PRE-PROCESSING OF REMOTE SENSING DATA

1|Page

INTRODUCTION What is pre-processing? Pre-processing can be defined as the act of processing data beforehand. The data is analysed and appropriate steps are taken before the data is processed. Therefore, preprocessing of remote sensing data involves processing remote sensing data i.e. imageries before further analysis and information extraction is done. In pre-processing, corrections are applied to the remote sensing data. TYPES OF CORRECTION Pre-processing is grouped into two: Radiometric correction Geometric correction RADIOMETRIC CORRECTION Radiometric correction involves correcting data for sensor irregularities and unwanted sensor and atmospheric noise. This data is then converted so they accurately represent the reflected or emitted radiation measured by the sensor. It also involves the re-arrangement of the DN (Digital Number) in an image. The digital number is a number showing the degree of brightness in an image. This is done in order for all the areas of an image to have the same linear relationship between the DN and either radiance or back-scatter. Radiometric distortions are introduced by the atmosphere between the surface and the sensor. Scattering in the atmosphere causes fine detail in image data to be obscured, and the effect is larger at the edges of the swath (the path cut by a single sweep of the satellite). Scattering depends on wavelength and is also a function of relative humidity, atmospheric pressure, temperature and visibility (a measure of the concentration of larger particles or aerosols in the atmosphere). Radiometric correction is concerned with improving the accuracy of surface spectral reflectance, emittance or back- scattered measurements obtained using a remote sensing system. Brightness values inconsistencies caused by the sensors and environmental noise factors are balanced or normalized across and between image coverage and spectral bands. There are five primary objectives for applying radiometric corrections to digital remotely sensed data. Four of these reasons pertain to achieving consistency in relative image
2|Page

brightness and one involves absolute quantification of brightness values. Relative correspondence of image brightness magnitudes may be desired for pixels: (1) within a single image (e.g., orbit segment or image frame), (2) between images (e.g. adjacent, overlapping frames), (3) between spectral band images, and (4) between image dates. Before listing the types of radiometric errors, it is essential to mention the error sources. Error sources include: Internal errors: This kind of errors are introduced by the remote sensing

system. They are generally systematic (predictable) and may be identified and then corrected based on pre-launch or in-flight calibration measurements. For example, n-line stripping in imagery may be caused by a single uncalibrated sensor. In many instances, radiometric correction adjusts for detector miscalibration. External errors: They are introduced by phenomena that vary in nature

through space and time. These include atmosphere, terrain elevation and slope. Some external errors may be corrected by relating empirical ground observations i.e. radiometric or geometric ground control points, to sensor measurements.

Types of radiometric errors Types of radiometric errors include: Sensor error (Internal error) Atmospheric error (External error) Topographic error (External error)

Correcting sensor error Ideally, the radiance recorded by a remote sensing system in various bands is an accurate representation of the radiance actually leaving the feature of interest (e.g., soil, vegetation, atmosphere, water, or urban land cover) on the Earths surface or atmosphere. Unfortunately, noise (error) can enter the data acquisition system at several points. For example, radiometric error in remotely sensed data may be introduced by the sensor system itself when the individual detectors do not function properly or are improperly calibrated. Several of the more common remote sensing systeminduced radiometric errors include:

Random bad pixels (Shot noise) Line start/stop problems

3|Page

Line or column drop-outs Partial line or column drop outs Line or column stripping

Random bad pixels (Shot noise): Sometimes an individual detector does not record spectral data for an individual pixel. When this occurs randomly, it is called a bad pixel. When there are numerous random bad pixels found within the scene, it is called shot noise because it appears that the image was shot by a shotgun. Normally these bad pixels contain values of 0 or 255 (in 8-bit data) in one or more of the bands. Shot noise is identified and repaired using the following methodology. It is first necessary to locate each bad pixel in the band k dataset. A simple thresholding algorithm makes a pass through the dataset and flags any pixel (BVi,j,k) having a brightness value of zero (assuming values of 0 represent shot noise and not a real land cover such as water). Once identified, it is then possible to evaluate the eight pixels surrounding the flagged pixel, as shown below:

BVi,j,k = int BVi

i=1

The above mathematical equation makes it possible to evaluate the eight pixels surrounding the bad pixel. This formula is used after the bad pixel must have been flagged i.e. noted.
4|Page

a) Landsat Thematic Mapper band 7 (2.08 2.35 m) image of the Santee Delta in South Carolina. One of the 16 detectors exhibits serious striping and the absence of brightness values at pixel locations along a scan line. b) An enlarged view of the bad
5|Page

pixels with the brightness values of the eight surrounding pixels annotated. c) The brightness values of the bad pixels after shot noise removal. This image was not destriped.

