Using ArcGIS Pro for Time Series Analysis of Landsat Data

Project

Run a time series analysis to analyze spectral recovery following the 2000 Jasper Fire in South Dakota.

Data

For this project data was obtained from June 1- August 31 each year from 1995 to 2022. Landsat 5, 7 and 8 sensors were all included and data was selected from Earth Explorer. Data was converted to the NBR spectral index, projected to UTM Zone 13, and clipped 3 meters of the burn boundary using ESPA. 

After downloading, peak of greenness (POG) was estimated using the MTBS NDVI online tool. Python was used to narrow files to those which fell within 5 days in either direction of POG. After the datasets were selected, ArcGIS Pro was used to decode the QA pixel for each dataset and select the clearest images that fell within range of the POG. Each dataset was rated subjectively on a scale of 1-3 with 1 being unsuitable for analysis, 2 being suitable but still very cloudy and 3 being mostly clear. If multiple datasets were suitable for analysis and fell within the 5 day range, the one closest to the actual data was selected. If POG lasted for a longer period of time and multiple clear datasets fell within POG the central dataset was selected. For 2013 a dataset within 5 days of POG was not possible due to limited data sets that covered the burn boundary and fell within the peak of greenness dates, so the dataset that was closest to POG and acceptable for analysis was selected.

Creating and Populating a Mosaic Dataset

Once a suitable POG dataset was selected for each year of analysis, datasets were compiled into a mosaic dataset in ArcGIS Pro using the Create Mosaic Dataset tool followed by the Add Rasters to Mosaic Dataset tool. 

Examples of the Create Mosaic Dataset and Add Rasters to Mosaic Dataset tools

The following steps outline how to populate and mask a mosaic dataset.

After creating a Mosaic Dataset add the desired raster types. The tool has to be run separately for each of the Landsat types. Set the raster type to the appropriate Landsat sensor. The default raster type is "Raster", but if you leave this setting you will pull the .tif files instead of the .xml/ MTL files. If you want to add a cloud mask to your files later on it is necessary to pull the data from the .xml/MTL file. If you pull the .tif files the group name isn't populated so you will be unable to match the appropriate QA band to the appropriate raster.

 For the purposes of my analysis I wanted to add the Spectral Indices as my processing template, which would pull my NBR file. I changed input data to "Folder" and selected the folder with my chosen dataset. I then selected to calculate mosaic statistics and clicked run to add the files to my dataset. Repeat this for all Landsat sensors represented in your datasets.

After all the spectral index files are added to the mosaic dataset, add the QA files. This process is similar to adding the indices but instead of selecting "spectral indices" as the processing template select "QA". After selecting the processing template, click the edit button beside the template. This pulls up the raster functions pane.

 In the raster function page, search for the "remap" function. Drag the remap function to the function editor pane. Remove the identity function and click and drag your mouse from the QA raster to the remap function. An arrow should appear, linking your raster to the function. 

Double click the remap function in the function editor to bring up the remap properties pane. Set the variables to match with the desired pixel values according to the landsat documentation for the specific sensor you selected, Landsat 8 has different pixel values than Landsat 5 and 7. Set the output value to 1. For my map I wanted all clear pixels to remain visible and all others set to "No Data". Make sure the  "Change Unmatched Values to NoData" option is selected.

The raster functions pan can be seen on the left, the function editor appears on the bottom of the screen and the remap properties pane appears above it. The pixel values in this example align with the "clear" pixel values obtained from Landsat 8 documentation on the USGS website. Landsat 5 and 7 use different values to represent clear pixels.

Hit the save button at the top of the function editor. This should bring you back to the "Add Raster to Mosaic Dataset Pane". Deselect the "Estimate Mosaic Dataset Statistics" and click run to add the QA raster. Repeat for all sensors.

This is important for masking: in your attribute table for your mosaic, make sure the GroupName column for your rasters is present. Your raster group name should match up to the corresponding QA pixel group name. The Tag column should match the processing template chosen when uploading your rasters.

Attribute table for the mosaic dataset. Can be accessed by right clicking the mosaic dataset and selecting "Open Table"

To create a mask template access the catalog pane and right click your mosaic dataset. Select "Manage Processing Templates". Click the hamburger menu button and select “Create New Template”. Under raster functions search for clip and add clip raster function to the raster function editor. Double clip the clip function to open the clip properties pan. Click the variables tab and set the appropriate GroupName variables as the "Raster" and "Clip" features, these variables will align with your "Tag" column in the raster attribute table. In my case the Raster to be clipped was the SpectralIndices and the clipping raster was the QA raster. Click Okay. 

Save the raster in the function editor. This will pull up the "Save" pane. Add desired name and description information. Set the "type" field to item group. Set the group field as "GroupName" and the tag field as "Tag". Uncheck the match variables option.

Example of the Clip Properties and Save panes

After saving the Clip Function, return to the Manage Processing Template pane and click to star to set it as your default template. Your mosaic dataset should be masked, but if for some reason the processing template is not applying you can make sure it is activated through the ribbon. With your mosaic dataset selected in your contents pane, select the "Data" tab at the top of  the ribbon. Under the "Processing" group select "Processing Templates" and make sure your masking template is selected.

Create a Multidimensional Dataset

To add multidimensional data to a mosaic dataset utilize the Build Multidimensional Information Tool. Set the variable field to the product name and set the variable name to the appropriate layer, in my case I set it to Spectral Index. Set the Dimension field to the Acquisition date.

After running this tool you should be able to adjust the slider along the top of the map view to view your masked data over time.

Finally, use the Copy Raster Tool to convert the multidimension dataset to a CRF file. Select the options to "Process as Multidimensional" and "Build Multidimensional Transpose". You now have a .CFR file suitable for time series analysis. 

The Copy Raster Tool converts the multidimensional dataset to the .CRF format to optimize for analysis.