Using LandSat data to make color composites

We will be working today with LandSat images. As you learned in lecture, LandSat is an example of multi-spectral imagery, which records reflectance of EM radiation from the earth's surface in visible wavelength bands, as well as in several near- to mid-infrared bands.

Before we start to work with the LandSat images, though, we should learn a little about image data in a GIS.

Thematic data vs. images

You worked with vector data last week, which is made up of points, lines, or polygons drawn on a map. The second major kind of GIS data is called raster data, which is made up an array of "pixels", which are usually square cells. Each pixel has a location recorded at its center, and a data value associated with the pixel. Raster data can be used in the same way as our land use layers last week, to represent different categories of land cover. If we had a raster representation of the land use in 2016 at the mouth of the watershed, it might look something like this picture to the right. Each color represents a different cover type, and the pixels are large enough that you can see them individually. Since the pixels hold codes that indicate the cover type, this is considered thematic raster data - it is not just raw measurements of light reflecting off of the surface, it has been interpreted and converted into a set of land cover categories.

The picture to the right shows you how the pixel values would be assigned - the black line represents an edge between two adjacent polygons with different cover types, and the grid represents the pixels that will be used to represent these cover types. All of the gray pixels have their centers in the polygon to the left of the black line, and all the blue pixels have their centers in the polygon on the right. The assignment of pixels to cover type is based only on the center of the pixel, and pixels overlap into the adjacent cover type at various points along the edge. Pixels that cross over the line are are only partially accurate, because they cause some gray to fall into area that should be blue, and vice versa. Because of this effect, in a thematic raster map the most accurate part of the pixel is the center, but edges of pixels may not accurately reflect the cover type.

Vector data is the opposite - when you digitized polygons in the last lab, you recorded vertex locations by clicking along the boundaries of the features you saw on the screen, so the place where you were making positive choices to record information about cover type was at the edge. The insides of polygons are treated as though they are homogeneous, containing whatever cover type is assigned to the polygon, and because of this you can expect to find small patches of other cover types within a land cover polygon (any patch of cover that is smaller than the minimum map unit won't be recorded, for example). Vector polygon data is most accurate at the edges.

Digital images are another form of raster data. Images are structured the same way as thematic raster data - images are arrays of pixels with a single value each - but images are a little different in a couple of ways:

• First, the imaging sensors record all of the photons reflected off of the earth's surface within the entire pixel, so it's best to think of the pixel value as an average intensity within the entire pixel, rather than being the value at the center.
• Second, although we may have no trouble interpreting what feature or cover type is present in the image when we look at it, the image only actually contains data about reflectance of EM radiation within pixels. There is no thematic information in an image, we have to supply that by interpreting the image in some way.

We are going to work with LandSat images today to learn how to use various bands to create false-color composites. We will still derive meaning from the images based on manual interpretation of the images - false-color composites are a visualization technique that make it easier for us to see particular cover types, but they do not convert the image into a thematic raster map. However, use of false-color composites can greatly enhance our ability to manually interpret images by allowing us to see bands that are outside of the visible part of the EM spectrum. Cover types that reflect visible light similarly (i.e. are a similar color) may reflect infrared bands differently, which can make it much easier to tell those cover types apart. We will learn to classify pixels into cover type categories based on their "spectral signatures" to make new cover type maps soon, but for now focus on understanding how the available LandSat bands give us different information about cover type that we can use to see what would otherwise be invisible.

LandSat

LandSat has contributed tremendously to our ability to monitor our world. LandSat data has always been available to the public, but before widespread availability of broadband internet, distribution charges were fairly high. In the last few years the USGS has made LandSat data available for download free of charge to anyone who wishes to use it. The image to the left shows the Rim Fire, which burned near Yosemite National Park in 2013. Between the two LandSat satellites that are currently operational (LandSat 7 and LandSat 8), each location on the ground is imaged once every 8 days. LandSat 7 developed technical problems that make the data difficult to use, but LandSat 8 records each location every 16 days. LandSat satellites have been in operation since the early 1970's, which makes it possible to use LandSat data to monitor changes on a variety of time scales, from decades-long trends in cover type conversion, to monthly or seasonal variation in vegetation, to nearly real-time monitoring of acute events like fires. Over the next two lab meetings you will learn how to use LandSat data to visualize various aspects of land cover on a landscape, and to quantify the amount of change in land cover over time.

