Food Production & HPC: Understanding Plant Growth From Above & Below
Applying Ones and Zeros to Fields and Farms
From Thomas Malthus in the 18th century to Paul Ehrlich in the 1960s, people have long feared that population would outstrip the planet’s capacity to feed people. What has kept those predictions at bay has been technology—not only in the soil, but more recently in the field of high-performance computing (HPC). To learn where the frontier stands, I spoke to two leading researchers who are working to apply ones and zeroes to fields and farms.
Both use HPC to tackle previously insurmountable challenges of scaling: For one of the researchers in the interview below, that means taking data at the resolution level of a single local corn field and modeling it to apply across entire growing regions. For the other, it means creating the ability to make long-term determinations about climate with the same granular specificity that traditionally applies only to short-term weather.
Katherine (Kate) Evans is the Acting Division Director of the Computational Sciences and Engineering Division, and the group leader for Computational Earth Sciences, at the Oak Ridge National Laboratory (ORNL) in Tennessee. She leads efforts to improve earth system model (ESM) evaluation techniques for the large-scale, long-term climate trends that shape food production. As part of ORNL, Evans and her team are users of Summit—currently the world’s fastest supercomputer—which uses more than 27,000 GPUs and more than 9,000 CPUs to take on AI tasks of unprecedented size and complexity.
Kaiyu Guan is a Blue Waters Assistant Professor in ecohydrology and remote sensing in the Department of Natural Resources and Environmental Sciences (NRES), College of Agricultural, Consumer and Environmental Sciences (ACES) at the University of Illinois at Urbana-Champaign. His title reflects his biggest tool: the Blue Waters petascale supercomputer he and his colleagues use to enhance the power of satellite data and process-based models to provide real-time information on the health and yield of agricultural fields.
Michela: Can you give us an overview of your work and how it relates to food production?
Kate: My group uses Summit, the largest computer in the world, to develop, execute and analyze large scale coupled earth system models. These are the largest, most complicated models, used for all kinds of predictions and data analysis, including of climate change.
The models for weather and climate are similar, but they run differently. Weather models can use finer spatial resolutions because they only simulate about 10 days forward. Climate models, on the other hand, need to simulate the earth over tens to hundreds of years. So we can’t operationally run with the same resolution as weather—but with this computer, we can run climate models with such fine scales that we’re basically able to resolve weather features and extremes over climate length scales.
Kaiyu: Our work starts with satellite data that contain visible, near-visible, thermal, and microwave electromagnetic readings from crops as they grow. These observational data help constrain our simulation models, so that we can also infer what’s going below ground, such as soil moisture and available nitrogen in the soil. To make a real-world impact, we have to go down to each field and even sub-field scales to provide actionable information for farmers. And we are not doing it for one field, but all the fields in a region, for example, the whole U.S. Corn Belt. So supercomputers are the necessary tools to allow us go to field-scale modeling and also scale up to the continental scales.
Using this computational power, we have the ability to monitor and virtually model individual fields and play what-if scenarios to assess various farmers’ practices, such as irrigation, nitrogen fertilizer application, and the use of cover crops [off-season plantings used to manage soil quality].
Michela: What are some of the specific techniques you’re using and results you’re achieving?
Kate: One example that’s near and dear to my heart is atmospheric blocking. Sometimes the jet stream becomes diverted for various meteorological reasons. That’s called blocked flow, and it moves typical weather patterns—if you’re in one location, you’ll get extra rain, and if you’re in another location, you’ll get droughts. If you’re able to predict block formation using weather models, that’s good information. But if you can capture their behavior on a climate scale, you can examine trends, for example how often will these blocks occur.
Let me give you an example of how that can apply to food production. One of our graduate students, Deeksha Rastogi, is using these models to research apple trees. They need so many cold days to be dormant, and then they need so many cool days to get the buds going, and so forth. All those processes require certain types of weather over a season. We’ve taken the data from our high-resolution climate models and applied it over the United States to look at current and future potential changes to the kind of weather that apple trees need to thrive.
