in sun illumination geometry and atmospheric effects. This improved the ability to mosaic adjacent <br />imagery and compare imagery over time (e.g., multi-date classification). Because some procedures <br />involve collecting in-situ atmospheric measurements and radiometric transfer code (RTC), they are very <br />difficult and expensive to implement. <br />Without calibrating satellite imagery into radiance or normalizing pixel values, the ability to mosaic <br />adjacent imagery and compare imagery over time would be diminished by the introduction of variability <br />within and between the imagery that was unrelated to the variability found on the ground. Crop <br />classifications using imagery that have not taken these factors into consideration during processing are <br />directed toward less than optimum results (Jensen, 1996). <br />The radiance conversion and pixel normalization procedure described here is image-based and requires no <br />additional information other than the imagery itself. First, DN values were converted to at-sensor <br />radiance and at-sensor reflectance while normalizing the solar elevation angle. In addition, animage- <br />based atmospheric correction was applied using the COST method developed by Chavez (1996). The <br />equation is as follows: <br />RN = DN_N *GainN +BiasN) - (DON *GainN + Bias_N)) * ~ * D~ <br />EsuN ~ (cos ((90 - e) ~ ~i iso~ <br />Where, <br />RN = Reflectance for Band N <br />DNN = Digital Number for Band N <br />DON = Digital Number representing Dark Object for Band N <br />D = Normalized Earth-Sun Distance <br />ESUN = Solar Irradiance for Band N <br />0 = Solar elevation angle <br />GainN +BiasN = Gain and bias constants for Landsat 7 and Landsat 5 described <br />correspondingly in Huang et al. (2002) based on Markham and Barker <br />(1986). <br />Page 4 of 45 ~R~versfde TecAnotogy, fnc. <br />4'JaYCr Resources Errgi~ecr:np an~i CansaFlrnp <br />