ENVI Tutorial: Multispectral Analysis of MASTER HDF Data

Overview of This Tutorial

This tutorial introduces you to the Hierarchical Data Format (HDF) and analysis of multispectral imagery from the MODIS/ASTER Airborne Simulator (MASTER) sensor. The first part of the tutorial shows you how to open and read HDF files, extract spatial and spectral subsets of MASTER shortwave infrared (SWIR) data, compare spectra to spectral libraries, and how to perform classic multispectral processing.

The second half of the tutorial explains the analysis of longwave infrared (LWIR) MASTER data, including how to examine LWIR spectra to define key spectral bands, how to perform decorrelation stretching of color composites to enhance LWIR spectral differences, and how to compare SWIR and LWIR mapping results.

Files Used in This Tutorial
CD-ROM: Tutorial Data CD #2
Path: envidata/cup99mas

File	Description
9900511f.hdf	MASTER data of Cuprite, Nevada, USA
header.txt	Text Header with data information

These and other MASTER data are available for purchase from NASA through the EROS Data Center, Sioux Falls, SD.

Background

MASTER is designed to collect simulation data in support of the ASTER and MODIS instrument teams for algorithm development, calibration, and validation. See https://master.jpl.nasa.gov/ for more information. MASTER is a 50-band scanner, operated by NASA, that covers approximately 0.4 to 14 μm. The instrument operates on the NASA Beechcraft B200, DC-8, or ER-2 aircraft to produce spatial resolutions of 5 to 50 m. Spectral band positions are designed to simulate both ASTER and MODIS. MASTER data are delivered via EROS Data Center as HDF files.

HDF is a multi-object file format for the transfer of graphical and numerical data between machines. This format allows you to create, access, and share scientific data in a form that is self-describing and network-transparent. “Self-describing” means that a file includes information defining the data it contains. “Network-transparent” means that a file is represented in a form that can be accessed by computers with different ways of storing integers, characters, and floatingpoint numbers. HDF data supported in ENVI include raster format images, images stored in 2D or 3D scientific data format, and plots stored in 1D scientific data format.

Shortwave Infrared (SWIR) Analysis

Before attempting to start the program, ensure that ENVI is properly installed as described in the installation guide.

Open HDF File and Select a Dataset to Plot

Using ENVI’s HDF reader to open the MASTER HDF file, you can select individual datasets to view and plot from the HDF Dataset Selection dialog.

1. From the ENVI main menu bar, select File → Open External File → Generic Formats → HDF. A file selection dialog appears.
2. Navigate to envidata\cup99mas and select 990051f.hdf. Click Open. An HDF Dataset Selection dialog appears with a list of the available HDF datasets in the file.
3. Select the HDF dataset labeled (50x1):Left50%ResponseWavelength and click OK to plot the data in an HDF 1-D Data Sets window. This dataset contains the center wavelengths for the MASTER HDF data.
4. Select Edit → Data Parameters from the HDF 1-D Data Sets plot window menu bar. A Data Parameters dialog appears.
5. Select a symbol from the Symbol drop-down list. Enlarge the plot two or three times its initial size by dragging one of the corners so that you can see the markers for the individual band centers.
6. Click and drag inside the plot window to read the wavelength values. The band center values appear at the bottom of the plot as you move the cursor. The right number represents the band center of the selected band in micrometers. The range of the MASTER data is approximately 0.4 to 14 μm.
7. Close the plot window when you are finished.

Open HDF File and Load Image

1. Repeat Steps 1-2 above.
2. In the HDF Dataset Selection dialog, select (716x50x2028):CalibratedData and click OK. An HDF Data Set Storage Order dialog appears.
3. Select the BIL radio button and click OK to load the image data into the Available Band List dialog. This dataset contains the image data for the Cuprite, Nevada, MASTER HDF data.
4. In the Available Bands List, select (716x50x2028):CalibratedData: Band 4 and click Load Band. ENVI automatically reads the HDF data.
5. Examine the image and select File → Close All Files from the ENVI main menu bar when you are finished.