Obtaining LandSat Data

To save time, I will be providing you with the data you need for today's lab, but take a few minutes to look at how LandSat data are distributed. There are a couple of web sites that USGS uses to distribute their publicly available data, one of which is the Earth Explorer web site. If you're interested in seeing how it would be done, read the gray box below - otherwise, move on to today's activity.

Downloading LandSat data

USGS asks that you register for a free account to download LandSat data, but you can go through every step of the process of finding and selecting images without one, you will just be unable to actually download the scenes you select - these instructions will walk you through how you would obtain LandSat scenes, but don't feel obliged to create and account unless you want to. You can reach the EarthExplorer web site here - the link should open in a new window with a world map on the right, and controls on the left.

The first step is to identify the location for which you would like to obtain data, which you can do by panning (i.e. click-holding the map and moving it to place the location of interest in the center of the map) and zooming (by clicking the + or - buttons, or moving the slider up and down on the zoom control). Zoom in until you have a nice view of coastal San Diego County (with Tijuana at the bottom of the map, and Oceanside at the top).

Now, you can identify a coordinate of interest by simply clicking on the point on the map you want. Click on San Marcos (the exact location isn't important, as each snapshot, called a "scene" in LandSat terminology, cover a fairly large area, and San Marcos is in the middle of the scene). This will place a marker on the map, and will record the coordinate in the "Polygon" control on the left, and then by clicking "Use Map" the coordinates of the corners of the visible area in your map are entered as the search area.

You can also specify the date range if you want, which by default will start at 1/1/1920 and end on today's date. It's not a bad idea to change the range to a narrower window, because search results are limited to 100 scenes at a time, and later data will not be shown if you start too early. We'll be looking at LandSat 5 TM scenes, so you can start with 01/01/1984 and end with 01/01/1994. You could also narrow the search by months if you were only interested in a particular time of the year, such as springtime.

Next, click on the "Data Sets>>" button at the bottom of the controls on the left. The controls switch to the next tab in which you get to specify which data you will download. Find the "Landsat" option and expand it, then expand Landsat Collection 1 - these are the best available images from this date range, with the least amount of cloud cover and processed to the point of being ready to use. Check the box next to L4-5 TM to see the LandSat 4 and LandSat 5 Thematic Mapper scenes available for your date range at this location. Collection 2 scenes are still made available, but they have defects that make them less usable, such as cloud cover or optical defects. For our region it is common for cloud cover to occur over the ocean and to cover part of the coast, which might put the image into Collection 2, but if your particular project area falles in the mountains of desert this may not actually be a defect from your perspective - depending on your needs, if you are unable to find what you want in Collection 1 it can be fruitful to also search Collection 2 images.

The other options are data from other satellites with different types of sensor; the L1-5 MSS are the earliest, dating back to 1972, but the sensors didn't have all of the bands we would like to use, so we'll stick to TM scenes (the LandSat 5 satellite had two types of sensors, the MSS and TM, so it's included in two different options here). Notice also that there are a lot of different types of data available - we don't have time to learn about all of them, but if you're interested the arrows in the orange boxes next to each option are links that will take you to web sites that describe what they are.

Now, click on "Additional Criteria>>" to go to the next tab. Here you can limit the scenes that you'll see to particular paths, amounts of cloud cover in the scene, and various other criteria. The path refers to the path of movement of the satellite with some overlap between adjacent paths, and the row refers to where the snapshot was taken. Since there is some overlap between adjacent paths and rows to make sure there is complete coverage of the ground, you may be working in an area covered by more than one scene, and can choose which you want to use here. It's fine to leave these criteria blank.

Finally, click on "Results>>" to see the list of scenes available. They are listed in reverse chronological order, with the first one listed taken on 01/01/1994, and the last one listed taken on 3/23/84.

