After any natural disaster—such as a flood, wildfire, or hurricane—one way to assess destruction and coordinate relief is to use satellite imagery for large-scale change detection, which uses a location’s past and present satellite imaging data to understand what was there before that isn’t there today.
“The easiest way to think of change detection is to imagine two images taken from above,” says Amanda Ziemann, a member of the Space Data Science and Systems group at Los Alamos National Laboratory. “The process of detecting changes digitally layers those images and looks for differences on the ground.”
For example, in before and after photos of a house in the middle of the woods, the before picture might show a reflective, metal roof and the green canopies of pine trees. The after photo, taken when a fire has already burned the area, might show a charred structure and matchstick tree trunks. A computer then calculates values at each pixel in each image, then calculates the differences between the two images. Wherever the difference is highest, scientists can expect a significant change to the environment.
“The easiest way to think of change detection is to imagine two images taken from above. The process of detecting change digitally layers those images and looks for differences.”
But measuring changes through satellites has always suffered from an Achilles’ heel: the information gathered by one satellite often can’t be compared to the information gathered by another satellite because satellites gather images in different ways.
Some satellites with multispectral imaging use green filters to take images of plants and blue filters to measure coastal erosion. Infrared satellites create images from temperature. Others use synthetic aperture radar to measure changes in surface height. The change detection method of computing values at each pixel, then measuring the difference, doesn’t work when two images are comprised differently. It would be like trying to compare numbers to letters or musical notes. Additionally, different satellite cameras are equipped with different numbers of signal measurements, or channels. So even if two satellites both have multispectral imaging cameras, they need the exact same number and type of color channels to compare data.
As a result, scientists often use the same satellite to collect data of an area once every orbit. In an emergency, however, waiting up to two weeks for the satellite to return to position means change detection is often too slow to be of immediate help.
Recently, Ziemann and her team developed a new approach called multi-sensor anomalous change detection (MSACD), essentially a universal satellite data translation device. It’s an idea so innovative that when Ziemann explained it to her former doctoral advisor at the Rochester Institute of Technology, he replied, “What? That doesn’t even sound possible.”
But Ziemann knew that although images might be comprised differently, what was being measured—a geospatial area—was the same. So using a computer algorithm, MSACD translates data from satellites and plots it on a joint distribution graph, which has an X and Y axis and is used to compare random variables. Wherever the dots vary widely, the most change has occurred.
In the future, scientists could aim a variety of satellites at a wildfire, for example. Within hours, MSACD could send automatic updates to emergency operations and aid groups that show how the fire changed direction or how many buildings were destroyed.
“We’ll be able to detect changes to the environment faster than ever before,” says Ziemann, noting that MSACD is in prototype demonstration stages of development. (It received funding in 2018.) “That has many applications—assessing damage to flooded coastal communities, detecting an adversary’s clandestine military base, spotting illegal loggers thinning the rainforest, determining the damage from a wildfire, and much more.”