Line start/stop problems: Occasionally, scanning systems fail to collect data at the beginning or end of a scan line, or they place the pixel data at inappropriate locations along the scan line. For example, all of the pixels in a scan line might be systematically shifted just one pixel to the right. This is called a line start problem. Also, a detector may abruptly stop collecting data somewhere along a scan and produce results similar to the line or column drop-out previously discussed. Ideally, when data are not collected, the sensor system would be programmed to remember what was not collected and place any good data in their proper geometric locations along the scan. Unfortunately, this is not always the case. For example, the first pixel (column 1) in band k on line i (i.e.,BV1,i,k) might be improperly located at column 50 (i.e.,BV50,i,k). If the line-start problem is always associated with a horizontal bias of 50 columns, it can be corrected using a simple horizontal adjustment. However, if the amount of the line-start displacement is random, it is difficult to restore the data without extensive human interaction on a line-by-line basis. A considerable amount of MSS data collected by Landsats2 and 3 exhibits line-start problems.

Infrared imagery of the Four Mile Creek thermal effluent plume entering the Savannah River
6|Page

Line or column drop-outs: An entire line containing no spectral information may be produced if an individual detector in a scanning system (e.g., Landsat MSS or Landsat7 ETM+) fails to function properly. If a detector in a linear array (e.g., SPOT XS, IRS-1C, QuickBird) fails to function, this can result in an entire column of data with no spectral information. The bad line or column is commonly called a line or column drop-out and contains brightness values equal to zero. For example, if one of the 16 detectors in the Landsat Thematic Mapper sensor system fails to function during scanning, this can result in a brightness value of zero for every pixel, j, in a particular line, i. This line drop-out would appear as a completely black line in the band, k, of imagery. This is a serious condition because there is no way to restore data that were never acquired. However, it is possible to improve the visual interpretability of the data by introducing estimated brightness values for each bad scan line. It is first necessary to locate each bad line in the dataset. A simple thresholding algorithm makes a pass through the dataset and flags any scan line having a mean brightness value at or near zero. Once identified, it is then possible to evaluate the output for a pixel in the preceding line (BVi1, j, k) and succeeding line (BV i+1, j, k) and assign the output pixel (BVi, j,
k)

in the drop-out line the average of these two brightness values

BVi, j, k = int BVi1, j, k + BV i+1, j, k 2

Partial line or column drop outs: This is similar to line or column drop outs but, in this case only a portion of the line or column is affected.

Line or column stripping: Sometimes a detector does not fail completely, but simply goes out of radiometric adjustment. For example, a detector might record spectral measurements over a dark, deep body of water that are almost uniformly 20 brightness values greater than the other detectors for the same band. The result would be an image with systematic, noticeable lines that are brighter than adjacent lines. This is referred to as n-line stripping. The maladjusted line contains valuable information, but should be corrected to have approximately the same
7|Page

radiometric scale as the data collected by the properly calibrated detectors associated with the same band. To repair systematic n-line stripping, it is first necessary to identify the miscalibrated scan lines in the scene. This is usually accomplished by computing a histogram of the values for each of the n detectors that collected data over the entire scene (ideally, this would take place over a homogeneous area, such as a body of water). If one detectors mean or median is significantly different from the others, it is probable that this detector is out of adjustment. Consequently, every line and pixel in the scene recorded by the maladjusted detector may require a bias (additive or subtractive) correction or a more severe gain (multiplicative) correction. This type of n-line stripping correction a) adjusts all the bad scan lines so that they have approximately the same radiometric scale as the correctly collected data and b) improves the visual interpretability of the data. It looks better. To repair non-systematic stripping, there is no easy way.

Correcting atmospheric error There are several ways to atmospherically correct remotely sensed data. Some are relatively straightforward while others are complex, being founded on physical principles and requiring a significant amount of information to function properly. This discussion will focus on two major types of atmospheric correction: Absolute atmospheric correction Relative atmospheric correction

Absolute atmospheric correction: Solar radiation is largely unaffected as it travels through the vacuum of space. When it interacts with the Earths atmosphere, however, it is selectively scattered and absorbed. The sum of these two forms of energy loss is called atmospheric attenuation. Atmospheric attenuation may 1) make it difficult to relate hand-held in situ spectroradiometer measurements with remote measurements, 2) make it difficult to extend spectral signatures through space and time, and (3) have an impact on classification accuracy within a scene if atmospheric attenuation varies significantly throughout the image. The general goal of absolute radiometric correction is to turn the digital brightness values (or DN) recorded by a remote sensing system into scaled surface reflectance values. These values can then be compared or used in conjunction with scaled surface reflectance values obtained anywhere else on the planet.
8|Page

Radiative transfer-based atmospheric correction algorithms Much research has been carried out to address the problem of correcting images for atmospheric effects. These efforts have resulted in a number of atmospheric radiative transfer codes (models) that can provide realistic estimates of the effects of atmospheric scattering and absorption on satellite imagery. Once these effects have been identified for a specific date of imagery, each band and/or pixel in the scene can be adjusted to remove the effects of scattering and/or absorption. The image is then considered to be atmospherically corrected. Unfortunately, the application of these codes to a specific scene and date also requires knowledge of both the sensor spectral profile and the atmospheric properties at the same time. Atmospheric properties are difficult to acquire even when planned. For most historic satellite data, they are not available. Even today, accurate scaled surface reflectance retrieval is not operational for the majority of satellite image sources used for land-cover change detection. An exception is NASA's Moderate Resolution Imaging Spectroradiometer (MODIS), for which surface reflectance products are available. Most current radiative transfer-based atmospheric correction algorithms can compute much of the required information if a) the user provides fundamental atmospheric characteristic information to the program or b) certain atmospheric absorption bands are present in the remote sensing dataset. For example, most radiative transfer-based atmospheric correction algorithms require that the user provide:

latitude and longitude of the remotely sensed image scene date and exact time of remote sensing data collection image acquisition altitude (e.g., 20 km AGL) mean elevation of the scene (e.g., 200 m ASL) an atmospheric model (e.g., mid-latitude summer, mid-latitude winter, tropical) radiometrically calibrated image radiance data (i.e., data must be in the form W m2 mm-1 sr-1)

data about each specific band (i.e., its mean and full-width at half-maximum (FWHM) local atmospheric visibility at the time of remote sensing data collection (e.g., 10 km, obtained from a nearby airport if possible).

9|Page

These parameters are then input to the atmospheric model selected (e.g., mid-latitude summer) and used to compute the absorption and scattering characteristics of the atmosphere at the instance of remote sensing data collection. These atmospheric characteristics are then used to invert the remote sensing radiance to scaled surface reflectance. Many of these atmospheric correction programs derive the scattering and absorption information they require from robust atmosphere radiative transfer code such as MODTRAN 4+ or Second Simulation of the Satellite Signal in the Solar Spectrum (6S). Examples of these atmospheric correction programs include ACORN, ATCOR, ATREM, FLASH etc.

a) Image containing substantial haze prior to atmospheric correction. b) Image after atmospheric correction using ATCOR (Courtesy Leica Geosystems and DLR, the German Aerospace Centre).

Empirical line calibration Absolute atmospheric correction may also be performed using empirical line calibration (ELC),which forces the remote sensing image data to match in situ spectral reflectance measurements, hopefully obtained at approximately the same time and on the same date as the remote sensing overflight. Empirical line calibration is based on the equation: Reflectance (field spectrum) = gain x radiance (image) + offset

Relative atmospheric correction:

10 | P a g e

Relative atmospheric correction is done when the required data is not available for absolute atmospheric. Relative atmospheric correction may be used for Single-image normalization using histogram adjustment. Multiple-data image normalization using regression.

Single-image normalization using histogram adjustment : The method is based on the fact that infrared data (>0.7 m) is free of atmospheric scattering effects, whereas the visible region (0.4-0.7 m) is strongly influenced by them. Use Dark Subtract to apply atmospheric scattering corrections to the image data. The digital number to subtract from each band can be either the band minimum, an average based upon a user defined region of interest, or a specific value.

Multiple-data image normalization using regression: This involves selecting a base image and then transforming the spectral characteristics of all other images obtained on different dates to have approximately the same radiometric scale as the based image. Selecting pseudo-invariant features (PIFs) or region (points) of interest is important. Important things to note include: Spectral characteristic of PIFs change very little through time (deep water body, bare soil, rooftop). PIFs should be in the same elevation as others. No or rare vegetation. The PIF must be relatively flat

After this, the PIFs will be used to normalize the multiple-date imagery.

