Quantifying Land Cover Changes in Maine

Transcription

1 Quantifying Land Cover Changes in Maine! STUDENT HANDOUT Introduction Change detection tools enable us to compare satellite data from different times to assess damage from natural disasters, characterize climatic and seasonal changes to the landscape, and understand the ways in which humans alter the land. In this exercise, you will study a real-world situation in which change detection techniques were applied to track an economically important land cover change in Maine the expansion of commercial blueberry farming. Municipal and regional planning officials, conservation organizations, and agricultural scientists and economists often need to know not only where agricultural practices occur, but also how agricultural practices change over time. Unfortunately, most land cover data provided by state governments are static, and federal agricultural census data are not geographically referenced. By combining satellite data and ground observations with GPS and other tools, we can document and quantify such changes. Maine is the nation s largest producer of wild blueberries. Very large areas of originally wild blueberry barrens are now managed commercially, especially in the northeastern region known as Downeast, which is well-suited to this crop because of its thin acidic soil. In recent years, the price of blueberries has risen, prompting Downeast farmers to clear additional land for blueberry production. Many are removing rocks and flattening their barrens to allow for machine harvesting, since migrant labor has become scarce. Many are also installing below-ground irrigation systems in order to increase yields with automated watering. These changes in the types and extent of blueberry culture in Maine have led to several concerns that can be addressed with remote sensing technologies. First, blueberry cultivation near waterways can negatively impact water quality with erosion, removal of shade trees, and run-off contaminated by pesticides. Such damage can be particularly problematic for the area s rare and endangered fish populations and coastal shellfish areas. Also, because blueberries account for a sizable portion of Maine s economy, agricultural scientists and economists need to be able to predict annual yields and advise growers and buyers on pricing. Remote sensing can help us examine not only the extent and type of blueberry cultivation, but also how it is changing over time. This allows for better management, pricing and planning for the future of the industry and the region. In this exercise, you will use Landsat data for Downeast Maine to look for changes in vegetation cover over a four year period. The scenes for this exercise have already been downloaded for you in the data package that comes with this module, along with a shapefile defining the study area and a model to use in ArcMap for preprocessing. You will be using a Landsat 5 scene from 2010 and a Landsat 7 scene from 1999, so that you can quantify changes over that time period. Developed by the Integrated Geospatial Education and Technology Training (igett) project, with funding from the National Science Foundation (DUE ) to the National Council for Geographic Education. Opinions expressed are those of the author and are not endorsed by NSF. Available for educational use only. See for additional remote sensing exercises and other teaching materials. Created 2008; last modified January 2012.

2 Part 1: Prepare your data set. Step 1. Download and unzip the module folder in your workspace. Open the module data folders and unzip the L5_10_29_2010.tar.gz and L7_10_29_1999.tar.gz files containing the data for this exercise. You will need to unzip each of these two times, first to create a single.tar file and again yielding the individual bands. Step 2. Open ArcMap. In ArcMap, add bands 1 through 5 and 7 of the 2010 Landsat scene for path 10/ row 29: Click the ArcCatalog tab (if your ArcCatalog tab is not available, click Windows> Catalog). Navigate to the module folder, expand the folder, expand the L5_10_29_2010 folder. Select and drag band 1 into your map. Repeat for bands 2 through 5 and 7 and the study_area.shp shapefile. The bands should appear in your Table of Contents. If your map takes a long time to redraw after the addition of each layer, click the Pause Drawing button in the bottom left corner of the map window. Hint: Having trouble understanding the band file names? This guide will help. This will stop the map from redrawing. Once all your layers are listed in the table of contents, turn off all but band 1, then if necessary, click the Pause Drawing button to allow the map to redraw. Step 3. In this step you will clip the bands to include only our study area and combine them into one file called a layer stack to allow easy symbolization and band compositing. Make sure the Spatial Analyst extension is turned on: Click Customize> Extensions. Ensure that the Spatial Analyst check box is checked and close the Extensions window. Open the ArcToolbox window open the following tool: ArcToolbox> Data Management Tools> Raster> Raster Processing> Composite Bands. (You can also do a search for the Composite Bands tool.) Drag all the bands into the Input Raster field, and if necessary, re-order them. It is VERY important that the bands are in the correct order in the layer stack. Click the browse button and navigate to your project folder. Name the output layer "L5_10_29_2010_09_01_STACK.TIF." Click the Environments button and set the Processing Extent to "Same as layer study_area." This will clip the bands to our study area. Click OK to apply the extent and then OK to run the composite bands tool. Wait for the process to complete and for your composite layer to appear in the Table of Contents. You may remove the individual, unclipped bands from your map. 2

3 Step 4. Note that the new layer looks strange. We will resymbolize the layer to make it look more natural. In the Table of Contents, right click on the new layer and choose Properties. Select the Symbology tab. Under "Draw raster as an RGB composite," click the down arrow for the Red channel and select band 3, the red Landsat band. The Green channel can remain as band 2, the green Landsat band. Change the Blue channel to band 1, the blue Landsat band. Click OK. This will make the image look more natural. In ArcMap, click Windows> Image Analysis. In the Image Analysis window, click on the stack layer to select it. If necessary, expand the Display panel. Then click Stretch> Percent Clip, as shown here. If you are asked to create a histogram, click Yes. Step 5. Take a moment to explore the image by zooming and panning. When you are finished, save your map. Try it yourself! Repeat this process with the 1999 Landsat 7 scene. When you are finished, save your map in your workspace folder as "cover_change.mxd." 3

4 Part 2: Convert Digital Number to Radiance When examining change detection, it's important to normalize the images so that you can make comparisons between them. This involves two steps, taking the digital number (DN) values in each pixel and converting them to radiance and then to reflectance. To do this, you will need to collect some information about your Landsat scenes to use as inputs for the equation to convert DN to radiance and then to convert radiance to reflectance. There are tables at the end of this module that you can use to find some of the values, and others are available in the metadata for your Landsat scene. DN to Radiance In this step, you will convert the DN in your scene to radiance, the amount of energy in watts at the satellite's sensor for each cell on the ground. Here is the equation to convert DN to radiance. Lλ = ((LMAXλ - LMINλ)/(QCALMAX-QCALMIN)) * (QCAL- QCALMIN) + LMINλ L is the spectral radiance. So, LMAX and LMIN represent the highest and lowest possible values of radiance, which vary with gain state. This value is saved for each band in the MTL file saved with your Landsat scene. You can open the MTL file with Word Pad or any other word processing program. It's a good idea to note the values for all the bands in a table like the one at the end of this section. MTL File for 2010 p10r29 scene: This file comes in the Landsat package and contains information you will need to calibrate the Landsat scene. INPUTS FOR CONVERSION FROM DN TO RADIANCE & REFLECTANCE Acquisition Date 2010_09_01 Mission Landsat 5 Path 10 Row 29 Sensor TM Band source LMAXλ MTL LMINλ MTL QCALMAX MTL QCALMIN MTL Leap Year? No Day of Year Table 2a or b 244 Earth/Sun Distance (d) Table ESUNλ Table Sun Elevation MTL Solar Zenith Angle θ 90-sun elev Use a table like this one to note the values required to convert DN to radiance and reflectance (which is covered in the next section. This table shows the values for the 2010 scene. A blank one is below, and an Excel file is included with the tutorial data. 4

