Correct Images for Atmospheric Effects

来源: https://vis-webcontent.s3.amazonaws.com/quickguides/Image+Preprocessing/Atmospheric+Correction/index.html#/

Lesson content

Lesson 1 of 1

Correct Images for Atmospheric Effects

In this quick guide, you will use the following tools to mitigate atmospheric effects in an Earth surface Mineral dust source InvesTigation (EMIT)(opens in a new tab) hyperspectral image:

•

Fast Line-of-sight Atmospheric Analysis of Hypercubes (FLAASH®) * •

Quick Atmospheric Correction (QUAC®) * •

Dark Subtraction

You will also compare spectral profiles of the resulting QUAC and FLAASH reflectance images.

Sample Data

Download sample data below. Then extract the contents of the .zip file to a local directory.

[EMIT_L1B_Radiance.zip

307.9 MB

DownloadArrow down with horizontal line beneath it](assets/EMIT_L1B_Radiance.zip)

Background

Atmospheric processes such as absorption, transmission, scattering, and reflectance must be taken into account before doing any spectral analysis, as they can degrade image quality and affect the accuracy of image interpretation. Atmospheric correction methods do not completely remove the effects of these processes. Rather, they minimize their effects in imagery.

Multispectral sensors such as Landsat and Sentinel-2 are designed to account for gas absorption features, because their bands correspond to atmospheric windows. Hyperspectral sensors, in contrast, cover the entire visible to shortwave-infrared (SWIR) spectrum, including bands where gas absorption is present. It is imperative to compensate for these effects in hyperspectral imagery.

A further pre-processing step that often accompanies atmospheric correction is calibration to surface reflectance. Any application that analyzes the reflectance properties of materials should be conducted with an image that has been corrected for atmospheric effects and calibrated to surface reflectance.

Many data vendors today distribute surface reflectance images that are “analysis-ready,” meaning that they have already been corrected for atmospheric effects and calibrated to surface reflectance. Examples include Landsat, WorldView, PlanetScope, Sentinel-2, AVIRIS, and EMIT. If you don't want to use these products and would rather perform atmospheric correction yourself, ENVI provides several empirical and model-based methods to do this.

Open and Display an EMIT Image

1

Select File > Open from the Menu bar. An Open dialog appears. 2. 2

Go to the directory where you downloaded the sample data, and select the file EMIT_L1B_Radiance.dat. Click Open. The image is added to the Layer Manager and displayed in the Image window.

The image is located near Yosemite National Park in California. Bridgeport Reservoir is located in the bottom part of the image. It is a spatial subset extracted from a larger NetCDF-4 image and converted to ENVI raster format (.dat).

The image is not georeferenced to a standard map projection. See the Georeference Images Using Geographic Lookup Tables (GLTs) quick guide for instructions on georeferencing EMIT images that are in NetCDF-4 format.

3

Select File > Data Manager from the Menu bar. The Data Manager appears. 2. 4

Scroll through the list of bands. The values in parentheses are center wavelengths, in nanometers (nm). Each band covers a wavelength range of approximately 7.5 nm.

5

In the Layer Manager, right-click on EMIT_L1B_Radiance.dat and select Profiles > Spectral. The pixel values in the image represent at-sensor radiance. You can tell this is a radiance image because the Spectral Profile curve mimics a solar irradiance curve, also called the solar spectrum; for example:

Solar irradiance spectrum. Moderate Resolution Imaging Spectrometer (MODIS) bands are shown in orange, for reference. The solar spectral irradiance curve at the top of the atmosphere is colored green. Solar irradiance transmitted through the atmosphere to the Earth’s surface is colored brown. Source: NASA, 2012 (public domain).

"Bad" bands are marked with yellow triangle symbols. They contain striping and other noise because they correspond to gas absorption regions in the visible to shortwave-infrared spectrum.

6

Close the Spectral Profile and Data Manager.

Run FLAASH

FLAASH is included with the ENVI Atmospheric Correction module. It is a rigorous method that uses MODTRAN® radiative transfer code to simulate atmospheric properties such as water vapor, distribution of aerosols, and scene visibility. Hyperspectral images, in particular, typically provide enough spectral information to measure atmospheric water vapor absorption bands. This can improve the accuracy of a model-based method. FLAASH also accommodates multispectral sensors. It contains many parameters and requires a lot of user input, particularly when using hyperspectral imagery. For a thorough description of parameters, refer to the ENVI Help(opens in a new tab).

1

In the search window of the Toolbox, enter flaash. 2. 2

Double-click the FLAASH Atmospheric Correction tool. The Data Selection dialog appears. 3. 3

Select EMIT_L1B_Radiance.dat and click OK. The FLAASH - Rigorous Atmospheric Correction dialog appears. The scene acquisition date and time are automatically filled in. 4. 4

In the Output Raster field, enter an output file name of EMIT_FLAASH_Refl.dat.

Enter Sensor Parameters

5

Click the Sensor tab. 2. 6

Click the Sensor Type drop-down list and select Unknown. 3. 7

Enter a value of 1 in the Input Scale field. Scale factors are used to convert radiance values to units that FLAASH expects, namely, μW / (cm2 * nm * sr). The EMIT radiance image is already in these units, so scaling is unnecessary.

Enter Geometric Parameters

8

Click the Geometric tab. 2. 9

In the Lat field, enter 38.16. 3. 10

In the Lon field, enter -119.27. 4. 11

In the Sensor Altitude field, enter 407 km. 5. 12

In the Ground Elevation field, enter 2.4 km. The average scene elevation was estimated using Google Earth.

ENVI will automatically determine the Solar Azimuth and Solar Zenith values based on the scene's geographic coordinates and acquisition time. It does not display them here.

Select an Atmospheric Model

13

Click the Model tab. 2. 14

Click the Atmospheric Model drop-down list and select 1976 U.S. Standard Atmosphere. This model is appropriate when the estimated air temperature is 15°C (59°F).

You will keep the default options in the Water tab.

Enter Aerosol Parameters

15

Click the Aerosol tab. 2. 16

Click the Aerosol Model drop-down list and select High-Visibility Rural. 3. 17

Leave the remaining parameters at their default values.

Enter Miscellaneous Parameters

18

Click the Misc. tab. 2. 19

Click the Spectral Polishing drop-down list and select Polish using statistical detection of spectral artifacts. Spectral polishing is a technique that reduces spectral artifacts in hyperspectral data. The selected method calculates the polishing factors using a comparison between a spectrum and a low pass-filtered version of the spectrum. 3. 20

Set the Polishing Width value to 9.

Run FLAASH and Display a Spectral Profile

21

Click OK in the FLAASH - Rigorous Atmospheric Correction dialog. When processing is complete, the FLAASH-corrected reflectance image is added to the Layer Manager and displayed in the Image window. It looks the same as the radiance image. 2. 22

In the Layer Manager, right-click on EMIT_FLAASH_Refl.dat and select Profiles > Spectral. The Spectral Profile window appears with a reflectance curve of the pixel underneath the red crosshairs. 3. 23

In the Go To window of the Toolbar, enter pixel coordinates 224, 924. Press the Enter key. The crosshairs jump to a vegetation pixel, and the reflectance curve updates accordingly.

24

Keep the Spectral Profile window open. Later, you will display a Spectral Profile of the same pixel location in the QUAC reflectance image so that you can compare them.

Run QUAC

QUAC is also included with the ENVI Atmospheric Correction module. It is a scene-based method, meaning that atmospheric correction parameters are derived strictly from the pixel spectra and not from reference spectra. Because direct measurements of atmospheric properties are rarely available, the properties can be inferred from image pixels. QUAC is simple to use and accommodates a wide range of sensors and wavelengths (VNIR to SWIR, approximately 0.4 to 2.5 µm).

The image should contain a variety of spectrally diverse materials—at least 10—such as water, soil, vegetation, and man-made structures. It performs best when the image is uniformly illuminated, such as in clear-sky conditions or when airborne sensors fly under complete cloud cover.

1

In the search window of the Toolbox, enter quac. 2. 2

Double-click the QUAC - Quick Atmospheric Correction tool. The Data Selection dialog appears. 3. 3

Select EMIT_L1B_Radiance.dat and click OK. The QUAC dialog appears. 4. 4

Click the Sensor Type drop-down list and select Generic / Unknown Sensor. 5. 5

In the Output Raster field, enter a file name of EMIT_QUAC_Refl.dat.

6

Click OK. When processing is complete, the QUAC-corrected reflectance image is added to the Layer Manager and displayed in the Image window. 2. 7

In the Layer Manager, right-click on EMIT_QUAC_Refl.dat and select Profiles > Spectral. 3. 8

In the Toolbar, put your cursor at the end of 224, 924 and press the Enter key. The crosshairs jump to the same vegetation pixel as before. The reflectance curve of the QUAC Spectral Profile updates accordingly. 4. 9

In the QUAC Spectral Profile window, click the small arrow icon to the right of the plot. This expands the plot properties. 5. 10

Click the Curve tab. 6. 11

Click inside the Color field. A small drop-down arrow appears. 7. 12

Select a dark blue color. The plot line changes to that color.

