Remote Sensing

Looking at the Hanksville (UT) area with Landsat data and the Davinci program!

Introduction

My work with Team Argon would greatly benefit if a detailed geologic map of the area surrounding the competition site was made. This project attempts to identify the mineralogy of this area, to the best extent Landsat will allow with its limited spectral resolution. The command-line program Davinci was used to do this project. Code inputted to the program is displayed in bold. The data were given as raw digital number, so they have to be converted to reflectance before the investigation can begin. This page will reference landsat spectral bands; a more detailed description of each band, including its wavelength, can be found Here. Note that this investigation used Landsat 7 data with 6 spectral bands: 1-5 and 7. Band 6 (thermal) was not used, and from here on out if the term "band 6" is used, it actually refers to band 7.

Figure 1: the visual image. The competition site is inside the "triangle" seen in the image, and is in the eastern half of this triangular-shaped region.

converting digital number to radiance

Converting raw digital number to radiance is done by searching the metadata to find the maximum and minimum radiances, as well as the maximum and minimum quantized calibrated pixel values, for each spectral band of data. An equation was used to convert the data to radiance, and the radiance data was concatenated to a file contained within the structure in Davinci. The specific equation used was b1=((LMAX_λ – LMIN_λ) / (QCAL_max – QCAL_min)) * (temp.data[,,1] – QCAL_min) + LMIN_λ where the filename was temp.data . The Lmax, Lmin, Qcalmax, and Qcalmin files were found in the metadata. [,,1] means the entire image file in the first band. This was done for all 6 Landsat bands, with the number in brackets, as well as the number after 'b', changing. After this was done, each of these new files (b1 through b6) was concatenated along the z axis using the following equation: temp.radiance = cat(b1,b2,b3,b4,b5,b7,axis=z) .

Figure 2: This is a spectral plot of vegetated farmland near Hanksville. Healthy vegetation has a very distinct spectral shape, so it is being used to ensure these methods are being done correctly. This plot is normal for spectra that have not been atmospherically corrected, which is the next step of this investigation.

atmospherically correcting the radiance data

In order to perform the atmospheric correction, a dark pixel in the shadow of a cloud is to be used, preferably over deep water. Since there is no deep water in the image, the darkest pixel must be selected from shadows of clouds on land. I plotted the spectra of several dark pixels in the image and chose the darkest spectra to be used for my atmospheric correction, which was made by subtracting the dark pixel from the rest of the data using the following equation: temp.atmcorr=temp.data-temp.data[XX,YY] where (XX,YY) is the pixel to be subtracted.

Figure 3: Atmospherically corrected radiance data. This shape is indicative of healthy vegetation, meaning this technique was done properly. The only remaining step is to convert radiance to reflectance. Once this is done, the true investigation can begin.

converting radiance to reflectance

The equation to perform this operation requires the Earth-Sun distance on the day the image was obtained, as well as the Sun’s angle of inclination. These values were obtained using the metadata. Also required is the spectral solar irradiance for each band. The equation to perform this conversion is b1=(3.14159*(d*d)*landsat.atmcorr[,,1])/([solar irradiance]*sind([sun's angle of elevation])) Like when digital number was converted to radiance, this was done for each band and the bands were concatenated using the same concatenation equation as before.

Figure 4: Spectral plot of atmospherically-corrected reflectance. The red plot is of farmland near Hanksville, and is the same pixel used in the previous two images. The green plot is of vegetation west of Hanksville. These values make sense for the reflectance of healthy vegetation. This figure proves

that my data has been correctly calibrated to reflectance, and the data can now be analyzed with the goal of identifying the mineralogy of the region near Hanksville.

attempting to identify mineralogic properties of the hanksville region

To plot spectra, the pplot() command in Davinci is utilized. Specific code may look something like this: pplot({temp.reflectance[XX,YY,],temp.reflectance[XX,YY,],temp.reflectance[XX,YY,],temp.reflectance[XX,YY,],temp.reflectance[XX,YY,]},xaxis=temp.xaxis,{"1","2","3","4","5"}) This works for plotting 5 separate spectra, the maximum amount that can be done with labels. To plot only 1 spectra, the following command can be used: pplot(temp.reflectance[XX,YY,],xaxis=temp.xaxis

Figure 5: This is a plot of 5 spectra close to the compeition site. Wavelength in micrometers is on the x-axis, and reflectance is on the y-axis. These spectra are all extremely similar in shape, so I need to broaden my search!

Figure 6: This is a plot of five other spectra farther away from the competition site. While these spectra are still similar, their shapes contain enough differences to perform a more detailed analysis.

performing a decorrelation stretch on the image

A decorrelation stretch (DCS) is used to maximize differences in spectral bands. This technique is performed in three spectral bands. After studying figure 6, I decided that the three best bands to perform a DCS were the 3 (~0.8 micrometers), 5(~1.7 micrometers), and 6 (~2.2 micrometers) bands. I used the following code to produce the DCS (Figure 7): temp.dcs=do_dcs(temp.reflectance,3,5,6)

Figure 7: Decorrelation stretch of figure 1 in the 3, 5, and 6 bands. This is a false-color image used to identify variations in the different regions of the image.