5 QCALMAX and QCALMIN are the calibrated maximum and minimum cell values. These values are also listed for each band in the metadata. QCAL = Is the digital number, or the cell value to be calibrated. So for that term in the equation, you will specify the target band, and the program will use the cell values in that band to calculate the radiance for that cell. Your band math equation should look something like this, substituting the variables for their values (note that some values may be negative): ((LMAXλ (LMINλ))/(QCALMAX QCALMIN)) *((BAND LAYER (QCALMIN)) + LMINλ) For example, for Band 1 in the 2010 p10/ r29 scene, the equation would be: ((193.0-(-1.52))/(255-1))*((BAND1-1.0)+(-1.52)) In addition to the MTL file, these values can also be found in Table 1 at the end of this tutorial. For other Landsat missions, such as Landsat 7, consult this PDF document produced by NASA: Chander, Markham and Hedler. "Summary of current radiometric calibration coefficients for Landsat MSS, TM, ETM+, and EO-1 ALI sensors." Remote Sensing of Environment 113 (2009) Step 1. Now you will use the Raster Calculator to convert DN to radiance for each band. To do this, you will need to add the individual clipped bands to your map. If necessary, open ArcMap and open your cover_change.mxd file. In ArcMap, open the Catalog window, navigate to your project folder, and expand the layer stack for the 2010 scene that you made in Part 1 of this tutorial. Drag Band 1 into your map. Step 2. Open the ArcToolbox window and open the following tool: ArcToolbox> Spatial Analyst Tools> Raster Calculator. Write the following expression in the Raster Calculator: ((193.0-(-1.52))/(255-1))*((BAND1-1.0)+(-1.52)) Once the expression is pasted, highlight the term "Band1." Then find the Band 1 layer in the variables box and double click it. This will replace the "Band1" in the expression with the actual name of the layer. If you used the naming convention described in Part 1, your expression will look exactly as shown here. Click the browse button, navigate to your project folder, and name the output L5_10_29_2010_09_01_B1_RAD.TIF. Click Save and then click OK to execute the calculation. Wait for it to complete and for the new Band 1 radiance layer to be added to your map. 5

6 Step 3. Repeat this process for bands 1 through 5 and 7 in both the 2010 and the 1999 scenes. REMEMBER! The values for LMAXλ and LMINλ will be different for each band, and you will need to look these up in the MTL file for each scene. Also, remember that band 6 in your layer stack is actually Landsat band 7. When you are finished, remove the original bands from your map, leaving only the radiance bands in your Table of Contents. Save your map. You will continue with the calibration process in the next lab. INPUTS FOR CONVERSION FROM DN TO RADIANCE & REFLECTANCE Acquisition Date Mission Path Row Sensor Band source LMAXλ MTL LMINλ MTL QCALMAX MTL QCALMIN MTL Leap Year? Day of Year Table 2a or b Earth/Sun Distance (d) Table 3 ESUNλ Table 1 Sun Elevation MTL Solar Zenith Angle θ 90-sun elev. 6

7 Part 3: Convert Radiance to Reflectance Radiance to Reflectance: Top-of-atmosphere reflectance (ρ λ ) is a normalized, unitless measure of the ratio of the amount of light energy reaching the earth's surface to the amount of light bouncing off the surface and returning to the top of the atmosphere, to be detected by the satellite's sensors. Reflectance: ρ λ = π * L λ * d 2 ESUN λ * cosθ s Lλ is, of course, the spectral radiance at the sensor's aperture, that is the radiance value calculated in the previous lab for each cell in each band. So the radiance bands you created in the last step will be the input for this term in the expression. The variable d is the distance from the earth to the sun in astronomical units (AU). The earth's distance from the sun varies, depending on the date. To find the earth-sun distance, first use Table 2a or 2b at the end of this tutorial to determine the Julian day or "day of year" that the scene was taken (note that Table 2a should be used for scenes taken in non-leap years and Table 2b should be used scenes taken on leap years). Then use Table 3 to find earth/sun distance on that day. For example, for a scene taken on September 19, 2008 (a leap year) we use Table 2b to determine that this was day 263 of that year. So from Table 3, we see that the earth was AU from the sun on day 263. Therefore, d = ESUNλ is the mean solar exoatmospheric irradiance. In other words, it is the mean amount of light of a particular band that makes its way to the sensor from space, without passing through the atmosphere. You could think of it as ambient light around the satellite that is picked up by the sensor. This value doesn't change over time and is constant for each band on the Landsat 5 sensor. These values can also be found in Table 1 at the end of this tutorial. For other Landsat missions, such as Landsat 7, consult this PDF document produced by NASA: Chander, Markham and Hedler. "Summary of current radiometric calibration coefficients for Landsat MSS, TM, ETM+, and EO-1 ALI sensors." Remote Sensing of Environment 113 (2009) INPUTS FOR CONVERSION FROM DN TO RADIANCE & REFLECTANCE Acquisition Date 2010_09_01 Mission Landsat 5 Path 10 Row 29 Sensor TM Band source LMAXλ MTL LMINλ MTL QCALMAX MTL QCALMIN MTL Leap Year? No Day of Year Table 2a or b 244 Earth/Sun Distance (d) Table ESUNλ Table Sun Elevation MTL Solar Zenith Angle θ 90-sun elev Use a table like this one to note the values required to convert DN to radiance (covered in the previous section) and reflectance. This table shows the values for the 2010 scene. A blank one is below, and an Excel file is included with the tutorial data. 7

8 The solar elevation and the solar zenith angle, vary with season and time of day. Images courtesy of NASA) θ s is the solar zenith angle. This is the angle between the sun and the satellite, which depends on how high the sun is above the horizon, i.e. the sun's elevation. The elevation of the sun over the horizon depends on both the time of day and the season, and it is recorded when the scene is taken. To find this value, find the sun's elevation in the MTL file for the scene, and subtract the sun's elevation from 90 o. For example, if the sun's elevation is o, the solar zenith angle is 90 o o = o. So, for Band 1 in the 2010 p10/ r29 scene, the entire process would look like this: Find the sun's elevation in the MTL metadata file. The equation for converting radiance to reflectance is... π * L λ * d 2 ρ λ = ESUN λ * cosθ s Here are all the variables for Band 1 of the2010 p10/ r29: θ s = o = o, and this will be converted to radians by multiplying it by π/180 in the expression below Day of Year (non-leap year) = 244, so d = ESUN λ for band 1 from Table 1 below is 1983 L λ is radiance band 1: L5_10_29_2010_09_01_B1_RAD.TIF To put it all together, you would enter the following expression in the Raster Calculator: ( * "L5_10_29_2010_09_01_B1_RAD.TIF" * * ) / (1983) * Cos( * / 180) 8

