Lab 6: Multispectral Image Processing Using Band Ratios

Size: px

Start display at page:

Download "Lab 6: Multispectral Image Processing Using Band Ratios"

Madlyn Stewart
5 years ago
Views:

Lab 6: Multispectral Image Processing Using Band Ratios due Dec. 11, 2017 Goals: 1. To learn about the spectral characteristics of vegetation and geologic materials. 2. To experiment with vegetation indices for mapping vegetation.

1 Lab 6: Multispectral Image Processing Using Band Ratios due Dec. 11, 2017 Goals: 1. To learn about the spectral characteristics of vegetation and geologic materials. 2. To experiment with vegetation indices for mapping vegetation. 3. To experiment with geologic indices for mapping purposes. 4. To gain experience obtaining and working with ASTER multispectral data. Introduction: The aim of this lab is to investigate index techniques for producing maps of spectrally distinct surface cover types. This method takes advantage of differences in the spectra of land cover types to do simple mapping using spectral indices that include band ratios and other arithmetic and Boolean operations on image data. In general, the input data should be calibrated and atmospherically-corrected multi- or hyperspectral image data, i.e. surface reflectance, surface emissivity, etc. Simple Topographic/Illumination Correction (Informational Only No Need to Implement This Extra Credit If You Do, Though) Besides calibration and an atmospheric correction, an additional radiometric correction that sometimes is important to apply to image data is an illumination correction. This is necessary in order to remove, for example, the effect of shadowing due to topography and the illumination conditions that can erroneously cause surface reflectance values to be lower than what they should be. One way to do this is with the cosine correction: cosθ BV tic l = BV o + c l cosi + c k (1), where BV l is the brightness value of a given pixel in band l of the input image, BV tic l is the topography- and illumination-corrected brightness value for that same pixel in band l of the output image, θ o is the solar zenith angle (with respect to vertical), i is the solar incidence angle (with respect to the surface normal), and c and k are constants. For our purposes, we will assume c = 0 and k = 1. Figure (1) illustrates the geometry. Note that, given θ o, you will need to also know the surface slope (φ) of the terrain (smallest acute angle in the triangle labeled Terrain in Figure 1) in order to calculate i, i.e.: Figure 1. θ o = φ + i (2). 1

2 Spectral Indices Spectral index methods can be used to map out different surface cover types in a simple, semi-quantitative fashion. An index is typically defined as a simple ratio of two bands, although indices can be more complicated (involving addition, subtraction, more than two bands, false-color image combinations, etc.). However sophisticated, this approach to combining bands leads to a grayscale image (or a false-color image) that can be interpreted or further enhanced. The idea is that, by combining multiple images, you create an image map that can offer insights that the individual input images, by themselves, could not. You can implement a variety of these mathematical operations between bands (e.g. division, multiplication, addition, subtraction) using ENVI s Band Math function (see the ENVI help for more information). [In addition to enhancing spectral differences, band ratio indices have the added advantage of being relatively independent of scene illumination conditions, topography, and some atmospheric effects. All three are, to some extent (but not entirely), cancelled out by the ratio.] Spectral Characteristics of Vegetation and the NDVI Cells in plant leaves are effective scatterers of light due to the high refractive index contrast between water in cells and air spaces in between. Because of chlorophyll and other pigments, healthy vegetation is dark in the visible (0.4 to 0.7 µm), absorbing 80-90% of incident irradiance (Figure 2). Vegetation is green because it is reflects slightly more efficiently at about 0.55 µm (Figure 2). Between 0.7 to 1.3 µm, vegetation is, in comparison, very bright, reflecting 40-50% of incident irradiance (Figure 2). This is due to the transition from electronic to molecular excitations in the near-infrared, the concomitant lack of absorption, and the resulting dominance of scattering from the internal leaf structure. Beyond the red edge (~0.7 µm), between 1.3 and 2.5 µm, vegetation is a bit darker (but brighter than in the visible), with characteristic absorption features (at 1.4, 1.9 µm, etc.) due to water absorption. Figure 2. 2