Each available scene has a thumbnail image, and you can click on it to get a larger preview image of the scene, and to get a look at the "metadata" that describes the characteristics of the scene. Depending on your needs there are several pieces of information that might be of interest, but the big things are the quality values (9's are good), and the cloud cover (0 is good). Close the preview window if you opened it, and look at the buttons next to the thumbnail images. The first on the left is a foot - clicking on this will zoom to and show the outline of the "footprint" of the image on the map. The next one to the right superimposes the scene on the map, so you can see precisely what will be covered (the footprint is a square that includes the scene, but doesn't have the same dimensions as the scene - click both for the first image to see how they are related). The second from the right is the one you would click to download the scene. If you try it you'll be told you have to be logged in to download - signing up is free, but don't worry about this for now, I have already downloaded the data for today. You're allowed to select more than one scene to download at once, so the last button (red circle with a line through it) allows you to exclude scenes you aren't interested in, and download the rest.

If you did have an account and logged in, clicking on the download options would bring up a window that allowed you to select from a set of various color composite images, or the "Level 1 Product" that contains all seven of the available bands for the scene. We want all the bands, so you would choose the Level 1 product.

File format and organization of data

LandSat scenes are distributed as a "tar.gz" archive. This file extension is used for collections of files that have been combined together in "tape archive" (tar, for short) form, and then compressed with "gzip" compression (gz, for short). IF you had downloaded a file, then you would find it on your disk in Windows Explorer, and then right-click on it. In the popup menu, you would select "7zip", and unzip the tar.gz file into its own folder. I've done that for you for today.

The LandSat scene we will work with today was taken on 4/15/11. I selected that date because it was a Spring scene, with relatively cloud-free views of the project area over the San Dieguito River watershed.

To see the file formats and organization, start ArcCatalog and open your P: folder connection (these should be saved and available between uses, but if you lose the folder connection you can re-create it - the folder to connect to is "Public on Viking (P:)" → "Biology" → "kristanw" → "biol420620"), and then open "lab3". You will see that there are several files within it.

• The actual images are "TIFF" files (short for "tagged image file format", which uses a .tif filename extension), and there are seven of them, one for each band. I have re-named the tif files as "sp11_bandY.tif", where Y is the band number. These are actually a special version of TIFF file called a "geoTIFF", which are used for images of the earth's surface, and contain information about the location of the image on the earth's surface, the sizes of the pixels, and so on.
• The other file names are are all the same for the first string of 20 characters - before I renamed them to something more interpretable all the tif files had this same first 20 characters in their filenames. When I converted the tif files to geoTIFF these additional files became unnecessary, but I kept them in the folder for you to get a better idea of what is distributed in the tar.gz files. For example:
• The XXX_VER.txt file gives information about how the LandSat image was "registered" to the real world - the images are of a curved, topographically complex surface that was viewed at an angle, and the images have to be flattened to make them properly display on a 2-dimentsional surface, like a paper map or a computer screen.
• The "XXX_MTL" file gives the "metadata" associated with the file, including all the information needed to calculate reflectance from the raw data.
• The "XXX_GCP" file gives a report on how the image was registered to "Ground Control Points", which are identifiable features in the image with known locations on the ground.
• If I hadn't renamed them, the bands would have had names like LT50400372011105SPAC01_B1.tif for band 1, and the others would have been the same except for a B2, B3, etc. in place of B1  after the underscore.

If you recall, LandSat images record not just visible light, but also several bands in the infrared portion of the spectrum. For your reference, this table and the graph to the right of it give information about each band (the second row of boxes from the bottom, labeled TM, are the bands from the Landsat 5 TM sensor that we are using):

Band No.	Wavelength Interval (µm)	Spectral Response	Resolution (m)
1	0.45 - 0.52	Blue	30
2	0.52 - 0.60	Green	30
3	0.63 - 0.69	Red	30
4	0.76 - 0.90	Near IR	30
5	1.55 - 1.75	Mid-IR	30
6	10.40 - 12.50	Thermal IR	120
7	2.08 - 2.35	Mid-IR	30

Notice that the sixth band is an oddball - it covers a wider range of wavelengths and has a much bigger cell size than the rest. This is the "thermal infrared" band, which is used for measuring surface temperature. It's not terribly useful for the kind of land cover type assessments we will be doing, so after we take a look at it we're going to leave it out of the rest of our work.

