Usually, it is difficult to predict how ecosystems will respond to natural and human-induced changes beforehand. Let’s take the Amazon rainforest as an example, it is difficult to precisely conclude how it will respond to changing climate and possible future warming scenarios. Similarly, human activities such as land use changes disrupt the natural processes. For example, consider the construction of a dam.
The complex interactions in nature make it difficult to understand the full extent of such changes on ecosystems. Ideally, we would run a controlled experiment to observe the effects, but that’s only possible in a lab, not in the real world. Let’s stick to our dam construction example. If our goal is to understand the extent of the environmental impact of this and future changes, it is impossible to precisely predict those based on existing literature because, in reality, the response of ecosystems to such changes varies from place to place.
On the other hand, conducting a controlled experiment is impossible because, in reality, finding two catchments with exactly similar conditions is unrealistic. Even if we manage to achieve that, running an actual experiment, building a dam in one and leaving the other untouched, would be unrealistic, both ethically and financially
But we can’t simply throw up our hands and say, “We’ll deal with it when it happens.” Waiting for disaster to strike could lead to devastating human and economic losses. At the same time, it’s unrealistic and undesirable to stop all human interaction with nature.
What we need is a way to anticipate how natural and human-induced changes might impact ecosystems, and in turn, how those changes could affect our lives. Maybe not precisely, but to some reasonable extent, that helps guide our decisions. That’s where environmental models come in. They allow us to explore possible futures, assess risks, and make informed choices with the need to protect the natural systems we depend on.
With the rapid development in computational power over the years, the field of environmental modelling has grown tremendously. Take hydrological models as an example, back in the 1960s and ’70s, they were simple “lumped” models, treating an entire catchment as a single bucket, where water just moved from one compartment to another.
Today, we have highly complex models that simulate fine-scale physical processes and spatial patterns with a distributed grid approach. (If you’re curious about what makes a model complex, check out my previous blog post!) These advanced models offer a much deeper understanding of natural systems than their simpler counterparts. But the question I am asking in this blog is: Do we always need that level of complexity? I don’t think so!
My argument is based on a few key shortcomings of such complex models, including the following.
- One major challenge with complex models is that their physical understanding of environmental processes often comes from controlled lab experiments. These experiments do not always reflect real-world conditions. After all, nature doesn’t behave the same way everywhere. Environmental processes are incredibly diverse.
- As mentioned earlier, the fundamental purpose or application of these environmental models is to help us anticipate future changes we can expect in the future as a result of natural and human activities. However, complex models are very data-demanding. Unfortunately, one of the major concerns in the field of environmental sciences is the lack of data availability. Therefore, the transferability of such complex models is limited because of that.
- Building on that, the development of most models relies on historical observations. Therefore, the approach to develop a model that fits perfectly for past observations might not work for the future because of overparameterization.
- Another major shortcoming of these complex models is the demand for high computational processing power. When combined with the need for high-quality data, this makes them impractical for many settings, particularly in developing countries where such resources are limited.
- Additionally, the complexity of these models can act as a barrier to knowledge transfer. Advanced models often require specialized expertise to understand, implement, and interpret. This makes it difficult for decision-makers, practitioners, or even researchers outside of the immediate modelling community to engage with them meaningfully.
These are only a few, yet major obstacles with complex environmental models. At the end, we might get the perfect model, but what’s the point of investing hours and hours in developing that if it is not transferable and not contributing to the fundamental purpose, decision-making? So, what exactly should be the focus in the field of environmental modelling?
The answer is optimal complexity.
Which means, the goal of a model should be to identify the most influential natural processes that affect the model application and try to include those processes in the model in a way that minimizes input data requirement and computational demand. This approach ensures the transferability and understanding of the model. In my personal opinion, this is why it is important to define the application of the model upfront and then start making decisions related to model structure, because it is impossible to develop a model that serves all needs.
Take hydrology as an example. A model designed to predict floods will have a very different structure from one focused on water quality. The processes that drive flood events, like intense rainfall and runoff, are not the same as those influencing nutrient loading or pollution transport. Now, imagine a decision-maker whose main concern is the future variability of water quality. They don’t need a model that also simulates floods. This is why the development of models for specific purposes is important. A complex model might fail to serve in this example scenario because of its extensive input data demand, some of which has no relation to a water quality analysis.
The takeaway?
Environmental models should prioritise minimal complexity. That doesn’t mean oversimplifying but rather, building models that are fit for purpose and focused on specific tasks. Trying to solve too many problems with one model often leads to solving none of them effectively. The figure below perfectly summarises the message I want to deliver through this blog post.
Further Readings:
