CASA is a highly automated feature detection/extraction system. The system contains global libraries of expected spatial, visual, and near-infrared spectral and contextual signatures for the feature of interest (in this case clouds and cloud shadows). For example, depending on the Sun’s angle and viewpoint of the sensor, all of the clouds in any particular image should have shadows in similar locations. CASA uses information like this to confirm or reject elements in a given image as clouds. The system goes through an iterative, hierarchical self-learning process, identifying clouds based on a comparison to the characteristics in its global library. Definitive clouds are logged in a local library for additional comparison, helping CASA learn with increasing accuracy how to identify clouds and cloud shadows. The system produces a raster mask, showing per pixel cloud and cloud shadow contamination of the image. Clouds of different types can be assigned different color shades.
When tested on Landsat 5 and 7 images, CASA comes within 10 percent of the visual cloud estimate for 94 percent of the images tested—comparable to and often exceeding ACCA’s capabilities.
“CASA would be the perfect technology for the next-generation Landsat sensor,” says Hulina.
Since establishing CASA’s effectiveness with Landsat imagery, GDA has taken advantage of the system’s versatility and high level of automation to provide cloud detection services for a host of space-based sensors. GDA has proven the CASA software effective for cloud and cloud shadow detection in high-resolution imagery from commercial satellites such as QuickBird (owned and operated by DigitalGlobe), SPOT (Spot Image), IKONOS, and OrbView (both GeoEye). The company also completed a NASA Dual-Use Technology Development contract with Stennis to adapt the CASA algorithms to the Indian Remote Sensing-P6 Advanced Wide Field Sensor, or AWiFS, data. CASA’s automation enables it to process large data sets in short periods of time, allowing GDA to guarantee return of cloud masks within minutes of receiving the raw data on its servers. Using the CASA output, GDA can also remove identified clouds and cloud shadows on a per pixel basis, backfill them, and radiometrically normalize the changes from one image to another—all without the lengthy and expensive process of removing them by hand. The company counts a number of private remote sensing imagery firms among its CASA clients, as well as the U.S. Department of Agriculture (USDA) and the National Geospatial Intelligence Agency.
“The NASA SBIR contracts allowed us to get our start and develop our intellectual property using sound science,” says Hulina. “To have the research and development funding and the backing of NASA to go out into the commercial market has been key for us.”
Hulina also notes that GDA has since branched out into other categories of feature detection. Using the same technology it applied for CASA, GDA can provide detection services for virtually any predefined class of features. The company has entered into Phase III SBIRs with the USGS and USDA Forest Service for the development of feature detection systems for stream networks, riparian buffers (vegetated areas along streams or rivers), and certain land covers. The company also has contracts from the USDA Foreign Agricultural Service for detecting crop fields around the globe. GDA provides these clients with outputs like crop acreage maps or flood maps derived from the satellite imagery, allowing them to accurately evaluate key information about everything from agricultural production to the impact of pollution.
As concerns about the effects of climate change and population growth lead to a greater need for the valuable data acquired through remote sensing, GDA’s NASA SBIR-developed feature detection capabilities will likely be in high demand.