13

Click the small arrow to the right of the FLAASH Spectral Profile plot window to expand its plot properties. 2. 14

Drag and drop the plot name from the FLAASH Spectral Profile to the plot properties of the QUAC Spectral Profile.

Both reflectance curves are plotted in the same window. Again, they represent the same vegetation pixel location. The numbers along the Y-axis are reflectance values, scaled by 10,000. For example, a value of 3,000 represents 0.3, or 30% reflectance.

Overall, the QUAC reflectance values are higher than those of FLAASH, especially from 750 to 1,200 nm.

QUAC and FLAASH may show different reflectance values because of differences in their algorithms and assumptions. FLAASH uses a more rigorous physics-based first-principles method, which takes into account complex atmospheric conditions like molecular and aerosol scattering, yielding more accurate reflectance estimates. QUAC performs a more approximate correction that assumes a uniform atmosphere and constant aerosol properties, which can cause overestimation of reflectance, particularly in areas with varying aerosol properties. Typically, QUAC produces reflectance spectra within +/-15% of the physics-based approaches. QUAC is faster and requires less user input than FLAASH and performs well if there are at least 10 diverse materials in the scene and enough dark pixels to accurately estimate the baseline spectrum.

15

Close both Spectral Profile windows.

Both QUAC and FLAASH create surface reflectance images, with values scaled by 10,000. If you will be conducting analyses that compare image spectra to reference spectra—such as a spectral library or spectrometer data—then reflectance values should ideally range from 0 to 1. See the Scale and Mask Reflectance Images quick guide for instructions on how to do this.

Run Dark Subtraction

Unlike the previous methods, Dark Subtraction does not produce a reflectance image. It simulates atmospheric correction without calibrating imagery to surface reflectance. It attempts to remove the effects of atmospheric scattering from a scene (especially in the blue wavelength region) by subtracting a specific value from every pixel. This value represents a background signature. It can be the band minimum, a mean value based on an ROI, or a user-specified value.

The Dark Subtraction tool is included with ENVI and does not require the Atmospheric Correction module.

1

In the search window of the Toolbox, enter dark. 2. 2

Double-click the Dark Subtraction tool. The Dark Subtraction Correction dialog appears. 3. 3

Click the Browse button next to Input Raster. The Data Selection dialog appears. 4. 4

Select EMIT_L1B_Radiance.dat and click OK. For this exercise, you will use minimum band values to simulate atmospheric correction. ENVI will choose that method when you leave the ROI and Values fields blank. 5. 5

In the Output Raster field, enter a file name of EMIT_DarkSub_Radiance.dat.

6

Click OK. ENVI computes image statistics and creates the dark subtraction image. 2. 7

In the Layer Manager, uncheck the EMIT_QUAC_Refl.dat and EMIT_FLAASH_Refl.dat layers to hide them. 3. 8

Toggle the EMIT_DarkSub_Rad.dat layer on and off to compare it with the EMIT_L1B_Radiance.dat image. They look identical; however, their data values are slightly different. 4. 9

Click the Cursor Value button in the Toolbar to display the Cursor Value dialog.

10

Compare the "Data" values in the Cursor Value dialog.

For each image:

•

The first number is the radiance value in the band assigned to red (Band 36). * •

The second number is the radiance value in the band assigned to green (Band 24). * •

The third number is the radiance value in in the band assigned to blue (Band 13).

The radiance values in EMIT_DarkSub_Rad.dat are lower than those of EMIT_L1B_Radiance.dat. For example, the Band 36 (red) values are 1.22145 in the dark subtraction image and 1.849405 in the original radiance image. This is one way to verify that Dark Subtraction subtracted the effects of scattering from the original radiance image.

You can also compare spectral profiles of both images, using the technique shown in the earlier QUAC section.

You cannot compare spectral profiles of the dark subtraction image with those of FLAASH or QUAC. The dark subtraction image is still a radiance image, while the others are surface reflectance images.

11

This concludes the exercise.

In summary, FLAASH is a rigorous, model-based method that takes into account aerosols, water vapor, and solar angles. Like QUAC, it produces accurate results but requires more user input.

QUAC is an empirical technique that is easy to run and works well in a variety of cases. If you are working with Sentinel-2 or Landsat top-of-atmosphere (TOA) reflectance images, QUAC is a good option for calibrating them to surface reflectance.

Dark Subtraction is a simple empirical method used to simulate atmospheric correction. Other empirical atmospheric correction tools include Empirical Line Calibration(opens in a new tab), Flat Field Correction(opens in a new tab), Internal Average Relative (IAR) Reflectance Correction(opens in a new tab), and Log Residuals Correction(opens in a new tab).

Your input is important to us, please take a few moments to fill out ourQuick Guide Feedback(opens in a new tab)form.