9 Step 1. Now you will convert each radiance band to reflectance. To begin, open your cover_change.mxd file. Make sure that radiance bands 1 through 5 and 7 are in the Table of Contents for both the 2010 and 1999 scenes. Step 2. Open ArcToolbox, and then open Spatial Analyst Tools > Map Algebra > Raster Calculator. For raster calculations using trigonometric functions, you will need to set environment variables. Click the Environments button, expand the Processing Extent section and set the Extent to Same as layer "L5_10_29_2010_09_01_B1_RAD.TIF." Expand the Raster Analysis Section, and set the cell size and mask to the "L5_10_29_2010_09_01_B1_RAD.TIF" layer. Click OK to apply the settings. Enter the expression above for band 1 of the 2010 p10/ r29 image. Click the browse button, navigate to your project folder, name the output file L5_10_29_2010_09_01_B1_REF.TIF. Click Save and then OK to execute the expression. Wait for the process to complete. The output layer will be added to your map. Step 3. Repeat this process for bands 1 through 5 and 7 in both the 2010 and the 1999 scenes to convert them all to reflectance. REMEMBER! The value for ESUN λ will be different for each band, and you will need to look these up in Table 1 for each calculation. Remember to replace the input band with the correct one. Also, remember that band 6 in your layer stack is actually Landsat band 7. Step 4. Now combine the reflectance bands for the 2010 scene into a layer stack to allow easy analysis and band compositing. Make sure the Spatial Analyst extension is turned on: Click Customize> Extensions. Ensure that the Spatial Analyst check box is checked and close the Extensions window. Open the ArcToolbox window open the following tool: ArcToolbox> Data Management Tools> Raster> Raster Processing> Composite Bands (you can also do a search for the Composite Bands tool). Drag all the bands into the Input Raster field, and if necessary, re-order them. It is VERY important that the bands are in the correct order in the layer stack. Click the browse button and navigate to your project folder. Name the output layer "L5_10_29_2010_09_01_REFSTACK.TIF." 9

10 When you are finished, remove the original bands from your map, leaving only the composite reflectance stacks in your Table of Contents. Save your map. INPUTS FOR CONVERSION FROM DN TO RADIANCE & REFLECTANCE Acquisition Date Mission Path Row Sensor Band source LMAXλ MTL LMINλ MTL QCALMAX MTL QCALMIN MTL Leap Year? Day of Year Table 2a or b Earth/Sun Distance (d) Table 3 ESUNλ Table 1 Sun Elevation MTL Solar Zenith Angle θ 90-sun elev. 10

11 Part 4: Create a Cloud Mask We have one more step in preparing the data for analysis. It isn't always possible to get Landsat imagery that is cloud-free for the time frames needed for a particular study. While 0% cloud cover is ideal, sometimes it is necessary to use images with clouds and the shadows cast by clouds. If our study is looking at land cover, however, we need to exclude the clouds and shadows from our analysis. To do this, we create a cloud mask and apply it to the scene, assigning the pixels marred by clouds and shadows a value of NoData. The same process can be used to mask scan line gaps in Landsat 7 data collected after May 2003 when the scan line corrector on the Landsat 7 satellite failed. For more information about the L7 scan line corrector failure, visit this NASA webpage: Step 1. Open a new ArcMap document, and add the 2010 reflectance stack: L5_10_29_2010_09_01_REFSTACK.TIF. Symbolize the 2010 reflectance stack using a false color band combination (4,3,2). Note the clouds and shadows of clouds in this scene. We must mask them so that we don't mistake them for land features when we analyze and classify our image. To begin, you will create a polygon shapefile and digitize polygons over the clouds and their shadows. In the Catalog window, navigate to your project folder, right click on the folder, and choose New > Shapefile. Name the new shapefile cloud_mask. Under Feature Type, choose Polygon. Click the Edit button under Spatial Reference pane. Click Import. Navigate to your project folder, and select Import. Navigate to your project folder, click on one of the two reference stacks and click Add. Click OK and OK again. The new empty shapefile will be added to your table of contents. Step 2. Now you will digitize polygons over the clouds and their shadows. Add the Editor toolbar. Right click the cloud_mask.shp layer in your table of contents. Choose Edit Features > Start Editing. The Create Features window will appear. In the Create Features window, click the cloud_mask template. Under Construction Tools on the bottom of the Create Features window, click Polygon. Zoom in, as needed, and digitize polygons over each of the clouds and their shadows. To finish a polygon, double click. You don't need to be very precise, just make sure to completely mask the clouds and the shadows. You may need to check the aerial to confirm whether they are clouds or shadows. When you are finished digitizing all the clouds and shadows, click Editor > Stop Editing. Click Yes to save your edits. Step 3. Now you will convert the cloud_mask layer to a raster. Open ArcToolbox and open the following tool: Conversion Tools > To Raster > Polygon to Raster. Under Input Features, choose the cloud_mask layer. Under Value Field, choose Id. Save the output layer to your project folder and name it cloud_mask.tif. Set the Cell Size to

12 Click the Environments button. Expand the Processing Extent item, and under Extent choose "Same as layer L5_10_29_2010_09_01_REFSTACK.TIF." Expand the Raster Analysis item and set the Cell Size and Analysis Mask to L5_10_29_2010_09_01_REFSTACK.TIF. These environment variables ensure that the output raster will have the same cell size and extent as the reference stack layer. Click OK and OK. Wait for the process to complete. The cloud_mask raster layer will be added to your table of contents. Remove the cloud_mask shapefile from your map. Note that the cells corresponding to the cloud and shadow polygons have a value of 0. By default, cells with a value of NoData are displayed as transparent. To assure yourself that this new grid has the same cell size and extent as the reflectance stack, open the Properties window for the cloud_mask raster layer. In the Symbology tab, under Display NoData as, select any color. If necessary, zoom out to see the entire grid. The NoData cells should completely cover the reflectance stack layer. Step 4. Now you will reclassify the mask. Open ArcToolbox, and open the following tool: Spatial Analyst Tools > Reclass > Reclassify. Under Input Raster, choose cloud_mask. Make sure the Reclass Field is set to Value. Under New Values in the first row, type "NoData." In the second row type "1." Name the output cloud_mask_null.tif and save it to your project folder. Click OK and wait for the process to complete. The new layer will be added to your map. Note that now the clouds and shadows have a value of NoData, so they are displayed as transparent. The remainder of the cells now have a value of 1. Step 5. In this final step, you will apply the mask to the 2010 scene. To do this, you will need to add the individual reflectance bands you created in the previous section. So add the following to your map: L5_10_29_2010_09_01_B1_REF.TIF L5_10_29_2010_09_01_B2_REF.TIF L5_10_29_2010_09_01_B3_REF.TIF L5_10_29_2010_09_01_B4_REF.TIF L5_10_29_2010_09_01_B5_REF.TIF L5_10_29_2010_09_01_B7_REF.TIF (Alternatively, you can add the individual bands from the reflectance stack.) With the NoData cells displayed in yellow and cloud and shadow cells shown in red, the mask grid completely covers the 2010 reflectance stack. Open the Raster Calculator (Spatial Analyst Tools > Map Algebra > Raster Calculator). To build your expression, under Layers and Variables, double click the "cloud_mask_null" layer. Click the * button 12