Other relative atmospheric correction methods include: Flat field calibration: This is used to normalize images to an area of known "flat" reflectance. This is particularly effective for reducing hyperspectral data to relative reflectance. The method requires that you select a Region of Interest (ROI) prior to execution. The average spectrum from the ROI is used as the reference spectrum, which is then divided into the spectrum at each pixel of the image. IAR (Internal Average Relative) Reflectance calibration:
11 | P a g e

This is used to normalize images to a scenes average spectrum. This is particularly effective for reducing hyperspectral data to relative reflectance in an area where no ground measurements exist and little is known about the scene. It works best for arid areas with no vegetation. An average spectrum is calculated from the entire scene and is used as the reference spectrum, which is then divided into the spectrum at each pixel of the image.

Correcting topographic error Topographic slope and aspect also introduce radiometric distortion (for example, areas in shadow).The goal of a slope-aspect correction is to remove topographically induced illumination variation so that two objects having the same reflectance properties show the same brightness value (or DN) in the image despite their different orientation to the Suns position. Image acquisition geometry

Sun zenith angle is the angle of the Sun away from vertical. Sun elevation angle is the angle of the Sun away from horizontal. Sensor elevation angle is the angle away from horizontal. Sensor azimuth angle and Sun azimuth are clockwise from the north

12 | P a g e

Cosine correction: LH = LT cos0 cos i Minnaert correction: LH = LT cos0 k cos i

13 | P a g e

Statistical-empirical correction: LH = LT - m cos i b + LT C correction: LH = LT cos0 + c cos i + c Computing the cosine of the solar incidence angle cos (i) = sin()sin()cos(s) sin()cos()sin(s)cos() + cos()cos()cos(s)cos() + cos()sin()sin(s)cos()cos() + cos()sin()sin(s)sin() where = declination of the earth (positive in summer in northern hemisphere) = latitude of the pixel (positive for northern hemisphere) s = slope in radians, where s=0 is horizontal and s=/2 is vertical downward (s is always positive and represents a downward slope in any direction) = surface azimuth angle. is the deviation of the normal to the surface from the local meridian, where = 0 for aspect that is due south, = -for east and = + for western aspect. = -/2 represents an east-facing slope and = +/2 represents a west-facing slope. = -or = represents a north-facing slope. = hour angle. = 0 at solar noon, is negative in morning and is positive in afternoon

14 | P a g e

(a) Original image (b) Result of cosine correction (c) Result of Minnaert correction (d) Result of 2-stage normalization

GEOMETRIC CORRECTION Geometric error in the image arises from several sources. These include the curvature and rotation of the earth, the wide field of view and platform instability (both of which are bigger problems for airborne sensors than for satellite instruments), and panoramic effects of scanning instruments. Radar data is affected by the relationship between terrain slope and look angle. While the theory behind the correction of geometric distortions is usually straightforward, its implementation may not be. One problem is registering the image to a rectified grid. (The same problem arises when two or more images or maps from different sources are overlain, another common preliminary to data analysis). Polynomial
15 | P a g e

interpolation methods are satisfactory for single-band images and have been used for multispectral broadband data as well. For hyperspectral data, however, simpler nearestneighbour resampling schemes may be preferred because these do not distort spectral characteristics which will be used for spectral matching and detailed identification of objects in the scene. Detailed geometric correction models are developed for instruments with high spatial resolution. Some of these errors can be corrected by using ephemeris of the platform and known internal sensor distortion characteristics. Other errors can only be corrected by matching image coordinates of physical features recorded by the image to the geographic coordinates of the same features collected from a map or global positioning system (GPS). Geometric errors that can be corrected using sensor characteristics and ephemeris data include scan skew, mirror-scan velocity variance, panoramic distortion, platform velocity, and perspective geometry.

Geometric errors Remote sensing data is affected by geometric distortions due to sensor geometry, scanner and platform instabilities, earth rotation, earth curvature, scan skew, mirror scan velocity variance, panoramic distortion, platform velocity, perspective etc. A few of these errors shall be discussed.

Scan skew: This is caused by the forward motion of the platform during the time required for each mirror sweep. The ground swath is not normal to the ground track but is slightly skewed, producing cross-scan geometric distortion.

