InSAR Water Masking

The  Alaska Satellite Facility  offers free  On-Demand InSAR processing for Sentinel-1 . When ordering InSAR products through  Vertex  or  programmatically , there are a few  processing options available . One of these options is to apply a water mask during phase unwrapping.

This tutorial demonstrates the impact that water masking can have on your phase unwrapping results, and discusses changes that ASF has made to its water masking approach over time.

Originally published September 20, 2022, with subsequent updates.

One potential source of error during phase unwrapping is when large areas of open water exhibit high coherence from one acquisition to the next. Phase differences over the water can be unwrapped, making it appear that deformation occurred over the water, impacting phase unwrapping results on land as well.

Waterbodies are often filtered out by the  coherence threshold  applied during phase unwrapping, but there are times when the coherence values over water are high enough to pass through this threshold. Applying the water mask is a way to ensure that these areas are not considered valid during phase unwrapping.

The water mask used for processing is  included in the InSAR product package  for reference, even if you choose not to apply water masking during phase unwrapping.

Move the swipe bar to see the difference between an unmasked unwrapped interferogram (left), and the same pair with a water mask applied during phase unwrapping (right). Red indicates areas of relative uplift, and blue are areas of relative subsidence.

Some areas display very different displacement values when the mask is applied compared to when there is no mask applied.

In some locations, the unmasked interferogram displays much greater displacement than expected, while in other locations the displacement signals are muted or even showing the opposite direction of displacement than expected.

In February 2024, we updated the source data used for generating the water mask. We now use a combination of  OpenStreetMap  and  ESA WorldCover  data. Areas within Canada, Alaska, and Russia are primarily covered by ESA WorldCover data, while the rest of the world is covered by OpenStreetMaps data.

The water mask identifies coastal waters and most inland waterbodies. All remaining pixels (land, islands in large lakes, very small inland waterbodies, and landfast Antarctic ice) are considered to be not water. Source data for the water mask is only available from 85°S to 85°N. Areas north of 85°N are all treated as water, and areas south of 85°S are all treated as not water.

To learn more about the transition to this new water mask, skip ahead to the  OpenStreetMap & ESA WorldCover  topic in the Changes to Water Mask Strategy section. There is a  comparison exercise  that demonstrates the difference the change to the reference water mask file has made to the output InSAR products.

Up until the new OSM/WorldCover mask was implemented, we generated the water mask used for On Demand InSAR using the  Global Self-consistent, Hierarchical, High-resolution Geography Database (GSHHG)  dataset.

We combined the GSHHG full-resolution L1 (boundary between land and ocean) and L5 (boundary between Antarctic landfast ice and ocean) datasets, and removed the L2 (boundary between land and large inland waterbodies) dataset minus the L3 (islands) dataset.

We have made a couple of major changes to our water masking strategy over the years that On Demand InSAR has been available. Click on the topics below to learn more about the changes and why they happened:

1.  Removing the shoreline buffer from the water mask (27-Sept-2022) 
2.  New water mask using OpenStreetMap and ESA WorldCover data (15-Feb-2024) 

In ASF's original implementation of a water mask for InSAR processing, we used a version of the mask where a 3-km buffer had been applied to coastal shorelines and a 5-km buffer had been applied to inland shorelines.

On September 27, 2022, we transitioned to using a water mask that did not have a buffer applied. Let's explore why...

ASF implemented the use of a new water mask for InSAR processing on February 15, 2024. The surface water extent data available from  OpenStreetMap  and  ESA WorldCover  were a significant improvement over the outdated version of  GSHHG  that we were using. The data is more recent, more detailed, and has fewer geolocation artifacts.

 OpenStreetMap  is a crowd-sourced open-data mapping effort. The  OSM database  of geographic features can be accessed by anyone, and it includes a number of categories that can be used to map surface water extent.