3 Dead or dormant vegetation has a higher reflectance than healthy green vegetation throughout the visible (0.4 to 0.7 µm) (Figure 2). Conversely, it reflects less than green vegetation in the near-infrared (0.7 to 1.3 µm) (Figure 2). Dry soil generally has higher reflectance than green vegetation and lower reflectance than dead vegetation in the visible, whereas, in the near-infrared, dry soil generally has lower reflectance than green or dead vegetation (Figure 2). Thus, the more green leaves a plant has, the more red visible light is absorbed, and the more near-infrared light is reflected. With this in mind, the NDVI (Normalized Difference Vegetation Index) was developed and is a commonly-used vegetation index. The NDVI is a modified ratio of the nearinfrared ( NIR ; ca µm) to visible red ( Red ; ca. 0.6 µm) bands: NDVI = NIR RED NIR + RED (3). NDVI values are unitless and range between -1 and 1 and indicate the amount of green vegetation present in the pixel. NDVI values close to 0.7 indicate more abundant, healthy, green vegetation. Spectral Characteristics of Geologic Materials As with vegetation, absorption features in the reflectance spectra of geologic materials are the direct result of electronic and vibrational processes arising from the interaction between electromagnetic energy and the atoms and molecules in the imaged material. The various electronic and vibrational interactions require different quanta of energy to proceed, so the absorption features they create are manifest at specific wavelengths. Because the absence or presence of specific absorption features is dependent on the chemistry of the material, the shape of reflectance spectra can be used to infer composition. In terms of reflectance, there is a strong dissimilarity between carbonate rocks and silicate rocks due to vibrational processes in molecular bonds. Carbonate minerals display absorption features between 0.4 µm and 2.5 µm. Calcite, for example, shows characteristic absorptions at 0.5 µm and at 2.5 µm with flat reflectance in between. By contrast, silicate minerals such as quartz generally have flat spectra in the visible to midinfrared. Other visible and near-infrared reflectance properties of rocks and minerals stem from electronic processes that take place in transition metals. Ferric iron (Fe 3+ ) is a very important transition metal in rocks. It has a pronounced absorption in the UV at 0.35 µm, with subordinate ones at 0.40, 0.45, and 0.49 µm. An additional absorption feature at 0.65 µm appears as a shoulder in spectra and is characteristic of goethite. Features between 0.85 and 0.95 µm are also characteristic of Fe 3+, for example the pronounced absorption at 0.85 µm in hematite and the pronounced absorption at 0.93 µm in goethite. In general, minerals with Fe 3+ have low reflectance between ca. 0.4 µm and ca. 07 µm, with progressively higher reflectance into the near-infrared. 3

4 For alteration minerals, Fe 3+ and hydroxl (OH - ) are important. Water in the form of both H 2 O and OH - is also important for the spectral signatures of hydrous minerals like clay. In the case of water, its interactions with electromagnetic energy do not involve electronic transitions, but instead include molecular vibrations and rotations. For example, bound OH - groups in hydrous minerals have characteristic absorptions in the ranges µm and µm. Between 2.10 µm and 2.35 µm, vibration of Al-O- H and Mg-O-H bonds create absorption features in clays that are particularly useful for discriminating hydrous minerals. Much work has been done in creating geologic indices for use with Landsat MSS and Landsat TM data. These indices, however, can be applied to other datasets as long as the bands corresponding to the correct wavelengths are used. In addition to what is in the readings posted on the class webpage (and in the class notes), I briefly summarize some Landsat band ratios below. Note that the bands mentioned are for the Landsat TM or Landsat MSS instruments. You will need to refer to the documentation for those instruments ( to see what wavelengths these bands refer to in order to translate these band ratios to other instruments. TM3/TM1 Fe-O charge transfer transition causes absorption at 0.55 mm. Maps out hematite and other red Fe oxides and hydroxides. Fe-oxides are relatively bright in TM3. TM3/TM2 Similar to above. Used for mapping ferric iron. TM5/TM4 Fe crystal field effects cause absorption between 0.85 and 0.92 mm. Maps out limonite and other Fe-rich rocks. Fe is relatively bright in TM 5. Maps Fe-rich vs. non Fe-bearing rocks. TM5/TM7 Al-OH and Mg-OH bond rotations (in clays) cause absorption in the thermal infrared (TM7). Clays are bright in TM5. Maps argillaceous vs. non-argillaceous material. In addition, clays are relatively bright in TM3, and dark in TM4. Fe is relatively bright in TM7. Most alteration minerals (Fe-oxides and hydrated minerals) are bright in TM5 (1.65 mm). RGB image of ratios using TM bands: 5/7 (R), 3/1 (G), 4/3 (B). The red channel is sensitive to clays and the green to Fe-oxides. This complex index is used for mapping hydrothermal alteration. RGB image of ratios using TM bands: 5/7 (R), 4/3 (G), 5/4 (B). The red channel is sensitive to clays, the green to vegetation, and the blue to Fe-silicates. RGB image of ratios using TM bands: 4/5 (R), 3/2 (G), 5/7 (B). Fe-oxides appear green. Clays and vegetation appear red. RGB image of bands using TM data: 7 (R), 4 (G), 1 (B). Red is sensitive to Fe-oxides, G is sensitive to vegetation, and B is sensitive to clay. The Fe-oxide ratio: Iron_Oxide_Ratio = ( TM3 min( TM3 ) ) / ( TM2 min( TM2 ) ) The clay ratio: Clay_Ratio = ( TM5 min( TM5 ) ) / ( TM7 min( TM7 ) ) TM5/TM1 Used for mapping granites. TM3/TM4 or TM5/TM4 Used for mapping low-grade metamorphic rocks. TM4/TM5 Used for mapping salinity. MSS2/MSS1 Exploits the ferric oxide absorption in MSS1 ( MSS4/MSS5 ) / ( MSS6/MSS7 ) Discriminates between limonite and sparse vegetation. RGB image of ratios using MSS bands: 4/5 (R), 5/6 (G), 6/7 (B). Used to map limonite in areas of sparse vegetation. Fe-oxides appear red and orange. 4