13 once. Then under Layers and Variables, double click reflectance band 1. Your expression should look like this: "cloud_mask_null" * "L5_10_29_2010_09_01_B1_REF.TIF" Name the output L5_10_29_2010_09_01_B1_REF_CF.TIF (the CF indicates that the image is now cloud-free), and save it to your project folder. Click OK and wait for the process to complete. The new layer will be added to your map. The cells corresponding to clouds and shadows now have a value of NoData. Use the Identify tool to make sure. Step 6. Repeat the raster calculation with the rest of the 2010 reflectance bands. The 1999 scene is completely cloud-free in our study area, so we do not need to apply a cloud mask to the 1999 scene. Step 7. When you have conducted the raster calculation for all of the 2010 reflectance bands, create a composite stack and name it L5_10_29_2010_09_01_REFSTACK_CF.TIF. Remove all layers from your map except L5_10_29_2010_09_01_REFSTACK_CF.TIF. Save your map as cover_change_cf.mxd. 13

14 Part 5: Use band combinations to symbolize and then compare your scenes. In this part of the tutorial, you will use ArcMaps RGB compositing capabilities to symbolize the Landsat bands in several different combinations. This allows you to highlight different land cover types of interest using bands in both the visible and infrared Landsat bands. There are many different ways to symbolize these images through the use of band combinations. A band is assigned to each of the red, green, and blue channels of the image, and the color composite of these channels shows details in the landscape pertaining to the three bands that were chosen. The band combination 3-2-1, with settings as shown in Part 1, reveals the image in natural color. This combination shows the land much as the human eye would see it from above, using visible wavelengths of light--red, green and blue each assigned to the appropriate channel. We'll start with natural color. Hint: When adding images with multiple bands to your map using the Catalog window, be sure to click once on the layer name to highlight the image file. Double clicking on the image will allow you to add the bands separately, in which case you will be unable to make band combinations. If you do double click on the image and see the individual bands listed, simply click the Up One Level button to return to the folder containing the composite layer. Settings for natural color band combination, with Landsat Band 3 (red) assigned to the red channel, Band 2 (green) assigned to the green channel, and Band 1 (blue) assigned to the blue channel. Step 1. Now you will symbolize the 2010 image to mimic natural color. Right click the L5_10_29_2010_09_01_REFSTACK_CF.TIF layer in the Table of Contents and choose Properties. In the Properties window, click the Symbology tab. In the Band column, use the pull-down menus to change the bands associated with each channel so that Red = 3, Green = 2, and Blue = 1. Make sure the 14

15 Stretch Type is set to the image to Standard Deviation with n = 2. Your settings should appear as shown above. Click OK to apply the changes and dismiss the Properties window. Notice how this changes the image. Step 2. Add the 1999 reflectance stack and repeat the symbolization process. Turn the top layer on and off by clicking its check box in the table of contents to view the changes between 1999 and Use the magnification tools to zoom in and inspect the images more closely. Check it out! What areas of the image changed the most in that time period? What might have caused the changes you observe between the two time frames? Were the changes natural or caused by humans? Another common band combination is known as false color, where the band order is This combination uses the near-infrared band (band 4) in the red channel. Since green vegetation strongly reflects near-infrared light, lush and healthy vegetation appears as a dark red instead of green; the darker the red the color, the healthier the vegetation. Now you will change the band combination for one of the image files to show the image in false color. Step 3. Right click on the 2010 image in the Table of Contents and select Properties. Choose the Symbology tab and select. Use the Band drop-down menus to set the red channel to band 4, the nearinfrared band. For the green channel, choose band 3, and for the blue channel, choose band 2. Click OK. Repeat this step for the 2010 image, then examine the images to see the result. Remember, vegetation will appear red in a false color image. Hint: Learn more about band combinations. This NASA site has a clear explanation: You may also wish to consult your remote sensing and image analysis textbook for a more in-depth discussion of band combinations. Be sure you can explain the advantages of the false color combination. Why do we use band 4, the near infrared, in the false color band combination? Try it yourself! Use some of the band combinations listed in the NASA reading above. Try your own combinations and see which features pop out as a result of your choices. Use the Spectral Sensitivity table in your textbook or the USGS Spectral Characteristics Viewer ( to try to figure out why certain features are more prominent as a result of your chosen band combination. Discuss your findings with a classmate. False color band combination with Landsat Band 4 (near-ir) assigned to the red channel, Band 3 (red) assigned to the green channel, and Band 2 (green) assigned to the blue channel. Note that vegetation appears pink or red. 15