Using LandSat in ArcMap

For today, we are going to focus on learning how to work with LandSat data, and how to combine layers into true- and false-color composite images.

1. Open ArcMap with a new map. Start ArcMap from CougarApps. When the "Getting Started" screen appears, check that the default geodatabase is still set to your monitoring.mdb file (if not you can set it now), then switch to the "New Maps" → "Standard Page Sizes" → "North American (ANSI) Page Sizes" → "Letter (ANSI A) Landscape" map template, and click "OK".

Once the map window appears, click on the "Data View" icon in the lower left corner of the map pane - you will no longer see the map template, and will only have one set of coordinates showing as you move your mouse over the map pane (if you recall, the data view only shows geographic coordinates, and doesn't show position on the map page).

2. Add LandSat bands to the map. We will add all seven bands to ArcMap.

• Click the "Add data" button and navigate to the lab3 folder on P:. 
• Navigate to the lab3 folder, and you'll see the seven bands are listed with icons to the left of them that are grids of squares - the icon represents pixels, and identifies the files as raster data. You can select them one at a time by holding down SHIFT while clicking on each name (all the sp11_band1.tif through sp11_band7.tif files, none of the others) and then clicking "Add".
• Put them in order in the table of contents, with band 1 at the top - if you end up with them out of order, you can drag them around in the table of contents. If you're asked when you add the layers to the table of contents whether you want pyramids built, say yes - this makes the rendering on the screen much faster.

3. Open the image analysis window. ArcMap organizes its image analysis into a single window, which you can activate from the "Windows" menu - the "Image Analysis" option is the last option in the list, like so:

You'll see a window pop up like this:

You can dock this window out of the way on the right side of the map panel - if you graph the window by its title bar (the blue-gray band at the top) and start dragging it around the window you'll see that at the right, left, top and bottom edges of the ArcMap window some indicators pop up that look like this: . This one is on the right side of the window. If you click, hold, and drag the image analysis window on top of this indicator, wait until you see a blue-gray rectangle appear to show you where the window will be docked, and then drop the window there, you will get the Image Analysis window to dock out of the way along the right edge.

4. Compare individual layers. Each of the seven bands should be listed in the box at the top, and if they are in order fro band1 at the top to band7 at the bottom the map is displaying Band 1 - like the table of contents, the top-most layer is drawn last, and is thus on top of the rest, and what you are seeing is the EM radiation recorded for Band 1.

If you look at the table above giving wavelengths recorded by each band, Band 1 is blue visible light. Since only a single band is being displayed, ArcMap is rendering the data as a gray-scale image. Light areas indicating high reflectance of blue light, and black areas indicating low reflectance of blue light.

Look over band 1 and notice where the variation in shade occurs in the image. Some features seem pretty easy to pick out - this suggests that have very different amounts of blue light in their color compared with their surroundings, and the contrast makes them easy to see. For example, you can probably identify highway 15, which is a thin white line about 1/3 of the way from the west edge of the image.

We can't use the trick we used last time to make the watershed polygon see-through so that we could see the image behind - all the images fill the entire area. One way of comparing them is to turn the top layer on and off - if you un-check sp11_band1.tif in either the image analysis window or in the table of contents you'll be able to see sp11_band2.tif, which is the next highest in the drawing order.

This doesn't always work so well, though, because big images can take a long time to draw. It's possible to get a better, more convenient comparison using the swipe tool. Do the following:

• In the Image Analysis window, select sp11_band1.tif.
• Click on the swipe tool to activate it - it's in the "Display" block of the image analysis window, just above "Processing, and it looks like this, .
• Now, click and hold over the image - you'll see that band 1 is displayed from the top down to your mouse pointer, but band 2 is displayed below that line. If you hold your mouse button down and move the mouse you'll see the dividing line between the layers moves with your mouse.

As you look at features that stand out clearly on Band 1, note that sometimes they are not as obvious in Band 2. This happens when the colors differ in the amount of blue (recorded by Band 1), but not in the amount of green (recorded in Band 2) that make them up, and because of that you can tell them apart using just Band 1, but not using only Band 2. The converse will also be true - some colors are very different in the amount of green they contain, but have the same amount of blue - you will be able to tell them apart using Band 2 but not Band 1.

