Other team members:
 

Scientific Focus of Data Sets: Earth systems science

Description of Data Sets:

This presentation primarily explores the uses of high-performance GPU visualizations for two classes of data: Sea Surface Temperature and Sea Surface Height Anomaly. For Sea Surface Temperature, the primary data sets used are the PO.DAAC GHRSST Level 4 MUR Global Sea Surface Temperature Data Set and the AVHRR Sea Surface Temperature Data Set. These large data sets are used to explore global and local trends in ocean temperature related to climate change and the El Niño/La Niña phenomena. The primary Sea Surface Height data sets are the ECCO ETAN Sea Surface Height Anomaly data set and the combined Topex/Poseidon/Jason/OSTM Sea Surface Height Anomaly SSH Data Set. These are also used to explore sea level rise due to climate change and local trends due to atmospheric anomalies and ocean circulation. The MODIS True Color Corrected Reflectance data set is also used to demonstrate machine learning applications to optical measurements for popular release. All data sets are examined over a two to three decade period from the early 1990s to 2017. All data sets are publicly available from NASA Data Active Archive Centers, accessed through the PO.DAAC website and NASA Earth Data Search.

Scientific Potential of Presentation:

Earth scientists today face a problem they never expected to have: too much data. The ever-increasing size and volume of NASA Earth Science data sets makes it harder and harder for scientists to quickly visualize and understand the scientific potential of new data sets. Typically, new satellite data products require computationally intensive preprocessing to produce large scale analytics and gridded map representations (L4) at a low enough resolution to be rendered and manipulated on a traditional CPU – this must be done every day for daily products. Even then, statistical analysis and visualization is too slow to be performed on-demand at a speed that makes data exploration intuitive and responsive. New technologies, like GPUs, may be able to change this. This presentation introduces a new open-source GPU-accelerated framework for on-demand data visualization and analytics for use with NASA Earth Science data products. The NASA GPU Data Visualization Prototype is able to dynamically produce imagery from raw floating-point datasets at multiple resolutions in real time, generate time-series and correlation analysis for enormous products far faster than traditional CPUs, and apply the latest advances in machine learning as part of the visualization process. Because these visualizations are generated in real time from raw floating point data, map visualizations can be reprojected, manipulated with machine learning algorithms, and subjected to full fidelity statistical analysis in a small fraction of a second. This increase in speed and responsiveness can transform the way scientists investigate data – reducing the barrier for scientists to ask and answer questions about large data sets and hastening scientific discovery. This system can be deployed in the cloud and manipulated in a web-browser, eliminating hardware constraints, and opens up powerful new avenues for science data visualization on NASA data products. This presentation emphasizes the scientific utility of this technology for analyzing large climate datasets, particularly PO.DAAC Sea Surface Temperature products, ECCO Sea Level data sets, and atmospheric greenhouse gas levels, visualized and analyzed over a nearly twenty year period for trends and correlations. We examine the 2015 El Niño event using dynamic analytics and graphics generated on a GPU from AVHRR data, and use this GPU framework to study trends in sea surface temperature (AVHRR) due to climate change. This presentation and prototype aims to demonstrate the potentially transformative power of GPUs to perform exploratory data analysis and visualizations in a fraction of a second which on CPUs might take many minutes, giving scientists the power to rapidly iterate and explore NASA data.