Figure 8: Spectral plots done based on figure 7. The numbers correspond to the numbers on figure 7 to show the location of each spectra plotted. Note that spectra cannot be plotted off the DCS; they must still be plotted from the reflectance image.

The spectra shown in figure 8 display some unique differences, which can further be used to analyze these regions. For instance, spectrum 4 (the pink area on figure 7) is the only area with a negative slope between bands 4 and 5. It seems that this is the only area on this image where this occurs. There seems to be the most variation in the first 3 bands, which can be helpful in determining things such as iron oxidation state. This, a 3-2-1 DCS was performed next.

Figure 9: 3-2-1 DCS of figure 1. The rest of this page will describe the spectral variations associated with each color of this image. Note that the areas near clouds will not be investigated due to interference from the clouds.

interpreting this geologic map

Using the pplot() command in Davinci and studying the Landsat spectra, here is what can be interpreted from this geologic map:

The peach-colored areas of this map correspond to healthy vegetation. They are few and far between in this area, which is mostly desert. Some areas with this color can be seen on the southern side of the "triangle", which correspond to farmland. See figure 4 for the spectra of two regions with this color.

The dark purple region to the north of this image is a very interesting region. This is the area corresponding to spectrum 5 (the bottom spectrum) on figure 8 (see figure 7 for the exact location of this region). This region has a spectrum which is extremely concave-up in the first three bands of the spectrum, which most likely means iron has been oxidized in this region. A spectrum was also taken on a region of similar color towards the southwest of the image, and the spectra were very similar, suggesting similar geologic makeup. It should be noted that spectra from regions with this color have a much lower reflectance than spectra in regions with other colors on the DCS...

The pink region of this area is also very unique. This area corresponds to spectrum 4 on figure 8 (it is not the magenta region near the clouds, which was not investigated) and appears to be oxidized. This is the only area on this map with a negative slope between the fourth and the fifth band. The exact meanings of this with regards to mineralogy is currently unknown at this time, although this is certainly something to be further investigated.

The red region of this area also contains oxidized iron, though its spectra isn't as concave as the dark purple region. This region corresponds with spectrum 1 on figure 8 and the surrounding area. The similar-colored areas to the east of the image also have a similar spectrum, though they are not quite as oxidized.

The yellow region of this area corresponds to areas that are not oxidized. There are two main regions on the map with this color (see figure 10). One area is to the northeast, and the other is to the southwest.

The orange regions on this map generally correspond to areas that are only slightly oxidized, or not oxidized at all. It seems from these spectra that areas with a more red-orange color are slightly oxidized, while yellow-orange areas are not oxidized. This makes sense considering pure yellow spectra are not oxidized, while pure red spectra are slightly oxidized.

The gray regions on this map also correspond to areas that do not contain oxidized iron. An example of a spectrum in this region is spectrum 2 on figure 8. While these regions generally have the same spectral shape as the yellow spectra in figure 10, their spectra are not as "steep" in the visual region, which may be why their color differs in the DCS. As with yellow and orange, spectra were plotted in multiple regions colored gray, and their spectra looked very similar, especially in the bands the DCS was done in.

The light purple regions on this map correspond to regions which are also not oxidized. Their spectra seem to be similar to the yellow and gray regions, and include some of the spectra in figure 5. The competition site seems to be in an area of this color. I took a spectra of an area I believe to be near the competition site, and my result is figure 11.

Finally, the green regions on the map also correspond to non-oxidized regions. I took spectra of regions with this color near the competition site (and to the south in this same region), as well is inside the "rabbit" shaped region slightly west of there and the yellow-green areas in the far northeast of the image. My results are shown in figure 12. Also shown in that figure is a spectrum taken in the dark green region to the southwest, which differs from the previous four spectra.

Figure 10: Spectra of regions colored yellow on figure 9. While their shape slightly differs in the infared, the straight line in the visual means there is no oxidation, and is likely why both of these regions have the same color.

Figure 11: This is a spectrum of an area colored light purple on the DCS, near the competition site (may be slightly north of it). Its spectra is a bit different in the near-infared. More research is required to know why this is.

Figure 12: Various spectra from regions colored green in the DCS. The four spectra to the top look very similar, despite being from different regions on the map. The pink spectra comes from the dark green region on the far southwest of the map and appears to be oxidized, similar to the dark purple region. However, this area is very close to clouds, so they may be interfering with this spectrum.

conclusions, implications, and future work

For this project, I used Landsat data and the Davinci program to make a geologic map of the region near Hanksville, UT and attempted to analyze spectra to determine the mineralogy. I found that most of the spectra near this area follow the same general shape, shown in figures 5,6,8,10, and 12. I also performed a DCS (figure 9) and found that areas in the DCS colored dark purple, pink, red, some orange, and perhaps dark green contain oxidized iron in their mineralogy. Future research is needed to determine what exactly the mineral makeup is of this image, and what each spectrum means. It should also be noted that the results outlined above are only for areas on the map with a large extent; zooming in on the map shows small regions of different colors, and spectra such as the one in figure 11 resulted from those areas. Analyzing these spectra in further detail may also yield interesting results.