Turn off band 1 (un-check it), and set band 2 to be the swiped band by selecting it in the Image Analysis window. As you swipe away band 2 you'll see that band 3 (red) shows different patterns than band 2. Particularly note how the water bodies change in appearance - band 3 is very strongly absorbed by water.

Continue to turn off layers and select the top-most visible layer to swipe until you have gotten through all of them. Note how the contrasts change as you look at each successive band. Pay close attention to Band 4 and above, because those are infrared and not visible to the human eye - there are some features that look quite different in the infrared part of the EM spectrum, and we can take advantage of that to help us distinguish land cover types with similar colors.

5. Create a "true color" composite. Now that we can see that each layer shows us something different about land land cover, we can combine three bands together into color composites that will be even more informative.

Remember from lecture that computer monitors make all the colors we see by mixing red, green, and blue light together - on the computer, we refer to these as the red, green, and blue channels. We can make a color image from our Landsat bands by assigning them to the color channels on the monitor.

To get ArcMap to do this assignment of bands to channels, we first need to make a color composite from the Landsat bands - simply, we just need to group together all seven bands into a single composite image, and then we can assign the layers as we like to the color channels as we like.

• Check the boxes next to all seven bands, and select all of them in the Image Analysis window.
• In the Processing section of the Image Analysis window (second block from the bottom), you will see several buttons, and the third from the left looks like three sheets of paper held in parallel with one another, like this: . If you hover over the button you'll see this is the "Composite Bands" tool. With all seven of the bands selected, click this tool.

You should see a new item called "Composite_sp11_band1.tif" added to the table of contents, and to the Image Analysis window. You should also see that the image now looks something like a color image, but the colors are off. To make a true color image, we need Landsat band 3 to be assigned to the red color channel, Landsat band 2 to be assigned to green color channel, and Landsat band 1 to be assigned to the blue color channel - when we made the color composite bands 1, 2, and 3 were assigned to Red, Green and Blue, which means that band 1 and 3 are assigned to the wrong color channels.

**Note: there appears to be a bug in ArcMap that changes all of the band names to Band_1 - this is confusing, but they are in the same order as in the Image Analysis window, so the first Band_1 is indeed Band 1, the second Band_1 is actually Band 2, and the third Band_1 is actually Band_3, and so on.**

To assign the bands to the right color channels in the map, click on the red square (which is the red color channel for the monitor) next to "Red: Band_1", and select the third entry in the list (that is, the third Band_1 entry in the list, which is actually Band 3). Now do the same for blue (which is the blue color channel for the monitor), but assign the first Band_1 to it (which is in fact Band 1, which is visible blue light). Band 2 was correctly assigned to green, so no change is needed. The result should look like an aerial photograph, with colors that match what you might see yourself if you were looking down from space.

The name assigned to the composite is pretty uninformative, but we can change it:

• Select "Composite_sp_11_band1.tif" in the Table of Contents, right-click, and select "Properties".
• In the "Layer Properties" window, switch to the "General" table, and change the "Layer Name:" to "Landsat_composite_sp11".

This is not a permanent file, by the way - if we quit without saving a map file and re-started it would be gone. That's not a big deal because it's easy to re-create. If you do save a map file for this lab and then open it again later, ArcMap will re-create the composite for you, but it would not save a seven-layer image to your computer, it would just make a composite for display "on the fly". If you wanted a permanent version of this seven-layer image you would need to export to a file, or to your monitoring.mdb database. We don't really need a permanent version of it, so we will just use this on the fly composite instead.

6. Compare the true color Landsat composite to a high-resolution layer.

Zoomed to the watershed level, the image looks pretty crisp, but remember that each pixel in this image is 30x30 m. We will use imagery that comes from the USDA's National Agricultural Imagery Program (NAIP), which produces high-resolution aerial imagery for much of the US every few years (we have an image from 2012). It has the same 1 foot resolution as the World Imagery layer we used for digitizing polygons last time, but it also has a fourth band that records near-infrared EM radiation - we'll make use of that in the next step. But first, let's see what Landsat's 30x30 m pixel size does to your ability to manually identify features on the ground.