Open HDF File with ENVI’s MASTER Reader

The following method of opening an HDF file is different from the previous steps because a HDF Dataset Selection dialog does not appear.

1. From the ENVI main menu bar, select File → Open External File → Thermal → MASTER. A file selection dialog appears. Select 990051f.hdf and click Open.

• Or, select File → Open Image File and double-click on 990051f.hdf. ENVI automatically recognizes the file format. The 50 MASTER bands and their associated wavelengths appear in the Available Bands List dialog.

2. Select the RGB Color radio button in the Available Bands List dialog, then select Band 5, Band 3, and Band 1. Click Load RGB to display a true-color image.

Spatially/Spectrally Subset Data

1. Move the Image box in the Scroll window to the Cuprite area. Refer to the following figure:
   
2. Perform simultaneous spatial and spectral subsetting. From the ENVI main menu bar, select Basic Tools →Resize Data (Spatial/Spectral). A Resize Data Input File dialog appears. Select 9900511f.hdr, but do not click OK yet.
3. Click Spatial Subset. A Select Spatial Subset dialog appears.
4. Click Image. A Subset by Image dialog appears.
5. The red box shows the area currently viewed in the Image window. To subset this area, click OK, followed by OK in the Select Spatial Subset dialog.
6. In the Resize Data Input File dialog, click Spectral Subset. A File Spectral Subset dialog appears.
7. Click Clear. Select Band 1, hold down the key, then select Band 25. Click OK, followed by OK in the Resize Data Input File dialog. A Resize Data Parameters dialog appears.
8. In the Enter Output Filename field, enter resized_image and click OK.

Empirical Reflectance Calibration

1. In the Available Bands List, select the RGB Color radio button. Select Band 5, Band 3, and Band 1 under
   resized_image, and click Load RGB.
2. From the Display group menu bar, select Tools → Profiles → Z Profile (Spectrum). A Spectral Profile plot window appears, showing a spectrum of MASTER bands 1-25 (0.46 - 2.396 μm). Note the shape of the spectrum, corresponding to the shape of the solar irradiance spectrum modified by atmospheric absorption. Select File →Cancel from the Spectral Profile menu bar.
   Perform a rough atmospheric correction using the empirical Flat Field method, by choosing a spectrally flat region
   of interest (ROI) and dividing the average spectrum for that ROI into the spectrum for each pixel in the image.
3. From the Display Group menu bar, select Overlay → Region of Interest. An ROI Tool dialog appears.
4. Draw a polygon ROI inside of Stonewall Playa, the white area near the bottom-center of the image. Click the left mouse button to define polygon segments, then right-click to close the polygon. Right-click again to accept the polygon ROI.
   
5. In the ROI Tool dialog, select the Off radio button to turn off ROI mouse control.
6. From the ENVI main menu bar, select Basic Tools → Preprocessing → Calibration Utilities → Flat Field. A Calibration Input File dialog appears.
7. Select resized_image and click OK. A Flat Field Calibration Parameters dialog appears.
8. Select Region #1… In the Enter Output Filename field, enter calibration. Click OK to start the Flat Field correction.

Display a Color Composite and Extract Spectra

1. In the Available Bands List, select the RGB Color radio button. Select Band 5, Band 3, and Band 1 under calibration. Click Load RGB.
2. From the Display group menu bar, select Tools → Profiles → Z Profile (Spectrum). A Spectral Profile plot window appears, showing a spectrum of MASTER bands 1-25 (0.46 - 2.396 μm). Note the corrected spectrum, corresponding to apparent reflectance or relative reflectance (relative to the Flat Field spectrum).
3. Drag the cursor throughout the Image window and examine the spectra. Move the cursor to some of the red areas in the image, which correspond to Fe+3 absorption features near 0.87 μm (Band 9), and examine their spectra.
4. Click-and-hold the middle mouse button in the Spectral Profile window, and draw a box around the 2.0 - 2.4 μm range. Drag the cursor throughout the Image window, and observe the absorption features near 2.2 and 2.3 μm caused by clays and carbonates, respectively.