16 Part 6: Calculating NDVI Scientists often use Landsat imagery to analyze and quantify land cover types and changes in land cover over time, and they have developed indices that compute ratios of reflectance among different bands to discern land cover in any given location. For example, the normalized difference vegetation index (NDVI) is an index based on the relative reflectance of the red (band 3) and near infrared (band 4). We will explore this index and others next. Now we will calculate the Normalized Difference Vegetation Index (NDVI) for each time frame using reflectance for bands 3 and 4 as inputs. We use band 3 (red) and band 4 (near-infrared) for this index because the chlorophyll the green, photosynthetic pigment in plant tissue reflects light of these wavelengths very strongly, allowing us to easily distinguish between plant material, bare ground and other cover types. Also, variations in the intensity of reflectance in the near-infrared band can indicate different types of vegetation, allowing us to differentiate between grass, deciduous forest and coniferous forest, for example. NDVI is calculated using the following equation with the red and near infrared bands: NDVI = (NIR - Red) / (NIR + Red) Note that if the reflectance value in the near infrared band is significantly larger than the reflectance value in the red band, the NDVI value will be high. This pattern, with NDVI values between approximately 0.4 and 1, indicates vegetation. On the other hand, if the reflectance value in the near infrared band is nearly equal to the reflectance value in the red band, the NDVI value will be near zero. If the reflectance value in the near infrared band is significantly lower than the reflectance value in the red band, the NDVI value will be negative. NDVI values less than 0.3 typically indicate non-vegetative land cover. However, it's important to remember that any such determinations must be confirmed by ground truthing. ArcMap does have an NDVI tool. However, for this exercise, we will create our own NDVI tool using Model Builder. This will allow us to reclassify the results to show vegetated and non-vegetated areas in our scenes. Note: The following instructions assume that you have had at least a basic introduction to geoprocessing with Model Builder in ArcGIS. If you have not used Model Builder before, the Executing Tools in Model Builder Tutorial in the ArcGIS Desktop 10 Help website will get you started. Step 1. Open a new ArcMap document. For the NDVI calculation, you will need to use individual bands. So add to your map the reflectance bands 3 and 4 from the 2010 scene. Step 2. In the ArcCatalog pane, navigate to your lab activity folder and right click on it. Choose New > Toolbox. Rename the toolbox "NDVI Toolbox," then right click on it. Choose New > Model. An empty model window will appear. In ArcCatalog, rename the model "NDVI Tool." 16

17 Step 3. In the model window, click Model > Model Properties. Select the Environments tab. Scroll down and check the box next to Workspace. Click the Values button. Expand the Workspace item. Set the Current Workspace to your project folder. Create a new folder called Scratch in your project folder, and set that as the Scratch Workspace. Click OK. Step 4. Drag the band 4 layer into the model. It will appear as a blue oval. Right click on the blue oval, and choose Rename. Change the name to "Band 4." Drag band 3 into the model; rename it "Band 3." In ArcToolbox navigate to Spatial Analyst Tools > Math, and drag the Minus tool into your model. You will use this to model the Band 4 Band 3 portion of the NDVI formula. Use the connect tool to connect the Band 4 oval to the Minus tool, and choose Input raster or constant value 1, as shown here. Use the connect tool to connect the Band 3 oval to the Minus tool, and choose Input raster or constant value 2. The Minus tool will turn yellow, and the output layer will turn green. Double click the yellow Minus tool to open it. Rename the output layer "B4_minus_B3.TIF," and click OK. Save your model. Step 6. Now you will use the Divide tool to model the division of the Band 4 Band 3 expression by the Band 4 + Band 3 expression. Drag the Divide tool from the Spatial Analyst Math toolset into your model. Use the connect tool to connect the B4_minus_B3.TIF oval to the Divide tool, and choose Input raster or constant value 1, as shown here. Step 5. Now you will use the Plus tool to model the Band 4 + Band 3 portion of the NDVI formula. Add the Plus tool from the Spatial Analyst Math toolset, and connect both Band 4 and Band 3 to the Plus tool. Open the Plus tool, and rename the output layer "B4_plus_B3.TIF." Click OK. Use the connect tool to connect the B4_plus_B3.TIF oval to the Divide tool, and choose Input raster or constant value 2. The Divide tool will turn yellow, and the output layer will turn green. Open the Divide tool and rename the output "L5_10_29_2010_09_01_NDVI.TIF," and click OK. Right click on the green L5_10_29_2010_09_01_NDVI.TIF oval and choose Add to Display. 17

18 Step 7. Save your model, then click Model > Run Entire Model and wait for the model to run. Each tool will turn red as it is executed, so that you can watch the process. The output will be added to your map. Save the model and close it. Step 8. Right click on the new NDVI layer in the Table of Contents and choose Properties. Click on the Symbology tab. Under Show in the left hand panel, click on Classified. Click the Classify button and enter the following break values: 0, 0.3, 0.5, 0.6 and 0.9. Choose a brown-to-green color ramp. You may change the first class of values below zero to a blue color, because this is mostly water. Click the Display tab, under Transparency, enter 60 and click OK to apply the new symbology and display settings. Turn off all layers except the new NDVI layer. Step 9. Add a basemap showing aerial imagery of the area and examine your results. Click the small down arrow next to the Add Data button and choose Add Basemap. Select the Imagery option (this step will require an active internet connection). Aerial imagery will load under your transparent NDVI layer. Step 10. Use display tools in the Image Analysis window to explore the data further. If necessary, open the Image Analysis window by clicking Windows > Image Analysis. In the Image Analysis window, click the 2010 NDVI layer in the top pane. In the Display pane, click the Swipe button. Then click at the top of the map and drag the cursor down. This will reveal the aerial image beneath the 2010 NDVI layer, so that you can compare them more easily. Note the values of NDVI in various cover types. Step 11. Now you will run the NDVI model for the 1999 image. Add bands 3 and 4 from the 1999 reflectance stack to your map. In ArcToolbox, navigate to the NDVI toolbox you created earlier. Expand the toolbox and right click on the NDVI Tool model and choose Edit (NOT OPEN). The model will open. Double click on the blue Band 4 oval, and in the pull-down, select the 1999 band 4 reflectance layer. Click OK. Double click the blue Band 3 oval, and in the pull-down, select the 1999 band 3 reflectance layer. Click OK. Double click the Divide tool and rename the output "L5_10_29_1999_09_03_NDVI.TIF." Click OK. Save your model and run it. The 1999 NDVI layer will be added to your map. 18

19 Step 12. Symbolize the 1999 NDVI layer using the same classification and color scheme (the easiest way to do this is to click the Import button on the Symbology tab and select the 2010 NDVI as the template for symbolization). Apply a 60% Transparency to the 1999 NDVI layer. Step 13. Zoom into areas where you see a lot of human activity in the image, and use the Swipe tool to switch between the two images. What do you notice? Try the Flicker button to flash between the two layers quickly. You can also use the Swipe and Flicker tools to view the aerial, taken near the time the 2010 Landsat scene was collected. When you are finished, save your map. Check it out! Read up on NDVI. This site offers a clear and concise overview of commonly used vegetation indices: 19

20 Part 7: Supervised Classification Now you will use the supervised classification toolset to classify land cover types in both scenes. In this classification method, you first create a training data set identifying specific cover types of interest. ArcMap can then analyze the spectral characteristics of the cover types in the training data set and identify other places in the scene where those cover types occur. Step 1. First, you will open the Image Classification Toolbar. Click Customize > Toolbars > Image Classification to add the Image Classification Toolbar. Under Layer in the Image Classification toolbar, select L5_10_29_2010_09_01_REFSTACK_CF.TIF. Step 2. Since we are interested in distinguishing blueberry barrens, in particular, we will zoom to a location where there is a large barren in our scene. On the Standard toolbar, click the scale pull-down and choose 1:100,000. On the Tools menu (that's the one with your zoom tools) click the Go To XY button:. Use the Units pull-down to select meters. In the X field type: In the Y field type: Press Enter to zoom to a large complex of commercial blueberry barrens. A Inspect the image. Note that some of the barrens appear red (A) in the false color image, indicating that they are covered with vegetation. Others appear green (B) indicating that they are bare. Blueberry growers harvest barrens on alternate years, burning or mowing the barrens in fallow years. The vegetated barrens shown in the image (A) produced berries during the 2010 season, while the green barrens (B) were burned or mowed and will produce berries the following year. It's important to know about the harvest cycle of blueberries, because barrens appear different in the satellite image depending on their stage in the cycle. Step 3. On the Image Classification toolbar, click the Training Sample Manager button. The Training Sample Manager window will open. Click the Draw Training Sample with 20

