by Sarah Evans
Ecological models allow us to predict important properties of ecosystems. This is an important task, as it’s extremely inefficient to measure these characteristics at every location on earth, and simply impossible to measure the properties in the future. For a long time we have organized the natural world by ecosystem or biome type – grasslands, tropical forest, tundra, and so on. These categories are also helpful in predicting properties using models because a lot of the same relationships exist in similar ecosystems, even when they are located all over the world. I wanted to help improve models that predict the amount of carbon and nitrogen stored in soil, which are essential to soil fertility. Soil carbon also has the potential to be released into the air as CO2 as temperature and decomposition increases with climate change. CO2 is already unusually high in the atmosphere, and if soil carbon were to contribute more to these levels under higher temperatures, climate and CO2 could start to feedback on one another in a vicious cycle.
With this in mind, I set out to see whether some of the models we’ve developed based on grasslands in the United States did a good job of predicting soil carbon levels in Chinese grasslands.
The US Great Plains and northern China’s grasslands have similar climates and vegetation types, so most models assume that relationships between climate and soil carbon in the US would serve well to predict carbon levels in China. We tested this by going back to the basics: measuring soil carbon in the field.
We spent several weeks in northern China sampling soil and collecting data on the climate, soil texture, and land use where the soil was sampled. We used the US model, replacing data from the US sites with data from Chinese sites, and found that the US model overestimated the amount of carbon in Chinese grassland soil and underestimated the amount of nitrogen (see figure).
This surprised us, as many models assume carbon storage is controlled by similar factors in all grasslands, and the US model even accurately predicted soil carbon levels in grasslands in Argentina. Even when we incorporated information about how the site was currently managed (e.g. intensively grazed, cultivated, etc.), which helps a bit in predicting carbon levels in the US, the model still failed.
So we tried to use a more complex model to produce better estimations. This additional complexity might not be very practical for predicting soil carbon levels on large scales as it would require that we measure more things, but it allowed us to learn which components were important that we might have missed. For example, we found that including a factor for nitrogen deposition from the atmosphere made a large difference. Nitrogen deposition levels are low in the US, and have historically been low in other grasslands, so it did not seem important to include in previous models. However, we found a few accounts that showed that population and industrialization in China in the last 50 years have increased N deposition, and our results suggest that perhaps this is something we should not ignore in models. Another thing we changed was the historical land use simulated by the model. In the US, grasslands have been farmed for a long time, but not nearly as long as in China. Because of the relatively short history of farming in the US, grazing pressure has not been an important factor to include in general models (though cultivation is important). But it turns out, the much longer history of land use in China may have left lasting effects on the soil carbon levels, even if grazing pressures are similar to US regions today. When we inserted a period of “intense grazing” several hundred years before the present, carbon estimates from the model were much closer to the ones we measured.
This may seem obvious: how you use the land will affect soil fertility, and how much nitrogen is deposited there will affect soil nitrogen. But the very subtle challenge in modeling is knowing what is important to include without including too much. If we included everything about an ecosystem in a model, sure, we’d get pretty accurate results. But we’d also have to go out to each ecosystem and measure all the properties of it, defeating the purpose of the model! In this study, we had a chance to point out that some models oversimplify relationships between climate and soil carbon in grasslands by assuming that grasslands in China behave similarly to grasslands in the US. Continuing to monitor the accuracy of these models will be especially important as the world is subject to more and more global changes, but I think gaining the ability to generalize, extrapolate and predict is certainly worth the effort.