Physics-guided spatio-temporal super-resolution of an advection-diffusion process
- Kyongmin Yeo
- Andres Codas Duarte
- et al.
- 2021
- INFORMS 2021
Dr. Levente Klein is a Research Staff Member in the Climate and Sustainability department at the IBM T.J. Watson Research Center, Yorktown Heights, NY. His research is focused on developing AI models to analyze massive amount of data and combine them efficiently with AI models. A special focus is on developing applications that can be used in industrial environment ranging from AI architecture to operational global environmental monitoring. Current research focus on developing digital twin for oil and gas industry by integration of physics constrained Neural Network and real time data with a final goal to develop observing system to analyze the optimal sensor placements and parameters that needs to be measured. To advance some of the digital twin work, foundational models are developed to trace Green House Gases source locations and quantify emission volumes.
Nature based carbon sequestration in vegetations using satellite imagery and AI models is a secondary area of research to enable a more transparent carbon trading markets, to minimize amount of CO2 in atmosphere. The carbon sequestration work is a continuation of Vegetation Management solutions that is a commercial offerings from IBM.
Previous work encompass the development of a large scale sensor networks for automation and integration of satellite imagery in daily operation of vineyards working closely with Gallo wine for variable rate irrigation. Integration of vegetation signal from Landsat data and weather data was combined to provide an irrigation advice 10 days in advance at 30 m spatial resolution. Digitization of the irrigation system allowed to relate a single Landsat pixel to the corresponding irrigation zone in the vineyard. Analytics was developed to forecast the water requirement and integrated into an automatic control system that dispense water at 30 m by 30 m spatial resolution cross a drip irrigated vineyard.
An irrigation system prototype was developed in collaboration with EJ Gallo winery on a 10 acre area in Lodi, CA and run for 3 years. The results demonstrated 20 % increase in grape yield and 10% improvement in water use by grapes. The work was awarded the 2015 Vintage Report Award for innovation in water use and adoption of technology in grape growing operations.
The variable rate irrigation work triggered the development of a geospatial data platform that spatially and temporally aligns data. The platform called Physical Analytics Integrated Repository and Services (PAIRS) can accelerate geospatial analytics by a factor of ten relying on curated data sets compared with cases when original and raw datasets are used. The PAIRS technology enable geo spatial analytics for data scientists without extensive geographical and cartography background. The PAIRS technology was awarded an IBM Outstanding Technical Achievement Award in 2017.
Levente is a member of the American Physical Society (APS), American Vacuum Society (AVS) and the NY Academy of Sciences.