Landsat Top of Atmosphere Reflectance

Raw Landsat data as received at the ground station is designated 'Level 0' by NASA and the USGS. Level 0 data is the raw telemetry from the satellite. In order to be usable, this data must be corrected and processed. All commonly available Landsat data, for example data that you can download from GLOVIS or GLCF has been radiometrically corrected to the Level 1 standard. The Landsat 7 Science Data Users Handbook is a very useful reference regarding the Landsat instrument and data. According the Handbook, Level 1 processing algorithms include:

• Payload Correction Data (PCD) processing
• Mirror Scan Correction Data (MSCD) processing
• ETM+/Landsat 7 sensor/platform geometric model creation
• sensor line of sight generation and projection
• output space/input space correction grid generation
• image resampling
• geometric model precision correction using ground control
• terrain correction

The terrain (T) and geometric (G) corrections are also described in the Handbook. This reference also states that "...during 1G product rendering [assume processing] image pixels are converted to units of absolute radiance using 32-bit floating point calculations. Pixel values [Digital Numbers or DNs] are then quantized to byte values prior to media output." The following equation can be used to convert DN's in a 1G product back to radiance units:

Ll = Grescale * QCAL + Brescale

The Grescale and Brescale values are available in the Landsat metadata that accompanies each data set. Published Landsat data is corrected using algorithms 1-8, listed above. The terrain correction using a DEM is only applied for L1T product, and not L1G. Terrain correction means that variations in surface reflectance due to the orientation of the terrain surface with respect to the solar elevation angle (i.e. the angle of incidence of the sun's rays) are corrected. This can be very useful and important for multispectral analysis. All ETM+ data available from GLOVIS is processed to the L1T standard. Some data available from GLCF may be processed to the L1G standard. It is possible to convert from DNs to radiances using the band gain and bias parameters in the Landsat metadata. (DNs or Digital Numbers are sensed radiation quantized to discrete levels, for example 0-255 for Landsat data.)

Conducting analyses using Landsat data (for example vegetation analysis) may or may not require further processing beyond the standard Level 1. NDVI or other image parameters can be computed using uncorrected DNs from L1T data and the index can be successfully used to highlight vegetative areas in an RGB color composite image. However, problems could be introduced when comparing derived image characteristics or statistics using Level 1 DNs between one Landsat scene and another, or between two Landsat scenes acquired at different dates. The problem is that the parameter that you are interested will probably be related to ground reflectance, i.e. the light reflected off the earth's surface if you were standing next to the plant and aiming a spectrophotometer at it. You could collect data during different seasons or during different years and draw valid conclusions from the readings if you took into account the differences in the incident solar radiation as a function of solar zenith angle, atmospheric characteristics and other factors.

Comparing DNs acquired by a satellite without correcting for atmospheric and solar angle effects can lead to problems because even though the Landsat DNs have been calibrated and processed, the DNs are related to, but not the same as the surface reflectance values that you would measure with a spectrophototmeter. The light incident upon the satellite sensors has traveled down to the earth through the atmosphere suffering wavelength-dependent scattering. It reflects off the earth's surface in a diffuse rather than direct manner. Then it travels back up through the earth's atmosphere, suffering more scattering. Moreover, the sun's angle with respect to the earth and the earth's distance from the sun may differ between images as well. Extracting information using DNs can lead to wrong conclusions.

Some of these factors can be corrected by working with Top of Atmosphere (TOA) reflectance values rather than DNs. TOA reflectance is not as good as surface reflectance, but can be more useful than DNs in many cases. (The difference between TOA and surface reflectances is atmospheric distortions.) There are of course a great many other factors that could influence the validity of such comparisons. The USGS Tutorial has a lot more information on atmospheric physics and corrections. Also see Measuring Urban Air Quality Using Worldview2 Multi angle Multispectral Band Data. PANCROMA has a utility for converting Landsat band file DNs to TOA reflectances. The procedure for carrying out such a conversion is as follows:

Conversions must be made one Landsat band file at a time. Open a band file by selecting 'File' | 'Open'. (PANCROMA can input Landsat TIFF and L1G format files) Then select 'PreProcess' | 'Compute Radiometric Corrections' | 'Compute Landsat TOA Reflectance'. The TOA Reflectance Data Form will become visible. This form will ask you to input certain parameters from the metadata file that accompanies each Landsat 7 scene. These parameters are needed to compute and render (if desired) the TOA data. The required parameters are the band number, the solar elevation angle, the acquisition date and a scale factor (to make the image display properly.) TOA reflectance is also used for cloud masking and an explanation of the Solar Elevation Angle and Acquisition Date is given in the Cloud Masking section of the Instruction Manual. Additional parameters include the Landsat band number (this is needed because the sensor gain and offset values are not the same for all Landsat bands). In addition, the form has check boxes for writing the TOA reflectance data to file and for rendering it as an image.

Note that the Landsat 7 algorithm actually wants the solar zenith angle, which is 90^o minus the solar elevation angle. However the solar elevation angle, not the zenith angle is archived in the metadata file so input the solar elevation angle and PANCROMA will take care of the conversion. PANCROMA will not convert the acquisition date from the Gregorian calendar to Julian days. You must do this yourself by either counting on your fingers or alternatively using the NASA link on the PANCROMATM Help menu (http://www-air.larc.nasa.gov/tools/jday.htm) labeled "Launch NASA Ordinal Date Converter".

If you elect to save your file to disk you will be prompted for an output file name, with the suffix 'bin', denoting a binary data file. PANCROMA will output the data in flat binary form using IEEE 32-bit FLOAT (little endian) format. The output file will be ((ROWS by COLUMNS * 4) ) + 8) bytes in size. The extra 8 bytes result from the number of rows and number of columns that precede the FLOAT data. These are output as 4 byte integers, again in little endian format. The file structure is shown schematically below.

Contact us if you need more information regarding the file structure or the C++ algorithms for reading it.

Another option is to save your reflectance data in GeoTiff format. If you check this box on the TOA reflectance form (shown below) you will be prompted for a TIFF file name. When this option is enabled, a GeoTiff reflectance file is written immediately after the TOA Reflectance array is computed. The file is in the same format that USGS Landsat Reflectance data is archived in, namely sixteen bit integer (I16) format. A scaling factor of 100000 is applied to the data. This is ten times greater than the USGS convention but I found it was necessary in order for the output files to display adequately in other applications. You will have to take this into consideration if you import this format into another application.

You can also elect to display the TOA reflectances as an image file. Since the TOA reflectances are small numbers compared to the DNs, they will need to be scaled in order to display properly. The default scale factor is 500. If the images are too dark, try increasing the scale factor. Black splotches on the image mean that the scaled values exceed 255, the maximum grayscale image quantization value so you may want to decrease the scale factor a bit. As long as you use the same scale factor between images, the TOA reflectance values can still be valid since they are scaled absolute numbers. If you decide to create such images, you can output them in any of the standard PANCROMA output formats. A TOA reflectance image, with scaleFactor=500 is shown below.