First, un-check the boxes next to all of the bands, but leave the composite turned on - every time ArcMap needs to update to reflect a change you've made it has to re-draw all of the checked layers, and that can be slow. Un-checking the layers you don't need to look at speeds up redrawing a lot.

Then, add the "NAIP 2012 4Band.lyr" file (which is actually a file that tells ArcMap where on the internet to get the imagery - it is hosted on the State of California's servers). Put it below the Landsat composite.

Also add the "Athletic_fields.shp" layer, then right-click on it and zoom to its extent - this will take you quickly to an area with some features that are easy to see with the 1x1 foot pixel NAIP image, but look blocky and pixellated using the Landsat image. Using the NAIP image (with the Landsat composite turned off) you will see San Pasqual High School in Escondido, and a little of the surrounding housing developments to the east, a golf course to the south, and Kit Carson Park to the west.

If you turn on the Landsat composite now there is still some color variation, and if you know what's there you can tell that the blob of whitish gray is due to the roofs of the high school buildings, and the greenish blobs to the south and east are golf course and athletic fields, but if you had only this image you would have difficulty knowing what you were seeing - a few 30x30 m pixels are all that is needed to cover the entire football field, and that is certainly not enough to tell what it is. But, minimally you can tell that there is some green over the football field and over the

7. Create a color infrared "false color" composite. Any composite that assigns bands other than red to the red channel, green to the green channel, and blue to the blue channel is considered a "false color" composite. We sort of did one already when we composited the Landsat layers and accidentally assigned the blue layer to the red channel, and the red layer to the blue channel. But, the purpose of false color composites is not to just make trippy images, it's to allow us to see parts of the EM spectrum that we can't see with our eyes.

A very common false color assignment is to assign the near-IR band to the red channel, the visible red band to the green channel, and the visible green band to the blue channel. This is the color infrared false color composite, and it is so commonly used that it is sometimes called the standard false color image.

Go ahead and do the assignment for the NAIP image - Band_4 (near IR) to red, Band_1 (visible red) to green, Band_2 (visible green) to blue (the band numbering is different for NAIP than for Landsat). You'll see lots of dark red in undeveloped areas, and really bright red in golf courses, and other well-irrigated places. This color infrared combination is good for identifying healthy, growing vegetation from unhealthy vegetation, and vegetated areas from unvegetated areas.

Zoomed into the althletic fields layer (do so if you weren't already) you'll see something interesting - even though the football field was brighter green than the baseball field using the true color assignment, it is not red at all when we use near infrared radiation. This is because the football field uses artificial turf, which is green in color but not photosynthetically active. We will be able to make important distinctions using infrared bands that we would not be able to make using only visible light.

Go ahead and make the color infrared assignment with the Landsat composite now - the bands are numbered differently, so the assignment is Band 4 to red, Band 3 to green, and Band 2 to blue. At this zoom level you will be able to tell that the blob that is the football field is blue-gray, while the blob that is the baseball field is red - the individual pixels do tell us something about what is happening on the ground at that point, but the coarse resolution makes Landsat imagery unsuitable for visually interpreting fine detail.

Zoom out to the extent of the Landsat composite, and you'll see again that variation in the intensity of red is revealing differences in healthy, photosynthetically active vegetation and dead or dormant vegetation that was not so obvious before.

8. Create additional false-color composites. Just as the Landsat 4,3,2 composite is good for green vegetation, other combinations are good for other things. A summary of some of the common ones is in this table:

Zoom out to the extent of the Landsat composite, and then do each one of these band combinations. For each one zoom in and out to do a closer inspection of some of the conspicuous features - pay close attention to the lines of separation between different colors, as those are edges between cover types that we would like to be able to see for mapping purposes - some edges will be easy to see with one composite but hard to see with another. Note that you won't be able to do these band combinations with the NAIP image because it only has four bands, but you can still use it to help you interpret what you're seeing in the Landsat imagery (the swipe tool is helpful for this).

We're just scratching the surface on this topic - if you're interested and would like to learn more about selecting band combinations for various uses, see this article.

All done

Save a lab3.mxd map file so you can refer to it in your mini project 1 report.