Categories
Models can sometimes get a bad rap as black boxes. To help demystify AgroIBIS, the principal model we’re using in FEWscapes, this post provides a brief peek under its hood.
AgroIBIS is a process-based model, which means it simulates processes that happen within agricultural ecosystems, such as photosynthesis, plant respiration, and nutrient cycling. Since these processes follow the laws of nature, no matter the time or place they’re in, and they are measurable in real life, process-based models have the advantages of flexibility and an ability to tell a more accurate story of what is happening on a landscape – and what could happen in the future.
“Using this model allows us to account for variability across a wide range of conditions,” says Kelsie Ferin, a postdoctoral researcher on the FEWscapes project.
Ecological processes don’t occur uniformly across a landscape, and neither are landscapes uniform – think of the variation in soil types, elevation, vegetation, etc. With process-based models, modelers can account for this variability by dividing the modeled landscape into smaller chunks, resulting in a grid.
While the same processes are simulated within each grid cell, or pixel, each tells its own story based on its respective chunk of the landscape.
In AgroIBIS, modelers can shrink or expand those pixels to nearly any resolution they desire, capturing pieces of the landscape from as fine a scale as 30 square meters to whole continents. The trick is setting a resolution that is good enough to answer the questions at hand.
Similar to a computer or television screen, a fine resolution in AgroIBIS can provide a much clearer picture, but also requires so much computing power it could take forever to run. But if the resolution is too large, you miss a lot of detail.
To simulate scenarios for the Upper Mississippi River Basin, the FEWscapes modeling team settled on a resolution of one square kilometer, a “nice round number,” according to scientist Eric Booth, that both aligns with the available data and can account for meaningful features of the landscape.
“It definitely tells a more interesting story and, I think, more accurately reflects the heterogeneity of what we see out of a plane window,” said Booth.
But no matter its size, “there’s a lot that goes on in a pixel,” said Ferin.
Each pixel’s story has two dimensions: what is happening across space and what is happening across time.
The spatial story includes events on, above, and below the ground and all the interactions in between, in what is called the soil-plant-atmosphere continuum.
Below the ground, the model simulates processes and conditions in the soil column, such as nitrogen cycling and soil moisture. For FEWscapes, we are simulating to a depth of 10 meters, but similar to pixel size, soil depth is adjustable.
These below-ground conditions interact with what happens on land, where AgroIBIS simulates processes involved in the plant life cycle and crop management, with the capacity to represent 15 different crops, grasses, trees, and shrubs. Of course, plants are also interacting with the atmosphere – exchanging gases through plant respiration, photosynthesizing, and contending with the weather – interactions that AgroIBIS also simulates.
Which brings us to the temporal story each pixel tells. AgroIBIS can compute these processes in the soil-plant-atmosphere continuum as they occur at varying time intervals – from hourly to seasonally to yearly.
For FEWscapes, most of the processes will be calculated at a daily interval, except the processes that are happening to plant leaves, such as water vapor loss and photosynthesis, which will be calculated at hourly intervals (or three times an hour, in the case of photosynthesis).
“This allows us to more accurately represent what’s actually happening in the field,” said Ferin.
The pixels can pick up plant stress more accurately at the hourly scale – for example, how fluctuations in temperature and soil moisture throughout a day affect a plant. As a result, AgroIBIS can provide relatively reliable predictions of crop yields.
One fine point to make is that each pixel can simulate only one crop at a time. For example, corn and soy can’t be running in the same pixel for the same time interval, but the crop within that pixel could be switched over time. That said, natural plant types can coexist in the same pixel – for example, forest and grass can cohabit a pixel meant to represent savanna.
Another important piece of the pixel story is that AgroIBIS simulates these processes in each pixel independently from one another. In other words, what happens in that pixel, stays in that pixel.
This is where AgroIBIS’ complementary model, THMB, enters the story. THMB is a hydrologic routing model, which means it takes the outputs from AgroIBIS and weaves together each pixel’s story by simulating the transport of water and nutrients from one pixel to the next, downstream through the system. This team effort between AgroIBIS and THMB is what gives us the big picture of landscape change and impacts on food, energy, water, and ecosystems.
The independent nature of pixels is one of AgroIBIS’ computational superpowers, because it makes it computationally efficient. Modelers can send the data from each pixel to its own computer processor, instead of relying on a single processor to munch on all the data from all the pixels at the same time, thereby saving computational time and enabling the analysis of how each pixel’s unique story unfolds.
Important to understanding these unfolding stories is AgroIBIS’ other superpower, its process-based nature. Since each pixel is simulating timeless ecological processes, they can handle just about any novel condition you can throw at them, even if those conditions have never occurred in the historical data.
This superpower is super important when you’re talking about climate change. For example, climate change is expected to elevate carbon dioxide in the atmosphere and temperatures, and AgroIBIS can simulate the stress those conditions could have on future crops – even down to the leaf level.
Of course, no model is perfect, and AgroIBIS can’t tell the full story.
“We have a lot of [factors that influence agroecosystems] in there, but not all of them,” said Ferin.
Factors that aren’t currently in the models include some greenhouse gas emissions (namely, nitrous oxide), crop pests, soil microbes (which are difficult to model), and some management practices, such as explicit tile drainage (which is also hard to model given limited data on where exactly tile drains lie and whether they’re still functioning).
This is not to say it would be impossible to simulate some of these missing elements. It’s more a matter of what is currently built into the model and what would need to be added.
“There are always some processes we just don’t understand fully or are too complicated or computationally expensive to model,” said Booth.
Learn more about the full suite of models used in FEWscapes here.