OSM data was used to generate the water mask for all areas except Canada, Alaska, and Russia. To extract the relevant water extent data from the OSM database, the following filters were applied:

• wr/natural = water
• landuse = reservoir
• waterway = *

In many cases, waterway features stretch from one bank to the other, so islands within those waterways would not be identified as land. To remove islands from the water mask extent, the following filters were applied to the extracted surface water dataset:

• place = island
• place = islet

The resulting list of features was exported as a shapefile, then converted to raster format for inclusion in the reference water mask.

In October 2021, the European Space Agency (ESA) released the first version of its global land cover dataset,  WorldCover . It uses remote sensing data from the Sentinel-1 and Sentinel-2 missions to generate land cover classes, including water. More information is available from the  ESA WorldCover 2020 website .

This dataset was used to generate the water masks for Canada, Alaska, and Russia. It includes one class for permanent water bodies. The version 1.0 source rasters were downloaded from the  ESA WorldCover 2020 Downloader site . They were reclassified so that all areas with a value of 80 (Permanent water bodies) were defined as water, and all other values were considered not water.

The water mask rasters generated from the OSM and WorldCover datasets were mosaicked together, then tiled to 5° latitude by 5° longitude for storage.

The water mask identifies coastal waters and most inland waterbodies. All remaining pixels (land, islands, very small inland waterbodies, and landfast Antarctic ice) are considered to be not water.

Source data for the water mask is only available from 85°S to 85°N. Areas north of 85°N are all treated as water, and areas south of 85°S are all treated as not water.

When an InSAR job is submitted for  ASF's On Demand processing , the coordinates of the four corners of the input Sentinel-1 scene are used to find the water mask tiles that cover the scene. If the scene crosses multiple tiles, the necessary tiles are mosaicked together. The water mask is then clipped to match the spatial extent of the input Sentinel-1 scene.

The values of the source water mask are defined to meet the requirements of the InSAR processing software. Water pixels are set to a value of 0, and all remaining pixels are set to a value of 1.

This mask is then used as an input, along with coherence values, to generate the  validity mask  used for phase unwrapping. The 0-value water pixels are excluded from use in phase unwrapping.

A copy of the water mask is always included in the InSAR product package for reference, even if the user chose not to select the option to apply the water mask. Water pixels have a value of 0, and all other pixels are assigned a value of 1.

In some areas, this change will have minimal impacts. In other areas, however, there is a significant improvement. Use the  Overall Comparison Swipe Tool  below to compare an interferogram masked with the new water mask (to the left of the swipe bar) to an interferogram masked with the old water mask (to the right of the swipe bar).

Both of the unwrapped interferograms displayed in the tool have the same value range applied to the color bar, so a particular shade of the color in the image represents the same pixel value in both interferograms.

Keep in mind that negative values indicate a shorter distance between the sensor and the ground surface, so the dark red pixels indicate uplift. Dark blue pixels indicate movement away from the sensor, while white pixels have undergone no change. Pale blue and red colors have undergone very little change.

Neither of the interferograms have strong subsidence signals (neither has a phase difference value greater than 20), but the value range was set to be equivalent above and below zero in order to make the visualizations more intuitive. This way the narrow band of white in the color bar is centered on zero, and the paler the color, the closer it is to zero.

You can pan and zoom to different areas in the map to compare the results in more detail. We will highlight a few key differences in the next section.

Let's take a look at a few specific areas where we see differences.

Get in touch with us!

If any of the changes we've made to the water mask have impacted your results, we want to know!

It would be great to hear that your results have improved, but we're particularly interested to hear if you have degraded results in your area of interest due to any of the changes that have been made. Does the mask not work well in a particular area of interest? Is land excluded from analysis because of inaccuracies in the mask?

Feedback from our On Demand InSAR users will help us determine if we need to make additional changes to our water masking approach.

Connect with the On Demand team on  Gitter , or send an email to  uso@asf.alaska.edu .

Find more information about On Demand InSAR processing at  ASF's HyP3 Documentation website , or learn more about the Alaska Satellite Facility at the  ASF Website .