21 Polygon button on the Image Classification Toolbar. Now click on the image to outline one of the productive red barrens. Zoom in, if necessary, to be sure you don't include other cover types in your sample. Double click to finish the polygon. Note that a row is populated in the Training Sample Manager window corresponding to the polygon you've just drawn. Draw another polygon around another red barren, and continue to do so until you've drawn polygons indicating four or five red barrens. You can change the color of a polygon by clicking the color patch in the Training Sample Manager window. If you make a mistake while digitizing, you can delete a polygon by clicking its number in the Training Sample Manager window and clicking the X button. Step 4. Since these polygons all indicate the same land cover type, we will combine them into the same training sample layer. In the Training Sample manager window under ID, click the number 1 to highlight that row. Hold down your Shift key and click the last row in the list. This will select all the polygons you have drawn note that each selected polygon is hatched. Click the Merge button on the Training Sample Manager window toolbar. The polygons are merged into a single layer and assigned the same color. Under class name, change the name of the newly merged layer to BB Producing. Step 5. Repeat this process with several green barrens, merge them, and name the new layer BB Fallow. These barrens were mowed or burned and are not producing in this scene. ArcMap can use training samples to conduct a process called supervised classification. The software conducts a statistical analysis of all the bands in the stack and then checks each of the other cells in the scene to determine if it is similar to the training samples. These layers will provide a good basis for identifying blueberry barrens throughout our study area. However, before we do our supervised classification, we should create training samples from locations that are distributed throughout our scene, since lighting conditions and agricultural practices may vary from place to place. Also, we will need to create training samples for other cover types. Step 6. For the following steps, you may need to refer to an aerial photo to be sure of your classifications. So in this step you will add a basemap. Click the drop-down button next to the Add Data button. Select Add Basemap. Choose the Bing Hybrid basemap and click Add. The basemap will be added to your table of contents. To see the basemap, turn off the overlaying false color image layer. Alternatively, you can download the NAIP aerial of the year closes to 2010 for Washington County, Maine. Step 7. We need to identify the following cover types for our study. The coordinates of examples of these cover types are included in the table below. Zoom to a scale of 1:10,000 and then use the Go To XY tool to zoom to the examples. Before you begin, check out the hints below. Use the table as a guide to assemble eight to twelve polygons for each cover type distributed throughout the scene, and merge the polygons for each cover type into a single layer. 21

22 Cover Type Coordinates of Examples (1:10,000) Layer Name Value X Y Producing Blueberry Barren BB Producing 1 Fallow Blueberry Barren BB Fallow 2 Bare Bare 3 Fresh Water Water 4 Forest Forest 5 Light Vegetation/ Recently Logged Light Veg 6 Wetland Wetland 7 Coordinates indicate a point at the center of the feature Hints on Digitizing Training Samples: There is no need to outline a sample feature perfectly when you are drawing sample polygons. It is more important to include a number of adjacent cells of the same cover type. For this analysis, ignore salt water. It's not relevant for our study. If you are unsure of the classification, don't include it in the training sample. Forest cover may be either deciduous or coniferous, which will appear light red or dark red This map shows an area with several training sample polygons. Note that the Count column includes the number of cells in each sample type. in the false color image. They can both be classified as forest for this study since neither is likely to be misclassified as blueberry barren. Check the aerial basemap to confirm forest cover. Try to include 500 or more cells in each training sample class (though it might be difficult to find areas that can be classified as bare). You can reorder the samples in the Training Sample Manager window by selecting a row and clicking the arrows on the toolbar. This allows you to group similar cover types. You may find it fastest to create many samples of a single cover type then merge them together. If you merge some polygons in error, you can unmerge them with the Split button next to the Merge button. If you have an aerial photo taken at the same time (or nearly so) as your Landsat image, you can use it to confirm your land cover classifications. Step 8. When you are finished creating training samples and have merged, renamed and colored the samples, use the arrow buttons in the Training Sample Manager window to reorder the classes as they appear in the table and in the example shown above. Click the Reset Class Values button: sure the class values match those in the table above. Click the save button on the Training Sample Manager toolbar to save the sample data set as a shapefile. Name the shapefile training_samples_2010.shp and save it in your project folder.. Make 22

23 Step 9. Now you will use the Interactive Supervised Classification tool to classify the 2010 scene. In this step, ArcMap will use a maximum likelihood classification method to analyze the spectral signatures of the samples and classify the entire scene by comparing the spectral characteristics of each cell with the spectral statistics for each class. It assigns the cell to the class with the highest probability of a match. If you would like to learn more, see How Maximum Likelihood Classification works on the ArcGIS Desktop 10 Resource Center website ( On the Image Classification toolbar, click Classification > Interactive Supervised Classification. A new layer will appear in your map. To make this layer permanent (it is currently saved in your temp folder), right click on it in the table of contents and choose Data > Export Data. Save the output in your project folder and name it MLClass_2010.TIF. When asked, add the exported layer to the map. You may remove the temporary classification layer. Note that ArcMap assigned numbers to the classes. Take a moment to rename each class in the table of contents so that it shows both the class number and class name: 1- BB Producing, 2- BB Fallow, and so on. You may want to change the colors as well. Step 10. Examine the results. Use transparency or the swipe tool to compare the classification layer with the satellite image and the aerial. Did ArcMap correctly classify the cover types? Why or why not? It is common to do image classification many times to achieve satisfactory results. If you were to do this classification again, what would you do differently and why? Note: Roads are sometimes misclassified as blueberry barrens. Why might that be? Our postprocessing should clean up the classification and address this problem. There are many small, isolated patches of land cover in the classified image leading to a "noisy," speckled appearance. We will clean up the image with post-processing later. Step 11. Clean up and save your map. You may wish to collect layers you created in classifying the 2010 scene in a group layer called Turn off the Bing basemap and all of the 2010 layers and save your map. Now you will repeat the supervised classification process with the 1999 reflectance stack. First, you will need an aerial taken at nearly the same time as the Landsat scene. 23