Compare Image Spectra to Spectral Library

1. From the ENVI main menu bar, select Spectral → Spectral Libraries → Spectral Library Viewer. A Spectral Library Input File dialog appears.
2. Click Open and select Spectral Library. A file selection dialog appears. Navigate to spec_lib\usgs_min and select usgs_min.sli. Click Open.
3. Select usgs_min.sli in the Spectral Library Input File and click OK. A Spectral Library Viewer dialog appears.
4. Select alunite1.spc Alunite GDS84 Na03. A Spectral Library Plots window appears, showing the spectrum for alunite. Then, select the following spectra:

• budding1.spc Buddingtonite GDS85 D-206
• calcite1.spc Calcite WS272
• kaolini1.spc Kaolinite CM9

5. Right-click in the Spectral Library Plots window and select Stack Plots.
   
6. Select MASTER spectra for the same minerals. From the Display group menu bar, select Tools → Pixel Locator.
7. In the Pixel Locator dialog, enter the pixel location (521, 1587), which is an alunite feature. Click Apply.
8. Right-click in the Spectral Profile and select Collect Spectra.
9. Enter the following pixel locations in the Pixel Locator dialog, and click Apply each time:

• (424, 1578) - Buddingtonite
• (239, 1775) - Calcite
• (483, 1674) - Kaolinite

11. Right-click in the Spectral Profile window and select Stack Plots. Right-click again and select Plot Key.
12. In the Spectral Profile and Spectral Library Plots windows, zoom to a range of 2.0 to 2.4 μm, using the middle mouse button to draw a box around the range.
13. Compare the image spectra to the library spectra. Again, note the absorption features near 2.2 and 2.3 μm.
    
14. For a more direct comparison, select Options → New Window: Blank from the Spectral Profile menu bar. Click-and-drag a spectrum name from the Spectral Profile window into the ENVI Plot Window. Then, click-anddrag the corresponding spectrum name from the Spectral Library Plots window into the ENVI Plot Window. Following is an example for kaolinite:
    

Image Processing with SAM

Use spectra collected from the image to perform a Spectral Angle Mapper (SAM) classification of the MASTER data. SAM measures the similarity of unknown and reference spectra in n-dimensions. The spectral angle is the angle between the spectra (represented as vectors) in n-D space. This method assumes that the data have been reduced to apparent reflectance and uses only the direction of the spectra, not their length. Thus, the SAM classification is insensitive to
illumination effects. See the Selected Hyperspectral Mapping Methods tutorial and ENVI Help for more information and examples.

1. From the ENVI main menu bar, select Classification → Supervised → Spectral Angle Mapper. A
   Classification Input File dialog appears.
2. Select calibration and click OK. An Endmember Collection:SAM dialog appears.
3. Click-and-drag the spectra names from the Spectral Profiles window to the Endmember Collection:SAM dialog.
4. Click Select All and Plot to confirm that you have the correct endmember spectra.
5. Click Apply. A Spectral Angle Mapper Parameters dialog appears.
6. In the Enter Output Filename field, enter sam_class_out. This is the SAM classification image that will be created.
7. In the Enter Output Rule Filename field, enter sam_rule_out. Click OK. This is the set of SAM rule images that will be created.
8. In the Available Bands List, click Display #1 and select New Display.
9. Select Sam under sam_class_out. Select the Gray Scale radio button, and click Load Band. The classes are color-coded the same as the spectrum plot colors:

Mineral	Color
Kaolinite	Blue
Calcite	Green
Alunite	White
Buddingtonite	Red

11. Compare the classification map to the true-color image. From a Display group menu bar, select Tools → Link → Link Displays. Click in an Image window to toggle between the true-color image and the classification map.
12. From a Display group menu bar, select Tools → Profiles → Z Profile (Spectrum). Verify the spectral match.
13. You can also use the rule images to evaluate the spectral matches. The rule images show the best matches (small angles) in black when first displayed. Since it is more intuitive to show the best matches as brighter values, you can invert the colors in the rule images. Steps 13-15 explain this process.
14. In the Available Bands List, click Display #2 and select Display #1. Select a rule band under sam_rule_out and click Load Band.
15. From the Display group menu bar, select Tools → Color Mapping → ENVI Color Tables.
16. Move the Stretch Bottom slider all the way to the right, and move the Stretch Top slider all the way to the left.
17. From a Display group menu bar, select Tools → Link → Link Displays. A Link Displays dialog appears. Click OK to link Display #1 (rule image) and Display #2 (classification image).
18. Click in an Image window to toggle between the rule image and classification image. Examine the spatial locations of specific mapped minerals.
19. When you are finished, select Window → Close All Display Windows from the ENVI main menu bar.

Longwave Infrared (LWIR) Analysis

Emission spectra of typical rocks indicate that the LWIR region (8 to 14 μm) is best suited for determining rock types. The emissivity minimum shifts from around 8.5 µm for framework silicates (quartz and feldspars) to progressively longer wavelengths for sheet and chain silicates, and isolated ${\text{SiO}}_{\text{4}}$.

View LWIR Color Composite and Subset Data

1. In the Available Bands List, select the RGB Color radio button in the Available Bands List dialog, then select Band 46, Band 44, and Band 41. Click Load RGB to display a true-color image.
2. In the Available Bands List, click Display #1 and select New Display.
3. In the Available Bands List, select any band under resized_image (the subset you created earlier), select the Gray Scale radio button, and click Load Band.
4. From the ENVI main menu bar, select Basic Tools → Resize Data (Spatial/Spectral). A Resize Data Input File dialog appears. Select 9900511f.hdf, but do not click OK yet.
5. Click Spatial Subset. A Select Spatial Subset dialog appears. You will use the same spatial subset you created earlier, but with a different spectral subset.
6. Click File. A file selection dialog appears.
7. Select resized_image and click OK, followed by OK in the Select Spatial Subset dialog.
8. In the Resize Data Input File dialog, click Spectral Subset. A File Spectral Subset dialog appears.
9. Click Clear. Select Band 41, hold down the key, then select Band 50. Click OK, followed by OK in the Resize Data Input File dialog. A Resize Data Parameters dialog appears.
10. In the Enter Output Filename field, enter resized_image_lwir and click OK.
11. Close Display #2.

Animate LWIR Bands

1. In the Available Bands List, select Band 41 under resized_image_lwir, select the Gray Scale radio button, and and click Load Band.
2. Animate all of the subsetted LWIR bands and examine their variability. From the Display group menu bar, select Tools → Animation. The Animation Image Parameters dialog appears. Click OK to accept the default values. Ten LWIR bands are loaded and animated.
   Most of the variability is caused by differential heating, not by spectral differences between bands. Ideally, these data should be atmospherically corrected and converted to emissivity to enhance spectral differences. However, for this tutorial, you will substitute selection of key image bands using LWIR spectra, and enhance them using a Decorrelation Stretch.
3. From the Resize (animation) menu bar, select File → Cancel.
4. If desired, display individual bands of interest in new display groups, and use image linking and dynamic overlays to compare them.

Review LWIR Spectral Features

ENVI includes spectral libraries from Johns Hopkins University, which contain spectra for selected materials from 0.4 to 14 μm. The apparently seamless reflectance spectra over this region of rocks and soils were generated using two different instruments, both equipped with integrating spheres for measuring directional hemispherical reflectance.

Under most conditions, the infrared portion of these data can be used to calculate emissivity using Kirchhoff’s Law. For this tutorial, however, highs in the reflectance spectra from the spectral library can be considered equivalent to lows expected in emissivity spectra.

