Lesson 9: Multitemporal Analysis Lesson Description Multitemporal change analyses require the identification of features and measurement of their change through time. In this lesson, we will examine vegetation change over time in Bale Mountain National Park using a multitemporal change detection analysis. You will learn how to calculate and quantify the difference between two images in two scenarios presented in this lesson. First, you will examine two classified forest cover layers from the dates 1987 and 2015 to measure forest cover change. Second, you will calculate the normalized difference vegetation index (NDVI) for two images and measure the differences in vegetation density and health. Objectives: The student will: 1) Calculate change between two classified images 2) Perform change analysis on two images using NDVI Keywords: Bale Mountains National Park; NDVI; Raster Calculator Resources Required: ArcMap Data Used: LC81680552016024LGN00: Landsat 8 imagery near Awassa, Ethiopia Background: Given the extended history of satellite remote sensing (with the Landsat mission dating back to 1972 and aerial imagery as far back as the early 20 th century), multitemporal change analysis can be one of the most powerful applications of remote sensing. Some of the most common applications include quantification of urban sprawl, deforestation, reservoir changes, land use change, effects from a natural disaster, and more. These changes are often measured on a pixel by pixel basis where changes in values can be measured to quantify change through time and the implementation of these techniques can vary from relatively simple (basic subtraction) to more complex (e.g. Independent Component Analysis (ICA), radar change detection). The use of multitemporal change analyses allows for the possibility to solve complex problems related to Earth monitoring at a multitude of scales. 1
Lesson: Step 1. Postclassification Change Detection We will begin by examining two layers that have both been classified to represent forest cover, one from 1987 and the other from 2015. 1.1 Copy the data folder into your local directory, then drag the files BaleMountainNP.shp, ForestCover_Bale_1987.tif and ForestCover_Bale_2015.tif into the ArcMap Table of Contents window. 1.2 The values in these rasters represent forested (value 1 or 10) or not forested (value 0). We can begin by adjusting the symbology for these layers. For each raster layer, open the Properties and on the Symbology tab, select Unique Values and set the 0 value (not forested) to white, and the 1 or 10 value (forested) to green. Set the Bale Mountain National Park boundary to a hollow fill and outline of your choice (Figure 1). Examine both the 1987 and 2015 forest cover layers. Has forest cover appeared to increase or decrease during this time? Figure 1. Bale Mountain National Park 2015 forest cover. 1.3 We will quantify the change between these two layers using a simple subtraction operation performed within the raster calculator. 1.4 Find the Raster Calculator (Spatial Analyst) tool using the Search window or use ArcToolbox and navigate to: Spatial Analyst Tools > Map Algebra > Raster Calculator 1.5 In the raster calculator, we want to subtract the 2015 layer from the 1987 layer. Double-click the 1987 layer, ForestCover_Bale_1987.tif, click or type the subtract symbol (-), then double click the 2015 layer, ForestCover_Bale_2015.tif. Your expression should appear as follows: "ForestCover_Bale_1987.tif" - "ForestCover_Bale_2015.tif" This operation will run on each individual pixel within the raster datasets. Save your output as ForestCover_Bale_Diff.tif and click OK. 1.6 Open the Properties of the new layer, ForestCover_Bale_Diff.tif, on the Symbology tab, click Unique Values (click Yes if a notification appears). Examine the values, -10, -9, 0, 1. Think about the subtraction expression performed above and what each of these resultant values represents; Answer Question 1. 2
