Can we predict the future of farming? The truth about biogeochemical models.

What are biogeochemical models?

There is growing interest in improving agricultural sustainability and implementing sustainable practices worldwide. In the agricultural sector, one of the most relevant questions is how to improve management practices to balance optimized crop production with minimal environmental impacts, such as nutrient losses and greenhouse gas emissions. Therefore, we are interested in utilizing biogeochemical models to predict crop and soil responses to management, land use, and climate change.

Researchers perform a wide range of studies, from laboratory experiments assessing detailed mechanisms at the micro-scale, to field studies, and to simulations of changes in plant production and nutrient inputs/outputs in response to global changes at the regional scale. Laboratory experiments focus on detailed mechanisms underlying soil structure formation and breakdown, the structure and functioning of soil biota, and nutrient cycles at the micro-scale. At the same time, many researchers have focused on how nutrient cycling is affected by plant-soil interactions and management practices on farm and decadal scales.

Typically, a biogeochemical model consists of several interconnected sub-models. Each sub-model consists of a set of equations that attempt to describe seemingly complex plant or soil processes, based on our quantitative and qualitative understanding of these systems. Model development is mostly data-driven.

Why modeling?

By nature, crop and soil processes tend to vary significantly in time and space. This is because climate, soil, and other environmental conditions change when moving from one location to another, as well as with time. This is what we call site heterogeneity, and it is considered a major source of uncertainty that has led to many debates in the crop and soil sciences. Let us imagine that we were fully resourceful and able to set up an experiment to accurately and continuously measure the high spatial and temporal variability of agricultural processes. If this were the case, we would not need to bother modeling what happens in a plant-soil system (Biosphere 2 may be a good example). Obviously, and unfortunately, we do not have such a luxury. Direct measurements of these processes can be time-consuming and costly. As a result, field experiments can only investigate the effect of a limited number of management practices on crop and soil processes. It is also only possible to collect snapshot data from these experiments. In comparison, biogeochemical models offer a practical option that allows scientists to predict how crop and soil processes will vary in time and space under various conditions. These models have been developed primarily to estimate crop productivity as well as carbon, nitrogen, and other nutrient cycles under land use or management changes. Recently, the use of these models has been expanded as a decision-making tool for policy makers, farmers, and researchers, and they are now used to produce data for national greenhouse gas inventories related to climate change.

Simple vs. complex models?

The representation of crop and soil processes involves some level of simplification, which compensates for current knowledge gaps in biogeochemical cycles. Nevertheless, over the last few decades, these models have increased in complexity, which in return allows them to reproduce a number of biochemical processes in greater detail and to be used to make predictions under unmeasured conditions.

It is ok not to trust modeling. Then what?

There are several biogeochemical models that have gained popularity in recent years. Most of these models utilize many (e.g., several hundred) parameters. These model parameters represent various rate constants, multipliers, limits, and scaling factors. Their values can vary within a large domain, based on long-term research, laboratory measurements, literature studies, and expert knowledge if detailed information is unavailable. Some of the parameter values are more certain than others. Therefore, the correct application of biogeochemical models requires parameter calibration. The accuracy with which a model reproduces agro-ecosystem processes depends on how well its parameters are calibrated to the available measured data. Although the complexity of biogeochemical models allows them to reproduce ecosystem processes in detail, this complexity is often disproportionate to the amount of data available for model calibration and validation. Apart from the baseline input data which are essential to characterize soil profiles, local weather conditions, and management practices, relatively limited amounts of data tend to be available to calibrate parameters and test how well a simplified model reproduces actual agro-ecosystems. Soil scientists and biogeochemical modelers are faced with the challenge of using very little data to adequately tune complex interactions in the models and produce accurate predictions (Figure 2). Furthermore, the data used in biogeochemical modeling are also accompanied by errors associated with sampling, analytical procedures, and data analysis. Consequently, no matter how well a model fits the measured data, errors in the measurements devalue the model predictions.

This suggests that even robust and complex models, which encompass our current knowledge about biogeochemical cycles, do not have the absolute ability to produce reliable and accurate predictions of soil organic carbon and soil emissions. However, once we estimate the uncertainty and error associated with their predictions, they provide an effective and inexpensive research tool that can support policy decisions at the local, regional, or national scale.