And this type of analysis can be applied to any kind of agricultural product. It’s showing how the changing patterns of a region in terms of weather transmits to the climate scale in terms of what’s going to happen to impact long-term agricultural plans.
Kaiyu: Our technology can tell you things such as how much water is in your field. We can use near-IR readings to track the chlorophyll and nitrogen content in the leaves, or measure solar-induced fluorescence that is an indicator of plant photosynthesis, or gauge the canopy temperature to see whether the corn plants are water-stressed. Indicators like that can tell you how much irrigation or nitrogen fertilizer you need to put on the field, and those decisions can help growers increase yield and produce more food. For policymakers, it can also help drive sustainability by detecting excess water or fertilizer use.
Michela: How does HPC contribute to making results like that possible?
Kate: Different agricultural products all require different bits of information. That’s why it’s important for our models to be developed on these big computers because you need a model that can do a little bit of everything.
If you had the largest computer that you could in the world—which we do!—what could you do that you couldn’t before? Models are never perfect. They’re always some degree of simplification of what the earth is actually doing. But the larger the computer you have, the fewer of those approximations you have to make. I think of high performance computing as a way to answer questions that we couldn’t using any other platform.
Kaiyu: Modeling for one field is already a pretty complicated job. To do it for the whole Corn Belt, you need a supercomputer. Looking at the entire Corn Belt this way, at the resolution that lets you measure individual fields, is a totally different question. The reason HPC and Blue Waters makes our work possible is scale—the number of nodes we can employ at one time. We tend to use thousands of nodes at one time for much of our work.
Michela: What’s one thing about HPC and agriculture that would surprise the average person?
Kate: A lot of what HPC does is drive the next generation of regular computing. What’s happening now at the Summit scale is going to soon be something we see on our desktops and eventually in our iPhones. My iPhone is what we used to have in a supercomputer here at Oak Ridge many years back. It always trickles down over time.
In agriculture, that means what is cutting edge now on Summit will eventually lead to multiple studies of agriculture and climate change using a growing number of computing systems that are more widely accessible. With that, we’ll be able to use models like ours to better adjust which crops would go where in the future. We’ll use improved data to better adjust the yield.
Kaiyu: People know the visible spectrum can show us things such as leaf area and biomass. Using those metrics, we can make inferences about things like photosynthesis, evaporative transpiration, or the yield.
But the grain to be harvested, in this case the corn cob, is below the leaf canopy and not visible from above. People may be surprised to learn the remote sensing data we use goes beyond the visible spectrum to show us other indications using near-infrared and microwave data. For example, microwave data can penetrate the canopy and even the soil the help measure moisture in real time.
Michela: Can you give us a glimpse around the corner at what the future might hold in each of your respective fields?
Kate: I think there’ll be more connections to air quality. So we will look at not just the effects of carbon dioxide, but other chemicals in the air. We’re also going to be much more data driven in our models, and that will allow us to incorporate more effects at scale than we couldn’t capture before, say for example, very high-resolution precipitation. Rain and snow are really hard to forecast, and that’s critically important for agriculture.
Kaiyu: One of our frontiers is communication. Right now we’re working with Illinois Corn Growers’ Association and Illinois Farm Bureau to build the channels that can get this information into the hands of growers who can put it to use. Eventually, we’d like to see growers across the entire Corn Belt able to freely access this real-time tracking information about their own fields through an established infrastructure. We’d also like to expand the application of these techniques to other parts of the world, such as South America or Africa, and into other crop areas, such as wheat, cotton, or potatoes. Our vision is for a farmer to be able use a smartphone app anywhere in the world to have this information in an efficient way.
–––
Michela Taufer, PhD, General Chair, SC19
Michela Taufer is the Dongarra Professor in the Min H. Kao Department of Electrical Engineering & Computer Science, Tickle College of Engineering, University of Tennessee, Knoxville.