Mirror scan velocity variance: In this case, the mirror scanning rate is usually not constant across a given scan, producing along-scan geometric distortion.

16 | P a g e

Panoramic distortion: Here, the ground area imaged is proportional to the tangent of the scan angle rather than to the angle itself. Because data are sampled at regular intervals, this produces along-scan distortion

Platform velocity: If the speed of the platform changes, the ground track covered by successive mirror scans changes, producing along -track scale distortion.

Earth rotation: Earth rotates as the sensor scans the terrain. This results in a shift of the ground swath being scanned, causing along-scan distortion.

Perspective: For some applications it is desirable to have images represent the projection of points on Earth on a plane tangent to Earth with all projection lines normal to the plan. This introduces along -scan distortion.

Some of these distortions are corrected by the image supplier and others can be corrected by referencing the images to existing maps. Remotely sensed images in a raw format contains no reference to the location. In order to integrate these data with other data in a GIS, it is necessary to correct and adapt them geometrically so that they have comparable resolution and projections as the other data sets. The geometry of a satellite image can be distorted with respect to a North-South oriented map: Heading of the satellite orbit at a given position on earth (rotation). Change in resolution of the input image (scaling). Difference in position of the image and map (shift). Skew caused by earth rotation (shear).

The different distortions of the image geometry are not realized in certain

17 | P a g e

sequence but happen altogether and therefore cannot be corrected stepwise. The correction of all distortions at once is executed by a transformation which combines all the separate corrections. The transformation most frequently used to correct satellite imagery is a first order transformation also called affine transformation. This transformation can be given by the following polynomials:

X = a0 + a1rn + a2cn Y = b0 + b1rn + b2cn Where rn is row number cn is column number X and Y are map coordinates

To define the transformation, it will be necessary to compute the coefficients of the polynomials e.g. a0, a1 and a2. For the computations, a number of points have to be selected that can be located accurately on the map (X, Y) and which are also identifiable in the image (row, column). The minimum number of points required for the computation of coefficients for an affine transform is three, but in practice you need more. By selecting more points than required, this additional data is used to get the optimal transformation with the smallest overall positional error in the selected points. These errors will appear because of poor positioning of the mouse pointer in an image and by inaccurate measurement of coordinates in a map. The overall accuracy of the transformation is indicated by the average of the errors in the reference points. The so-called Root Mean Square Error (RMSE) or Sigma. If the accuracy of the transformation is acceptable, then the transformation is linked with the image and a reference can be made for each pixel to the given coordinate system, so the image is geo-referenced. After geo-referencing, the image still has its original geometry and the pixels have their initial position in the image with respect to row and column indices. In case the image should be combined with data in another coordinate system or georeference, then a transformation has to be applied. This results in a new image where the pixels are stored in a new line/column geometry which is related to the other geo-reference (containing information on the coordinates and pixel size). This new image is created by methods of resampling, by applying an interpolation method. The interpolation method is used to compute the radiometric values of the pixels in the new image based on the DN

18 | P a g e

values in the original image. After this action, the new image is called geo-coded and it can be overlaid with data having the same coordinate system. In case of satellite imagery from optical systems, it is advised to use a linear transformation. Higher order transformations need much more computation time and in many cases they will enlarge errors. The reference points should be well distributed over the image to minimize the overall error. A good choice is a pattern where the points are along the borders of the image and a few in the center. Affine transformation This is a six parameter transformation with the following unknown parameters: a, b, c, d, e, f. Each transformation requires a minimum number of reference points (3 for affine, 6 for second order and 9 for third order polynomials). If more points are selected, the residuals and the derived Root Mean Square Error (RMSE) or Sigma may be used to obtain the best estimates.

x = y

a b

u +

c d

It modifies the orthogonal type by using different scale factors in the x and y directions. It corrects for shrinkage by means of the scale factor, applies the translation to the shift of the origin and also performs rotation through angle (plus a small angular correction for non-orthogonality to orient the axes in the u, v photo system).

19 | P a g e

REFERENCES [1] Hogjie (2008). Lecture 4, Radiometric correction. Introduction To Remote Sensing [2] Medina, E. (2002). Lecture 6, GEOG371. Remote Sensing, Data Collection, Image Processing and Raster Data. [3] Olaleye, J.B. (2011). Lecture Note 11-1. Photogrammetry & Remote Sensing II, SVY517, Series In Geoinformatics, 39-49.

20 | P a g e

21 | P a g e