In policy analysis, we are almost always trying to quantify what might happen in a future scenario where a particular policy is enacted. A lot of the time, we are focused on short time frames, like what will happen in the next five or ten years after a policy is enacted.
However, sometimes we need to look much further ahead. There has been a recent push to study the impact today’s economic policies have on future generations in fields like poverty and education, but most research still focuses on the impacts people alive today will experience.
One area of policy analysis that is almost exclusively forward-looking is climate policy. With climate policy, policymakers today need to consider not only their short term needs, but the needs of people far into the future.
This presents a major challenge for analysts because our forecasting tools are largely designed for short-term thinking. We can’t just use the same tools to model the global climate into the future that we use for other questions.
So, let’s talk about climate models, and what sets them apart from the type of models you may be more familiar with.
When we talk about any type of forecasting model, one of the first things we need to understand is whether a model is fixed or random. The technical terms for this is whether a model is deterministic or stochastic.
In deterministic models, the same inputs will always return the same output. If you plug in X, you will always end up with Y. The downside of these models is that they usually require stricter assumptions and a better understanding of the underlying systems at work, but their advantage is that they enable you to look significantly farther out into the future.
With stochastic models, the same set of inputs doesn’t always yield the same output. These types of models allow analysts to account for some variability and uncertainty in their assumptions which can be useful in situations where we might be missing some information about what influences the underlying systems.
Models don’t have to be entirely stochastic or deterministic. A lot of the time, it is beneficial to include both types of elements. That way, analysts can take advantage of specific deterministic knowledge they do have, while accounting for some of the variability that exists in some of the less understood processes.
One example of a simple climate model comes from the University Corporation for Atmospheric Research. This is a deterministic model that just only takes into account the amount of carbon in the atmosphere. They use an established link between the amount of carbon and temperature to estimate what the global climate will look like in the future.
This model highlights another key characteristic of climate models, their resolution.
The University Corporation for Atmospheric Research model makes global climate predictions. If you were interested in understanding what might happen in your city, state, country, or even hemisphere, this model can’t help you.
Instead, you might be interested in the Representative Concentration Pathways, sometimes called “RCPs.” This set of models separates the globe up into over 500,000 grid cells, each with their own emissions data. This enables researchers to study how local changes can impact global climate change.
There are multiple different Representative Concentration Pathways that each come with a slightly different set of assumptions about what the world is going to look like in the future. Using a scenario analysis like this is one way that modelers can capture some of the stochastic uncertainty without specifically defining a variable process. You can just test what happens with a different set of assumptions.
One challenge that all predictive models face is that they are using past data to predict the future. This assumes that the underlying processes the these models are based on aren’t going to fundamentally change anytime soon.
Climate models push the boundaries of traditional forecasting by demanding models that account for both immense complexity and long-term uncertainty. As we confront the realities of climate change, improving our models and adapting our analytical tools will be essential. Good models give us good information, and good information is the key to good policy decisions.