1. From the ENVI main menu bar, select Spectral → Spectral Libraries → Spectral Library Viewer. A Spectral Library Input File dialog appears.
2. Click Open and select Spectral Library. A file selection dialog appears. Navigate to spec_lib\jhu_lib and select minerals.sli. Click Open.
3. Select minerals.sli in the Spectral Library Input File and click OK. A Spectral Library Viewer dialog appears.
4. Select the first Quartz option. A Spectral Library Plots window appears, showing the spectrum for quartz. Then, select the second Calcite option.
5. From the Spectral Library Plots window menu bar, select Edit → Plot Parameters.
6. Enter Range values from 8 to 13 and click Apply. Quartz (silica) shows a maximum near 9 μm, while Calcite does not. You can use this information to help find all of the silica-rich areas imaged by the MASTER data.
   

Design and Display Color Composites

1. Click in the Spectral Library Plots window to examine the nature of the two spectra. Look for contrasts between 8 to 14 μm, and relate them to the MASTER LWIR spectral bands.
2. Bands 46, 44, and 41 (10.085, 9.054, and 7.793 μm, respectively) bracket the prominent silica feature near 9.0 μm. In the Available Bands List, select the RGB Color radio button, select Band 46, Band 44, and Band 41 under resized_image_lwir, and click Load RGB.
3. Relate the image colors to the expected relative contributions, based on the library spectra. The colors do not match the expected colors very well because the effect of temperature overwhelms the spectral differences. The bands are highly correlated because of differential heating of the rocks and soils.

Perform Decorrelation Stretch

Use Decorrelation Stretch to enhance color differences, by removing the high correlation commonly found in LWIR multispectral data. ENVI provides a Decorrelation Stretch tool, but you can obtain similar results by computing a forward principal components (PC) analysis, performing contrast stretching, and computing an inverse PC analysis.

1. Decorrelation stretching requires three bands (a stretched color composite) for input. You will use the currently displayed color composite.
2. From the ENVI main menu bar, select Transform → Decorrelation Stretch. A Decorrelation Stretch Input File dialog appears. Select Display #1 and click OK. A Decorrelation Stretch Parameters dialog appears.
3. In the Enter Output Filename field, enter lwir_stretched and click OK.
4. In the Available Bands List, select the RGB Color radio button. Select R DS, G DS, and B DS, and click Load RGB. Try other color combinations.
   

Compare LWIR and SWIR Results

Compare the SAM results from the first half of this tutorial with the decorrelated LWIR image.

1. In the Available Bands List, click Display #1 and select New Display.
2. Select Sam under sam_class_out. Select the Gray Scale radio button, and click Load Band. As a reminder, the classes are color-coded as follows:

Mineral	Color
Kaolinite	Blue
Calcite	Green
Alunite	White
Buddingtonite	Red

4. From a Display group menu bar, select Tools → Link → Link Displays and click OK to link the two images.
5. Click in an Image window to toggle between the two images. Compare the distribution of minerals mapped from the SWIR data versus the distribution of silica (red areas) on the decorrelated image.

Optional: Perform Combined SWIR/LWIR Analysis

Use the MASTER data in a combined SWIR/LWIR analysis (using bands 1-25 and bands 41-50, respectively) that follows ENVI’s hyperspectral processing flow. Refer to the tutorials Advanced Hyperspectral Analysis and Geologic Hyperspectral Analysis Case History for more information. Use the following steps as a general methodology.

1. From the file 990051f.hdf, extract the SWIR and LWIR bands using spectral and spatial subsetting to build a combined data cube.
2. Create a Minimum Noise Transform (MNF) output file, and perform MNF analysis by looking at the spectral bands and the eigenvalue plots. Select a reduced number of MNF bands for further analysis.
3. Run a Fast Pixel Purity Index (PPI) analysis with 10,000 iterations to find the key endmember spectra, thus reducing the spectral dimensionality. Threshold the PPI to a ROI with around 5,000 pixels.
4. Use the n-D Visualizer to select endmembers. Rotate the scatter plot in many dimensions, and select extreme pixels by drawing ROIs in the n-D Visualizer and exporting them to image ROIs.
5. Use an ENVI spectral mapping method with the combined datasets. Compare these results to the SWIR and LWIR results above.

Reference: http://www.spweather.com/enviuser/tutorial/misc/HDF_Master_Data.pdf