24 Step 12. You will now add a web map service (WMS) to your map showing aerials of the region taken in the late 1990s. Open the Catalog window and navigate to your project folder. Drag the maine_orthos_ lyr file into your map. A grayscale image will appear in your map. Add the 1999 reflectance stack to your map and symbolize with a false color display (4,3,2). Change the target layer in the Image Classification toolbar to the 1999 reflectance stack, and if necessary, clear the old training samples from the Training Sample Manager window. Refer to the aerial and the 1999 Landsat image to build a training data set. Remember, the aerial is a composite of images taken over three years. So there may be land cover changes visible in the 1999 Landsat image that are not visible in the aerial. For example, this was a period of vigorous logging activity in the region, so several areas in the Landsat image had been logged after the aerial was taken. Note: Classifying the 1999 image will be much easier if you have had some training and experience in aerial interpretation. Ask your instructor if you should conduct create the training samples yourself or if they will be provided. Things to remember as you create a 1999 training data set: Don't forget to change the target layer on the Image Classification toolbar to the 1999 reflectance stack. Producing blueberry barrens will appear red in false color. Fallow barrens will appear green. Without a color aerial to guide you, it may be more difficult to clearly identify good training samples. Don't be tempted to use samples if you're not certain of the cover type. The aerial imagery is scale-dependent, so you may need to zoom in very close to discern the cover types in the image. Be persistent and try to include at least 500 cells for each cover type. Remember to reorder and renumber the classes in the Training Sample Manager window to match the numbers we used in classifying the 2010 scene. Transparency settings or the Swipe tool may be useful in creating the training data set. When you are finished building a training data set, save it and run an Interactive Supervised Classification. If you are satisfied with the results, make the classification permanent by exporting it to your project folder. Name it MLClass_1999.TIF. Examine the results and compare them with the classified 2010 scene. What differences do you see? Note that the quality of the 1999 scene was somewhat lower than that of the 2010 scene. Is this reflected in the resulting classifications? If so, how? Save your map. In the next section, you will do some post-processing of the classifications to eliminate the noise. 24

25 Part 8: Post-Processing Image Classification Now we will begin post-processing of the classified image. Post-processing will remove small isolated patches from our classified image. Many of these smaller patches are erroneously classified or are too small to concern us in our analysis. This will also smooth the boundaries between patches. Postprocessing involves a series of geoprocessing tools in the Spatial Analyst tools in ArcToolbox. This process is most easily done using Model Builder. To learn more about post-processing classified imagery, visit the Processing Classified Output page on the ArcGIS 10 Desktop Help website. Step 1. First, you will create a new toolbox to hold your new model. If necessary, open the ArcMap document you saved at the end of Part 6. Open the catalog window, navigate to your project folder, right click on your project folder, and choose New > Toolbox. Name the new toolbox Postprocessing.tbx and click Enter to apply the name. Right click on the Post-processing toolbox you have just made and choose New > Model. A new model window will appear. In the model window, click Model > Model Properties. Select the Environments tab. Scroll down and check the box next to Workspace. Click the Values button. Expand the Workspace item. Set the Current Workspace to your project folder. Set the Scratch folder in your project folder as the Scratch Workspace. Click OK. Drag and drop the Interactive Supervised Classification layer for 2010 (MLClass_2010.TIF) from your table of contents into your model. Step 2. Next you will add the Majority Filter tool to your model. This tool will change the classification of small, isolated patches in the classified layer to that of the majority of neighboring cells. For example, if a single cell classified as bare is surrounded by six forest cells and two water cells, that cell will be classified as forest in the output. Open ArcToolbox and expand Spatial Analyst Tools. Expand Generalization. Most of the tools you will need for this model are in the Generalization toolset. Drag and drop the Majority Filter tool into your model. When the Majority Filter tool appears in your model, use the Connect tool to connect the MLClass_2010.TIF oval to the Majority Filter tool and select Input Raster. The layers and tools will be filled with color. 25

26 In your model, double click on the Majority Filter tool to open it. Name the Output Raster maj_filter.tif and make sure it is saved to your Scratch folder. Leave the defaults for all other fields and click OK. Save your model by clicking Model > Save, and run it by clicking the run button. When the model is finished running, dismiss the results window, if necessary. Right click on the green maj_filter.tif oval, and choose Add to Display to add the maj_filter.tif layer to your map. Examine the new layer and compare it with the MLClass_2010 layer. Note that some of the speckling in the image is gone. Save your map. Step 3. We'll use the Boundary Clean tool next, to smooth the edges between patches of different classes. Drag and drop the Boundary Clean tool from the Generalization tool set into your model. Use the Connect tool to connect the maj_filter.tif oval to the Boundary Clean tool and select Input Raster. In your model, double click the Boundary Clean tool to open it. Name the output raster bndry_clean.tif and make sure it is saved in your scratch folder. Under Sorting Technique, choose Ascend. Uncheck the optional Run expansion and shrinking twice. Leave the Output Raster blank. Click OK. Save your model and run it. Add bndry_clean.tif to the display. Examine the results. Note that the boundaries between patches are smoother, but there are still many isolated, small patches. We will take care of them in the next step. Step 4. We will use three tools to eliminate all patches of with 50 or fewer contiguous cells (a little over 10 acres; the vast majority of commercial blueberry barrens will be larger than this). The Region Group tool will count the number of cells in all patches. Then the Set Null tool will select patches with fewer than 50 cells and give them a null value. Then the Nibble tool will use the Set Null output to identify the small patches and will assign them values of their nearest neighbors. So, for a patch consisting of 38 cells classified as bare and surrounded by a large forest area, the process will reclassify the 38 cells of the small patch as forest. Hint: If you close Model Builder in the middle of the tutorial, or if Model Builder crashes, do not double click the model to reopen it. Instead, right click on the model and choose Edit. 26

27 Drag and drop the Region Group and Nibble tools from the Generalization tool set into your model. In ArcToolbox, navigate to Spatial Analyst Tools > Conditional, and drag the Set Null tool into your model. Use the Connect tool to connect the bndry_clean.tif oval to the Region Group tool and select Input Raster. Double click the Region Group tool to open it. Save the output to your scratch folder and name it reg_grp.tif. Uncheck the box next to Add link field to output. Keep the default values for all other fields and click OK. Save your model and then right click the Region Group tool and choose Run. Wait for it to execute, and if necessary, dismiss the results window. Use the Connect tool Conditional Raster. to connect the reg_grp.tif oval to the Set Null tool and select Input Double click the Set Null tool to open it. Use the SQL button to open the expression builder and build the following expression: "Count" < 50. This indicates that region groups with fewer than 50 cells will be set to null in the output. Save the output to your scratch folder and name it set_null_50.tif. Click OK. Double click the Nibble tool to open it. Under Input raster, choose the bndry_clean.tif layer. Under Input Raster Mask, choose set_null_50.tif. Name the Output raster class_pp.tif. Save your model and run it. When the process completes, dismiss the results window, if necessary, and add the class_pp.tif to the display. Examine the results. 27

