Log Difference Tool

 This tutorial is one part of a larger workflow using On Demand RTC products from the Alaska Satellite Facility. 

Radiometric Terrain Corrected (RTC) products are generally visualized in grayscale, and while backscatter is sensitive to a variety of changes that can occur on the landscape, it may not be easy to identify differences visually.

Calculating the log difference between two images is a simple way to highlight areas where there are significant changes in backscatter over time.

The Calculate Log Difference Tool in the ASF_Tools ArcGIS Toolbox calculates the log of the ratio of the pixel values from two images collected over the same area. While it is optimized for two RTC images in amplitude scale collected over the same area on different dates, any two single-band rasters could be used as input.

Using RTC images in dB scale is not recommended when using this tool, as those datasets are already in a log scale. Refer to the  tutorial for the Scale Conversion Tool  to learn how to convert RTC products from one scale to another.

You can  download the toolbox  from ASF's website. Extract the zip file to the directory of your choice on your computer.

Step through the slideshow below to learn how to use the tool in ArcGIS Pro. This tool can also be used in ArcMap; the process is generally the same, but the interface looks different.

When using the tool in ArcGIS Pro, hover over any of the blue information icons in the Geoprocessing dialog for more information about the parameters or the tool in general. The full tool information can be viewed by right-clicking the tool in the Catalog window and selecting View Metadata.

When using the tool in ArcMap, click the Show Help button in the tool dialog box to see the information about specific parameters or the tool in general in the Help pane. The full tool information can be viewed by right-clicking the tool in the Catalog window and selecting Item Description.

The default output raster is a stretched grayscale image. In some cases where change is very significant and discrete, this may be sufficient to catch the eye, but in many cases it is helpful to change the display to a classified color ramp. A classified map allows you to quantify the change and draw attention to areas where significant change has occurred, rather than visualize the full spectrum of change values.

Heavy rainfall in Central Michigan in May 2020 resulted in a breach of the Edenville Dam (shown in the map above). This resulted in significant hydrologic changes as other dams subsequently failed and surface water was redistributed across the landscape.

We will use the same RTC products that were used in the  Reclassify RTC Tutorial , but focus on the Edenville area rather than the Shiawassee area further south.

Step through the slideshow below to learn how to use the Log Difference approach to identify and visualize areas with significant changes in backscatter.

The color classified version of the raster clearly brings attention to areas where change has occurred. The yellow pixels have undergone little change (log difference values of -0.15 to 0.15), light blue pixels have lower backscatter in the later acquisition than in the first acquisition, and dark blue pixels have undergone significant decrease in backscatter through time. Orange pixels display areas with increased backscatter, while red indicates areas with significant increases in backscatter through time.

You can adjust the number of classes and the intervals between the upper values as desired depending on the extent of change you are interested in detecting. If you are only interested in very large changes, you could use 3 classes and set the upper values to be quite high. If you want to detect small amounts of change, you can set the break values to be closer to 0.

Most of the change in this example represents a decrease in backscatter over time.

This can be caused by a number of different factors. The input rasters for the log difference calculation displayed in this demo are the co-polarized backscatter (VV), which are most sensitive to surface roughness and soil moisture. A decrease can be caused by rough surfaces becoming smoother, and/or a decrease in soil moisture.

• Significant decreases can indicate flooding, as unflooded areas generally exhibit higher backscatter than standing water, and this may be occurring in some areas (such as in the flooded portion of the Shiawassee National Wildlfe Refuge). Most of the blue areas in this difference raster are probably not due to standing water, however.
• After flooding or heavy rains, agricultural fields can become caked with dried mud as the floodwater recedes, which may leave a smoother surface with lower moisture content than the soil prior to the heavy rainfall. The peak flooding took place about halfway between the first and second Sentinel-1 acquisitions, so there would likely have been time for excess water to drain away, leaving behind harder, smoother, drier surfaces.
• Lake Huron had much lower backscatter on May 27 than on May 15. This is due to different wind conditions on the two acquisition dates. Large lakes are more susceptible to wind-driven wave action than smaller lakes or ponds, and can be extremely variable in backscatter from one acquisition to another.

There are some isolated areas with increased backscatter.

The most distinctive area with an increase in backscatter is associated with the Edenville Dam and other dams downstream. As you can explore in the map below, there is a sizeable reservoir upstream of the Edenville Dam visible in the RTC image from May 15. In the image from May 27, the reservoir has drained, leaving behind bare soil.

Standing water without significant wind impacts returns very little signal, so the reservoir area had very low backscatter on May 15. In the image acquired on May 27, the backscatter values in the drained reservoir area are similar to those of the agricultural fields in the area that exhibited lower backscatter in the May 27 acquisition than they did before the flooding. These values seem to indicate dried muck.