Analysis Ready and Interoperable: Taming Multi and Hyperspectral Imagery

Since 1972 scientists have used multispectral data to identify materials and detect processes on Earth and in space. Today hyperspectral imageries (HI), with possibly hundreds of contiguous bands and very high resolutions, are increasingly applied towards similar ends. Scientists are eager to extract information embedded in HI, however, their volume is a barrier to their use. Also, scientists want to integrate both multispectral and hyperspectral data into their research, e.g., for climate change questions and to create data records over time.

The Hylatis platform is a NASA AIST 2016 project targeting multispectral and hyperspectral imagery (MaHI), making them performant, analysis ready and interoperable. The project chose to focus on the HySICS hyperspectral dataset and also POLDER, GOES, and MODIS multispectral datasets. Leveraging the Spark big data engine, Hylatis provides a uniform interface to these datasets, with operations tailored for MaHI, including spatial and spectral subsetting and resampling for the purpose of interpolation onto a common grid.

Though similar in nature, MaHI datasets are highly varied in their spectral coverage, native grid choice, geolocation representation, storage format, etc. What does it take to make these disparate datasets analytics ready and interoperable?

In Hylatis, datasets are represented as mathematical sampled functions of independent and dependent variables. This common data model facilitates the representation of MaHI datasets and the provision of operations on them in the form a functional algebra that follows mathematical rules, such as variable extraction, filtering on data values, resampling, and operations to reshape the data for performance. This model also supports moving the computation to the data through function application. Metadata is integrated into the platform.

This talk describes Hylatis and further discusses providing analytics ready data and interoperability for MaHI.