Skip to main content

NOAA Satellites Provide Detailed Imagery of Michigan Floods

November 9, 2020
Image at NESDIS
Tuesday, May 26, 2020

During a recent major flooding event, scientists unveiled a newly-developed algorithm derived from satellite data to understand the extent of the damage.

Downtown Midland, Mich., is flooded May 20, 2020.
Downtown Midland, Mich., is flooded May 20. (Photo: Kelly Jordan and Junfu Han, Detroit Free Press)

OnMay 19, after three days of heavy rain and flooding in Michigan, two dams failed, flooding roads, engulfing homes and bridges, and forcing nearly 11,000 people in nearby towns to evacuate.

The Edenville Dam collapsedthat Tuesday afternoon, unleashing the Wixom Lake into the Tittabawassee River. Later that day, floodwater began gushing over the Sanford Dam downstream. By Wednesday evening, the Tittabawassee River had peaked at 35.05 feet, 10 feet above flood levels, beating the previous record of 33.9 feet set in 1986, according to the National Weather Service , which called the flooding “catastrophic.”

During a major flooding event, local forecasters and first responders rely on satellite data, along with other tools like aerial mapping, to understand the extent of damage.

In this case, the Operational Land Imager (or OLI, below) on the Landsat 8 satellite provided a detailed view of the Tittabawassee River and surrounding area. But Landsat, which captures a narrow swath just 180 km, or 111 miles wide on each pass, revisits a single spot in the U.S. about once every 16 days, and is therefore not always available to capture flooded areas when they happen.

NASA Earth Observatory image of the flooding in Midland, Mich. (light brown areas), as seen from the U.S. Geological Survey satellite, Landsat.
NASA Earth Observatory image of the flooding in Midland, Mich. (light brown areas), as seen from the U.S. Geological Survey satellite, Landsat. Credit: Joshua Stevens.

On the night the dams breached, the NOAA-20 and Suomi-NPP satellites of the Joint Polar Satellite System (JPSS) also observed the floods, allowing scientists to reveal the capabilities of an improved flood product.

Scientists at George Mason University have been working to create a product that gets daily measurements of the entire United States at a similar resolution to Landsat, using theVisible Infrared Imaging Radiometer Suite (VIIRS)instrument on the NOAA-20 and Suomi NPP satellites. This is especially important now, as the NOAA Climate Prediction Center in March f orecast widespread river flooding this spring , due to expected above-average temperatures and rainfall.

JPSS polar-orbiting satellites observe a width of 1,901 miles at every pass, which allows each to image the entire globe twice a day. And recent improvements to the flood algorithm mean the satellites can now see flooded regions at a resolution of 30 meters, or about 98 feet – that’s 12 times sharper than the previous VIIRS flood product.

“We’re demonstrating the new improvements by looking at the dam,” JPSS Chief Scientist Mitch Goldberg said. “And we’re showing that the new algorithm compares really well against Landsat.”

In 2016, when this technology first debuted during a joint JPSS/George Mason University seminar, the model could only produce preliminary unsteady results in specific regions. “In recent months, we've made great progress,” said Sanmei Li, the algorithm developer at George Mason University. “Now, the model can be used to generate 30-meter flood maps from VIIRS in the continental United States with pretty robust performance.”

In the map below, you can see Michigan’s Midland County and the Tittabawassee River, with flooding shown in red. The images show the original VIIRS flood product at a coarse 375-meter resolution (top) and the improved flood product at a 30-meter resolution (bottom), with much more detail on impacted areas.

Map showing flooded areas around the Tittabawassee River in red.
Image at NESDIS

The new flood product, funded by the JPSS Proving Ground initiative , relies on an algorithm that takes the satellite data of floodwater at a 375-meter resolution, and, using an understanding of terrain and surface elevations, gives a higher resolution (30m) estimate of flood extent within each satellite image pixel.

"VIIRS has such a large swath, and Landsat’s swath is much more narrow,” Goldberg said. “It takes 16 days for the whole world to be covered by the smaller Landsat swaths which is only 115 miles wide, versus VIIRS at 1,901 miles. This new technique allows us to improve resolution in flood mapping and still maintain global coverage.”