1.7 The following table shows the description of each value: Value Expression Description -10 0-10 Forest Increase -9 1-10 No change forested 0 0-0 No change not forested 1 1-0 Forest decrease We can now visualize where the most change has occurred and also quantify how much area has been affected in each of the categories. Open the Properties of the difference layer and go to the Symbology tab, select Unique Values. Set the color of each value to represent each respective description. I ve selected dark green to represent forest increase, light green for no change forested, tan for no change forested, and red for forest decrease (Figure 2). Also, note the Count column values, which store the number of pixels each value appears. 1.8 The count values should appear as follows: Value Count -10 43178-9 909736 0 1484999 1 123953 Switch to the Source tab, we can see the cell size for this raster is 30 by 30 meters. Figure 2. Forest cover change between 1987 and 2015. 1.9 To calculate area we simply take the number of cells multiplied by the cell size (30x30 m 2 ). The area of forest increase (-10) is thus: 43178 * 30 2 = 38,860,200 m 2 = 38.8602 km 2 1.10 Calculate the area for the remaining values to Answer Questions 2 and 3. 3
Step 2. NDVI Differencing We will now use Landsat imagery from the same dates, 1987 and 2015, to calculate the Normalized Difference Vegetation Index (NDVI) for each period in time, then measure the difference between the NDVI layers to assess changes in vegetation density. 2.1 Add the layers LT5_Bale_1987.tif and L08_Bale_2015.tif. These are top-of-atmosphere reflectance images of the park collected by Landsat 5 (1987) and Landsat 8 (2015). 2.2 We will begin by calculating NDVI for each layer. Recall from Lesson 6: Spectral Indices, NDVI is calculated as follows: NDVI = NIR - Red NIR + Red Also recall that Landsat 8 has the additional Coastal Blue band (band 1) not present on Landsat 5, therefore on Landsat 8, NIR is band 5 and Red is band 4; on Landsat 5, NIR is band 4 and Red is band 3. 2.3 In the Catalog window, click the + symbol to the left of each Landsat file to view the individual bands. For the Landsat 8 image, add bands 5 and 4, and for Landsat 5 image, add bands 4 and 3 (Figure 3). 2.4 We will calculate NDVI using the Raster Calculator. Find the Raster Calculator (Spatial Analyst) tool using the Search or use ArcToolbox and navigate to: window Spatial Analyst Tools > Map Algebra > Raster Calculator 2.5 First, calculate 2015 Landsat 8 NDVI, as follows: Figure 3. Landsat 8 and 5 bands in Catalog window ("L08_Bale_2015.tif - Band_5" - "L08_Bale_2015.tif - Band_4") / ("L08_Bale_2015.tif - Band_5" + "L08_Bale_2015.tif - Band_4") Save the output as NDVI_Bale_2015.tif. 4
2.6 Next, calculate 1987 Landsat 5 NDVI, as follows: ("LT5_Bale_1987.tif - Band_4" - "LT5_Bale_1987.tif - Band_3") / ("LT5_Bale_1987.tif - Band_4" + "LT5_Bale_1987.tif - Band_3") Save the output as NDVI_Bale_1987.tif. 2.7 Now that we have the NDVI layers, set a new color ramp in the Symbology tabs to better visualize the layers. Color ramps with a neutral zero color are best for visualizing these type of data (e.g. ). 2.8 We will use the Raster Calculator again to calculate the change between the NDVI layers. Open the Raster Calculator. Subtract the NDVI_Bale_2015.tif from NDVI_Bale_1987.tif as we did in step 1.6. Save the new layer as NDVI_Bale_Diff.tif. 2.9 In the new layer, higher values represent locations where vegetation density has decreased, while low values are locations where vegetation has increased. Set a color ramp with a neutral zero color to better visualize these changes (Figure 4). 2.10 Examine this layer and compare it with the forest cover difference map we created in Step 1; Answer Question 4. Figure 4. NDVI Difference layer between 1987 and 2015. Any use of trade, products, or firm names is for descriptive purposes only and does not imply endorsement by Colorado State University or any other collaborating individuals or agency. This tutorial was created for educational purposes and the data presented in these lessons may be incomplete or inaccurate. 5
Exercise Questions 1. What do each of the values (-10, -9, 0, 1) in the forest cover difference layer represent (e.g., forest increase, forest decrease, no change in forested, no change in nonforested)? -10-9 0 1 2. How much area falls within each forest cover class (m 2 or km 2 )? -10-9 0 1 3. What is the net decrease of forest cover for Bale Mountain National Park between 1987 and 2015 (in km 2 )? 4. How well do these data align within forested areas? Are locations where we found forest decreased represented in the difference NDVI image? 5. Aside from assessing vegetation change, what other applications could a multitemporal analysis be useful for? Are these applications relevant to your research or projects? 6