Retooling USDA imagery to cool cities
As global temperatures rise so, too, does the importance of finding ways to reduce sources of heat. Big data can help.
Scientists at Lawrence Berkeley National Laboratory have developed a program to analyze and display the albedo – or solar reflectance – of rooftops. According to the scientists at LBNL’s Heat Island Group, that’s important because roofs that aren’t efficient in reflecting light contribute to global warming in several ways. Not only do roofs that absorb sunlight heat up the building underneath, they also result in higher cooling costs and that increased use of power increases greenhouse gas emissions. And when many buildings are involved, “urban heat islands” are created, with attendant increases in pollution and illnesses.
The interesting thing about the Heat Island Group’s albedo project is that the researchers were able to do the analysis using existing data that had already been collected by the Agriculture Department’s National Agriculture Imagery Program. The NAIP data, originally collected for crop analysis, was gathered by airplanes equipped with Leica sensors that record light reflected from the surface.
“The NAIP imagery is freely available to the public in a form that looks like a nice photograph but it’s not scientific data so we couldn’t use that directly,” said George Ban-Weiss, a member of the Lawrence Berkeley team who is now professor of civil and environment engineering at the University of Southern California. “We went back and acquired the raw imagery before it was processed. We basically repurposed that data.”
The raw data is measurements of reflected light from the ground in four frequencies – corresponding to red, green, blue and near infrared – at a resolution of one pixel per meter.
Since many factors other than the composition and color of rooftops can affect the reflected light – including pollution and humidity in the air as well as the angle at which the light is detected – the team had to develop algorithms to make corrections in the sensor data.
“The atmosphere correction is based on a pretty complicated algorithm that figures out the properties of the atmosphere by looking at the measure of light over dark pixels,” Ban-Weiss said. “With dark pixels, you expect very little reflectance. When you end up measuring some reflectance, that’s because the atmosphere is reflecting sunlight even before it reaches the surface. The measure is of light you’re sensing where you expect there to be no light.”
The team also took ground measurements of albedo to further adjust the correction of the sensor data.
Once the data was analyzed, the team generated an interactive map of seven California cities: Los Angeles, Long Beach, San Diego, Bakersfield, Sacramento, San Francisco and San Jose. The map, which can be accessed at http://albedomap.lbl.gov, allows policymakers and consumers alike to check the albedo readings of an entire region or zoom into view the readings of a single roof. Clicking on roof will pop up a window that compares that roof’s albedo to the regional average.
The Heat Island Group has plan to further extend its albedo analysis. “Our next logical step would be to use this information to characterize the albedo of other man-made surfaces,” said Ronnen Levinson, LBNL staff scientist. The biggest obstacle the team faces, however, is surprisingly low tech – getting the appropriate shape files for the maps.
While the team was able to buy building shape files for current maps that had been created using LiDAR detection, shape files are in short supply for many areas and for many types of man-made surfaces. With pavements, -- the team’s next target surface -- Levinson said acquiring such shape files is problematic.
“The outlines of roads are easy to obtain, in most cases, because cities have maps, which are based on shape files bounding the roads,” Levinson said. “But roads are just one element of the picture. There are parking lots, sidewalks and driveways. Shapes for those are not readily available.”
When the rooftop project got under way, the team looked for software that could automatically extract such features from aerial imagery. “There were pretty reasonable solutions for identifying certain surface types, like impervious pavements versus vegetation, but distinguishing roofs from pavements was too challenging/expensive,” Levinson said.
That challenge, however, will be eased soon, judging from a recent blossoming of academic research on automatic feature extraction from images.
Posted by Patrick Marshall on Oct 14, 2014 at 11:38 AM