Image Preprocessing Learning Objectives Be able to describe

Image Preprocessing

Learning Objectives • • • Be able to describe when and why image corrections are appropriate or necessary Give examples of some common approaches to image correction Understand the processing steps of Landsat data

Image Preprocessing • Preprocessing is the removal of systematic noise from the data (Rees, 2001). It is the first step in the image processing chain and is usually necessary prior to image classification and analysis. GOAL : following image preprocessing, all images should appear as if they were acquired from the same sensor

Preprocessing Steps • • Noise reduction/data loss correction Atmospheric Correction Radiometric Calibration Geometric Correction

Noise Reduction • Two types of noise: global and local • Global noise = random DN variation at every pixel - can be reduced using filters (moving windows) or Fourier transform • Local noise may include errors such as : o Missing scan lines o Image striping

Missing Scan Lines • Cause: Sensor timing failure • Solution: Interpolate to fill in the missing data. • Missing scan line pixel values are estimated using the values of the pixels in the lines above and below the missing line (based on the principle of spatial autocorrelation)

Striping • Caused by an imbalance in detector gains and offsets • Solution: re-calibrate sensors (adjust pixel DNs from each detector to yield the same mean and standard deviation over the entire image)

Preprocessing Steps • • Noise reduction/data loss correction Atmospheric Correction Radiometric Calibration Geometric Correction

Atmospheric Correction • Reducing the effects of atmospheric conditions on the image values

• The value recorded at a given pixel includes not only the reflected radiation from the surface, but the radiation scattered and emitted by the atmosphere as well (path radiance).

Atmospheric Correction • Landsat 8 Cirrus band – Used to identify cirrus clouds which may not be visible with the naked eye – May want to remove those areas when conducting analyses/research

Atmospheric Correction - tools Correction methods for reducing atmospheric effects • Tools for improving both local and global effects • May improve some areas but cause artifacts in others (overcorrect) • ERDAS and ENVI both have modules that are used for atmospheric effects

Atmospheric Correction • IMPORTANT : Some atmospheric effects can not be fixed! – Dense clouds – Smoke – Heavy shadows • Best to exclude these areas from the analysis

Atmospheric Correction – Global Correction Techniques • Dark object subtraction • Conversion to surface reflectance • Image Normalization

Dark Object Subtraction • Dark objects have little to no reflectance observed by the scanner, so the DN values represent path radiance or the influence of atmospheric effects. By subtracting the value of the DN in each band, you remove that artifact. Dark object

Conversion to Surface Reflectance • Requires knowledge of aerosol conditions at the time of the image acquisition • Use of radiative transfer models based on optical depths of ozone and particulates in the atmosphere • Not usually possible for historical analyses

COST model by Chavez • Based only on image statistics – does not require field based data of aerosol conditions • Also utilizes dark object subtraction ERDAS Model available

Image to Image Normalization • Normalize one or more images to a ‘base’ image • Choose bright, mid and dark image targets that will be consistent in all images • Extract values and run linear regression • Use linear equation to adjust the values from the other, non-base images

Image 2 brightness value Image Normalization Image 1 brightness value

21

22

Comparison of sample 1986/2000 image bright target values BEFORE normalization Comparison of 1986 and 2000 bright target values AFTER normalization 23

Preprocessing Steps • • Noise reduction/data loss correction Atmospheric Correction Radiometric Calibration Geometric Correction

Radiometric Calibration • Reduce inconsistencies across detectors and reduce noise caused by sensor calibration, sun angle, and other conditions • Useful for comparing across sensors or comparisons across time

DN (raw value from the sensor) At-sensor radiance Calibrate based on gain and offset values Requires: Earth-sun distance Solar zenith angle Exoatmospheric irradiance TOA Reflectance Requires: Knowledge of aerosol properties Radiative transfer model Surface Reflectance

Conversion to Radiance = gain * DN + offset which is also expressed as: Radiance = ((LMAX-LMIN)/(QCALMAX-QCALMIN)) * (QCALMIN) + LMIN QCAL = DN From Landsat metadata

Lmin and Lmax values used to convert to radiance. Can also use the gain and offset coefficients

Conversion to Radiance Gain and offset values from Landsat metadata file

Radiance to Reflectance Conversion

Radiance to Reflectance Conversion ETM+ TM Band watts/(meter squared * µm) 1 1969. 000 1957. 00 2 1840. 000 1829. 00 3 1551. 000 1557. 00 4 1044. 000 1047. 00 5 225. 700 219. 3 7 82. 07 74. 52 8 1368. 000 Solar Exoatmospheric Irradiance Values

Radiance to Reflectance Conversion Solar Zenith Angle Zenith angle = 90. 0 – sun elevation

Preprocessing Steps • Noise reduction/correcting for data loss • Atmospheric Correction • Radiometric Calibration • Geometric Correction

Geometric Correction • Georeferencing • Image registration • Rectification/orthorectification

Georeferencing The process of assigning map coordinates to image data. The image data are not altered (i. e. DNs do not change assuming NN resampling). From Arc. GIS Resources online http: //resources. arcgis. com/en/help/main/10. 1/index. html#//009 t 000000 mn 000000

Image Registration The process of making one image conform geographically to another image. The process may or may not involve rectification. In most cases, some form of geometric transformation will be necessary. This involves ‘warping’ the image using mathematical models. From Arc. GIS Resources online http: //resources. arcgis. com/en/help/main/10. 1/index. html#//009 t 000000 mn 000000

Polynomial Transformations 1 st-order polynomial equations 2 nd-order polynomial equations 37

Orthorectification "Orthorectification is the process of removing the effects of image perspective (tilt) and relief (terrain) for the purpose of creating a planimetrically correct image. The resulting orthorectified image has a constant scale wherein features are represented in their 'true' positions. This allows for the accurate direct measurement of distances, angles, and areas. "

Once orthorectified, analyst can extract data from the image. Scale and distance are now true. 39

Once orthorectified, analyst can extract data from the image. Scale and distance are now true. 40

Steps for Orthorectification • 1. Use DEM to locate ground control points • 2. Compute mathematical equation (geometric transformation) • 3. Measure the accuracy of the transformation • 4. Create a new output image by applying the mathematical equation to the pixel data and resampling

Image Resampling 3 common options; 1. Nearest neighbor 2. Bilinear interpolation 3. Cubic convolution Process of converting the original image grid to a new image, either by projecting to a new coordinate system or altering the pixel dimensions 42

Resampling Methods Nearest Neighbor The pixel value of the output pixel is assigned to the closest input pixel (pixel center, really) 43

Resampling Methods Bilinear Interpolation The pixel value of the output pixel is based on the weighted distance to the 4 pixel values nearest the input pixel 44

Resampling Methods Cubic Convolution The pixel value of the output pixel is based on the weighted distance to the 16 pixel values (in a 4 X 4 array) nearest the input pixel 45

Preprocessing Landsat data are currently corrected by USGS/EROS to level 1 T which includes the following: • Geometric correction with ground control points for accurate ground location • Radiometric correction for accurate measurements at the sensor and no data loss I. may still want to convert to TOA reflectance for climate change studies

THE END