28 Step 5. For the last step in the model, we will conduct a reclass. Recall that we are specifically interested in identifying blueberry barrens in our study area. We will use reclassification to create a binary layer in which cells classified as blueberry barrens, either producing or fallow, have a value of 1. All other classes will have a value of zero. In ArcToolbox, navigate to Spatial Analyst Tools > Reclass, and drag the Reclassify tool into your model. Use the Connect tool to connect the class_pp.tif oval to the Reclassify tool and select Input Raster. Double click the Reclassify tool to open it. Under Reclass field, choose Value. Click the Unique button if it is not grayed out, so that individual cell values appear under Old Values in the Reclassification table. Recall that class 1 indicates BB Producing, and class 2 indicates BB Fallow. Under New Values for these classes, enter 1. For all other classes, enter 0. Name the output bb_rcls_2010.tif, and save it in your project folder, not your scratch folder (this is the final output of the model). Click OK. Save your model, run it, and add the final output to your display. Examine the results. Note that the blueberry barrens all have a value of 1, while all other classes have a value of 0. Save your map. Step 6. Now, you will run the model again using the 1999 classification layer, MLClass_1999.TIF. First, we must delete the intermediate files created when we ran the model to post-process the 2010 classification layer. In the Model Builder window, click Model > Delete Intermediate Data. The bb_rcls_2010.tif layer should remain in your table of contents, while the other layers produced by the model are removed. In your model, double click the Majority Filter tool to open it. Under Input Raster, select the MLClass_1999.TIF layer. Click OK. In your model, double click the Reclassify tool to open it. Change the name of the Output Raster to bb_rcls_1999.tif. Save the model and run it. It may take a few minutes to execute all the processes in the model. When it is finished, dismiss the results window, if necessary, and add the bb_rcls_1999.tif layer to your model. Compare it to the bb_rcls_2010. Do you see any areas of change between the two time frames? Are there more or fewer blueberry barrens in 2010 than in 1999? Are barrens larger or smaller in the newer image? Save your map. 28

29 Part 9: Change Detection Now you will use the ArcGIS Minus tool to identify changes in blueberry cultivation between 1999 and Using simple subtraction - later image minus initial image - ArcGIS will compare the classified, post-processed, and reclassified scenes to determine whether the amount of blueberry barren land has increased, decreased or remained the same. Step 1. If necessary, open a blank ArcMap document. Add the bb_rcls_2010.tif and bb_rcls_1999.tif layers you created in the previous section. Also, add the Bing Hybrid basemap. Step 2. Open ArcToolbox and open the following tool: Spatial Analyst Tools > Math > Minus. Minus is a very simple tool. It simply subtracts the values of overlapping cells in a pair of grids. Under Input raster of constant value 1, choose bb_rcls_2010 (the later image). Under Input raster of constant value 2, choose bb_rcls_1999 (the initial image). Click OK and wait for the process to run. Step 3. When the new layer is added to your map, take a moment to examine it. Note that there are three pixel values: 1, 0 and -1. Recall that land classified as blueberry barren in each layer had a value of 1, while anything else had a value of 0. So, the possible combinations are Difference Value Difference Type Description 1-1 = 0 No Change Blueberry Barren in 1999 & = 1 Increase New Blueberry Barren 0-1 = -1 Decrease Blueberry Barren is Gone 0-0 = 0 No Change Not Blueberry Barren, No Change Step 4. Now you will add a field to the attribute table and use it to calculate the area of the various classes. Open the attribute table for the bb_difference layer. In the Table Options pull-down menu, select Add Field. In the Add Field window, name the new field Area_Acres. Under Type, select Long Integer, and for Precision, type 7. Click OK. When the new field appears in the attribute table, right click on the field header and choose Field Calculator. In the Field Calculator window, build the following expression: ([COUNT] * 900) / Each cell in the grid is 30m X 30m = 900 m 2, so we multiply the number of cells, the count, by 900 to get the total area in square meters. Dividing by 4047 converts the area to acres, a common unit used in agricultural statistics. 29

30 Step 5. Consider the results. In the period between 1999 and 2010, according to your analysis, was there a net gain or loss of blueberry barren land? When you examine the difference layer and the Landsat images, does this result make sense? Why or why not? Step 6. Open the Reclassify tool. Under Input Raster select bb_rcls_2010. Under New Values, type "NoData" in the first row, and "2010" in the second row, as shown here. Name the output bb_2010.tif and save it in your project folder. Click OK and wait for the process to run. Repeat the reclass with the bb_rcls_1999 layer. For New Values enter 1999 in the second row as the new value for 1. Step 7. Open the following tool: Conversion Tools > From Raster > Raster to Polygon. Under Input Raster, select bb_2010. Make sure the Field is set to Value. Name the output bb_2010.shp, and save it to your project folder. Check the box for the option to Simplify polygons. Click OK and wait for the process to run. A polygon layer will appear in your map with polygons indicating the blueberry barren land cover class. Open the attribute table for the new polygon layer, and add a short integer field named Area_Acres with a precision of 5. When the new field appears, right click the field header and choose Calculate Geometry. Click yes, and the Calculate Geometry window will open. Under Property, choose Area. Under Units, choose Acres. Click OK. ArcMap will calculate the acreage for each of the polygons. Repeat this process for the 1999 layer. Step 8. Explore the results. You can right click on the header of the Area_Acres fields and select Statistics to learn more about the data. What is the mean size of the barrens in each of the different time frames? Did the mean size go up or down, or did it stay about the same? What does this indicate about trends in the blueberry industry? Do you see evidence of such change when you examine the original Landsat image? Step 9. Create a tabloid size layout showing changes you have identified in blueberry cultivation. You may choose to use the difference layer in a single data frame, or you may choose to have two data frames, one showing 1999 data and another showing Include a table derived from the attribute table of your difference layer and text discussing your findings regarding barren size. Also, include a brief statement about the potential sources of error in your analysis and recommended next steps. Save your layout, and export a PDF. Optional follow-up activities: 1) If you live in Maine, consider a ground truthing expedition with GPS and spectrometer. 2) Analyze an earlier or more recent Landsat image of the same area and look for additional changes. 30

31 Table 1 from Chander, Markham and Hedler. "Summary of current radiometric calibration coefficients for Landsat MSS, TM, ETM+, and EO-1 ALI sensors." Remote Sensing of Environment 113 (2009) p. 896, Table 3 Table 2a: JULIAN DAY TABLE, NON-LEAP YEAR (for leap year, see 2b below) DATE JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC From 31