The rising mists

Rain and other precipitation deliver essential water for life on Earth, but what comes down must also go up for the cycle to continue in a harmonious way. How much water we have for planting, drinking and many other activities depends on how much falls, but what returns to the atmosphere is just as important. Evaporated water, either from puddles on the sidewalk, water in a lake, river or ocean, and water that a plant or soil gives off in transpiration or soil evaporation, are all part of a process called evapotranspiration (ET). In order to accurately study how much water there is to go around, scientists need to know the difference between precipitation and ET. Measuring how much water falls is far simpler than measuring how much rises back into the atmosphere, so researchers from the University of Illinois Urbana-Champaign (U. of I.) have created a computer model that uses artificial intelligence (AI) to assist them in these predictions. The process involves remote sensing estimates of evapotranspiration, and it takes a supercomputer like NCSA’s Delta to make sure the results come in a timely fashion.

The research team is comprised of a cross-discipline group of experts, including Jeongho Han, a doctoral student in the Department of Agricultural and Biological Engineering (ABE), part of the College of Agricultural, Consumer and Environmental Sciences and The Grainger College of Engineering at Illinois and co-authors Jorge A. Guzman, research assistant professor in ABE. This project is also part of a larger project on soil erosion funded by the USDA National Institute of Food and Agriculture. Maria L. Chu, an associate professor in ABE, is the principal investigator on that project and co-author of the new paper.

“We used the NCSA Delta system to research, develop, and implement the Dynamic Land Cover Evapotranspiration Model Algorithm (DyLEMa),” said Guzman, speaking on behalf of the team. “Evapotranspiration (ET), the fluxes of water moving from the soil to the atmosphere, is challenging to measure or predict. In our approach, we used NASA, USGS, NOAA and USDA data for training and implementing machine learning models to predict unavailable remote sensing daily ET across Illinois for 20 years (from 2000 to 2020). To complete this data, we developed a computational framework that worked across ET’s spatial and temporal scales, requiring ample storage and rapid data access (data-intensive) and vast parallelized computation (computing-intensive). That is where the NCSA Delta system came to play an essential role in our work. Using the high-performance computing (HPC) resources at NCSA, we efficiently handled these data-intensive and computation-intensive tasks.”

Guzman and Chu have used high-performance computing (HPC) resources prior to this project and consider themselves very familiar with using them in their research. That wasn’t the case for the entire team. “Our graduate students are commonly unfamiliar with these systems,” said Guzman. However, they were all impressed with how well NCSA supported the various levels of expertise. Guzman’s team went through the U.S. National Science Foundation’s ACCESS program to get their Delta allocation, and for some of their team, this was the first time using HPC.

“NCSA, the ACCESS program, and Delta provide a fantastic infrastructure and support for Jeongho to help him rapidly learn the HPC environment and navigate it to resolve issues while developing his research goals,” said Guzman. “As he had no prior experience using HPC and NCSA resources, it was initially challenging, but NCSA provided good support and documentation, making the learning process more effective. Additionally, the support team from NCSA and ACCESS were always available and helped resolve the issues rapidly. Their support was invaluable when we encountered problems setting up model runs and transferring data, especially resolving the issues where small details arose impacting HPC performance.”

Supercomputers are a tremendous tool for researchers, hastening results in ways traditional computing can’t. “Utilizing HPC resources was significant in achieving and accelerating our research,” said Guzman. “Although we have dedicated computing and data nodes in our group, the goal of this research task was overwhelming. We ended up using more than 400,000 core hours on Delta and perhaps another similar on the Illinois Campus Cluster Program (ICCP) and dedicated resources. Compared to traditional computing resources, HPC allowed us to complete our goal in a shorter time by several orders of magnitude, which would otherwise take years.”

The group recently published their paper “Dynamic land cover evapotranspiration model algorithm: DyLEMa,” in Computers and Electronics in Agriculture. The US Department of Agriculture – National Institute for Food and Agriculture (NIFA) award number 2019-67019-29884, the NSF’s ACCESS program and the Illinois Campus Cluster provided funding for this research.

To read more in-depth details about the research, you can find the original story here on the College of Agricultural, Consumer and Environmental Sciences (ACES) website: Illinois researchers develop an AI model to reduce uncertainty in evapotranspiration prediction.

ABOUT DELTA

NCSA combines next-generation processor architectures and NVIDIA graphics processors with forward-looking user interfaces and file systems to create Delta, a powerful computing and data analysis resource that is part of the national cyberinfrastructure ecosystem through ACCESS. The project partners with the Science Gateways Community Institute to empower broad communities of researchers to easily access Delta and with the University of Illinois Division of Disability Resources & Educational Services and the School of Information Sciences to explore and reduce barriers to access. Delta is funded through NSF OAC 2005572.