5 Instructions: Data Acquisition and Pre-Processing: a. You will work with data from the ASTER ( instrument, specifically the VNIR and crosstalk-corrected SWIR surface reflectance (AST07XT) and TIR surface emissivity (AST05) data products. You can order these for download from (free registration is required and the data is available at no cost). You will need to do two separate searches, one for ASTER L2 Surface Emissivity V003 and another for ASTER L2 Surface Reflectance VNIR and Crosstalk Corrected SWIR V003. You can find these by choosing ASTER under Instruments and 2 under Processing Levels in the search form. Use the map in the search form to define an area of interest you can get data from anywhere on Earth you want (if it is available). If you wish, use the clock icon in the search form to select a specific time period, but note that even though ASTER has been in orbit since 1999, it has been operated as an on demand instrument, so imagery is not available for all of the Earth and imagery for any particular place may have spotty coverage in time. Be aware that SWIR data from ASTER obtained since 2008 until the present may be unreliable. Also be aware that ASTER has obtained imagery both during the day as well as at night (mainly TIR data). Night scenes, however, have limited value for the purposes of this lab, so you are better off only looking for daytime data. The search results will show you thumbnail version of each data granule so you can assess if there is cloud cover, etc. The search results will also show you the granule footprint on the map so you can assess proper geographic coverage. Be sure to order the same data granule in both searches. To help you do this, write down the granule ID number (it is something like AST_L1A#xxxxxxx_xxxx.hdf) and the START date so that you can easily find the scene you want in both searches. Select your data and submit an order for it by clicking the gear icon next to the thumbnail. Use the defaults for the processing options. Note that you will submit two separate orders, one for each search you need to do. b. The system will send you an alert when your data is ready to download. Data should be ready for download within a couple of hours from the time you place the order. From your two orders, you should have a total of three sets of data files to download: VNIR surface reflectance, SWIR crosstalk-corrected surface reflectance, and TIR surface emissivity. Be sure to promptly download your data, following the instructions in the you get. c. Note that while they all cover the same area, the three sets of data you have do not have the same spatial resolutions (VNIR is 15 m/pixel, SWIR is 30 m/pixel, and TIR is 90 m/pixel), and they are not all contained in the same data file. This is problematic because to plot image spectra, create false-color displays, do mathematical operations on bands, etc., the data should all be in the same file and/or have the same spatial resolution. To remedy this, you will need to layer stack your three data sets using the Layer Stacking tool (see the ENVI help). Choose to resample your data to 15 m/pixel and be 5

6 sure to put the bands in the correct order (shortest to longest wavelength) before stacking. d. An additional step is to edit the ENVI header file (see ENVI help) for your stacked data to add the wavelengths and FWHM values for each band. You can get the information needed to determine these values at Use the average of the listed range for the center wavelength value and half of the listed range for the FWHM value. This step is necessary for your spectra to plot correctly as well as to aid in resampling spectral libraries (see below). Vegetation Index: a. Perform the NDVI transform using the appropriate bands from your stacked dataset. Use Band Math to perform the NDVI, not the built-in NDVI tool. b. Open the Cursor Location/Value window and explore the resulting image. What is your interpretation of dark areas vs. light areas? What about patterns of light and dark? c. Extract statistics for ROIs in your image to assess the relative health of vegetation in different parts of your image. d. To improve the interpretability of the image, you can in addition to doing contrast stretches apply a color map, and/or perform a density slice (see the ENVI help for more on these simple enhancement processes). Describe what the enhanced vegetation image map shows you. Geologic Indices: a. Pick three target materials they can be a rock type, soil, or a specific mineral that you would like to map. Access to a geologic map or other a priori information about your area of interest will help you do this. b. Once you have defined the three materials you are looking for, find corresponding laboratory spectra in the ENVI spectral libraries. Note that since the spectral resolution of ASTER is quite different than that of the ENVI spectral libraries, spectral resampling of the libraries will be necessary. You can resample existing spectral libraries and build your own spectral libraries using ENVI tools (see the ENVI help). For this lab, please use only the copies of the ENVI spectral libraries that I gave you and not the built-in ones to avoid modifying the built-in files. Plot the resampled library spectra of your three target materials using the Spectral Library tool in ENVI. Pay attention to wavelengths where there are major absorbance features (use the NDVI as an example of what to look for). c. Based on what you see in the library spectra for your three target materials, define your own geologic index for each one using the NDVI as an example. Recall that an index can be a simple two-band ratio (like the NDVI). Or it can be a more complex ratio involving more than two bands. It can also include addition and subtraction of values. It can be an RGB color composite image of selected bands. Or it can be an RGB 6

7 image of selected band ratios. The possibilities are limitless. Your goal is simply to define a useful way of manipulating the bands in your image to highlight a ground target of interest based on the spectral properties of that target. Alternatively (or in addition), you can try out the Landsat ratios listed above ( translated to ASTER wavelengths, of course), or search the literature for others. Regardless of what you choose to do, be prepared to explain how you constructed the indices you will use for each of your three target materials, i.e. justify the bands you choose and your methods of combining them. Also explain how you will interpret the results, i.e. what range of output values are expected, what values are consistent with your target material, what ambiguities might exist, etc. Remember that you need to come up with three indices, one for each of your target materials. d. Once you have figured them out, write down equations for your three indices. Then, to implement your indices in ENVI, you will have to make use of the Band Math tool (see the ENVI help). e. Apply your indices to your stacked image data. The DN of each pixel in the resulting image map will be the value of your index at that location. You may find that a further processing step such as a contrast stretch (for grayscale and RGB images), a threshold/density slice (for grayscale images), or the application of a color map (for grayscale images) may improve the interpretability of your output index images. f. Assess the effectiveness of your indices, and, if necessary, make adjustments to them and re-process your data until you are satisfied with their performance (or are convinced of their uselessness). In either case, hypothesize on reasons for your success (or lack thereof). What To Turn In And How: Prepare a short report outlining the processing you did in ENVI. Limit your report to no more than 3 typed pages (normal margins, 12 point font, reasonable spacing), not including figures or references. Include in your report answers to any questions posed in the instructions and representative images (inline with the text is preferred) showing the results of your processing (e.g. vegetation image maps, lithologic image map, tables, spectra, etc.). Only include those images that are necessary to illustrate what you did i.e. if you don t talk about it in your report, I don t need to see it. Follow the general format of a Geology paper for your reports, including the formatting and presentation of figures. Submit your report as a PDF named lab6<yourname>.pdf that you should place in the dropbox. 7

Course overview; Remote sensing introduction; Basics of image processing & Color theory

GEOL 1460 /2461 Ramsey Introduction to Remote Sensing Fall, 2018 Course overview; Remote sensing introduction; Basics of image processing & Color theory Week #1: 29 August 2018 I. Syllabus Review we will