How to Think About Our Environmental Future

Shortcomings and Suggestions for Environmental Forecasting

{photo_credit}

A new paper by Breakthrough's Conservation Director Linus Blomqvist and Breakthrough Senior Fellow Barry Brook considers the problems with major environmental forecasts and proposes focusing more on process than endgame.

November 6, 2016 | Linus Blomqvist,

Where will the world be two, three, or four decades from now? Will carbon emissions have gone down to safe levels? Will the area of farmland have peaked and declined? Will the global population have reached 9, or 10, or 11 billion?

The future is unknowable, but that hasn’t stopped scholars from trying to answer these questions. Nor should it. Forecasting trends in resource use, population growth, and environmental impacts can help anticipate risks and opportunities, as well as assess the consequences of choices made today.

But the way forecasts are usually done today make them less illuminating than they might appear. In a paper published recently in Basic and Applied Ecology, Barry Brook and I review the state of the art in forecasting, identifying some benefits and strengths in different forecasting methods, as well as many pitfalls.

One reason forecasts fail to shed light on important questions is – to state the obvious – that most of them will inevitably turn out to be far off the mark. For example, a look at what people twenty or thirty years ago were predicting that today’s energy consumption would be doesn’t exactly inspire confidence. There are many reasons why predictions go wrong. Trends that once looked steady might be disrupted by unexpected events. Tiny differences in expected rates of change make a huge difference when compounded over decades.

But focusing only on the numerical predictions or scenario outputs themselves in some ways misses the point. What can be far more illuminating is to look at the process – the “how” and “why” the environmental future will unfold – rather than the prediction – where we expect to be in 20, 50, or 80 years. Constructing forecasts forces us to understand the patterns and mechanisms unfolding today that will shape trends in the future.
 
This is where many of the forecasting methods fall short. Take the Millennium Ecosystem Assessment’s scenarios, for example. These aren’t predictions per se, but they could serve the same purpose of giving insight into the processes and decisions that shape the future of the environment. Each scenario is based on a “storyline,” which is essentially a bundle of assumptions about the future state of the global population, the economy, technology, and so on. These assumptions are fed into Integrated Assessment Models that turn these inputs into outputs like the extent of future land-use change.
 
One problem with such scenarios is that assumptions about different factors often offset one another. In the same scenario, we might see crop yields go up and biofuels being widely adopted, or we might see organic farming expanding and meat consumption going down. Higher crop yields would shrink the area of cropland whereas expanding biofuels would have the opposite effect. When all of these are bundled together, more or less arbitrarily, it’s all but impossible to know the effect of each one. This can be exacerbated by the opaque nature of Integrated Assessment Models, whose inner workings are understood, at best, only by their creators.

When different factors offset each other, we are led to believe that the range of possible outcomes is quite narrow, when in fact other combinations of the same factors could generate far more divergent outcomes. Thus, while storyline scenarios can be illustrative and easy to communicate, they can also mask important dynamics at play within the model, and show only a narrow window of possible futures.

Another method consists in extrapolating trends from observed relationships between socio-economic and environmental variables. This can be useful in drawing attention to key drivers of environmental change, but the method also has its weaknesses. The seemingly contradictory conclusions of two studies looking at the relationship between deforestation and urbanization can illustrate this. One of these studies, by Joseph Wright and Helene Muller-Landau, found rural population density to be strongly associated with deforestation. For that reason, they predicted that urbanization would reduce pressures on tropical forests this century.

The other study, led by Ruth DeFries, saw a strong positive connection between urbanization and forest loss at the country level, thus reaching the opposite conclusion. Both are right in their own way, but neither really illuminates how urbanization affects deforestation, making it that much harder to make robust forecasts. As a result, it’s hard to know whether policies to encourage urbanization – all else equal – will lead to more or less deforestation in any given setting.

A third set of methods decomposes large-scale trends into a handful of drivers based on the IPAT identity, where environmental impacts are a product of population, consumption, and technology. This method has the advantage of being transparent and easily understood. For example, Jesse Ausubel, Iddo Wernick, and Paul Waggoner break down changes in cropland area into five factors: population, per-capita income, food consumption, the ratio of crop production to the total supply of calories consumed by people, and crop yields. Each of these change by some percentage per year, with the result, in this case, that global cropland area is forecast to decline from now on.

But this model too has its shortcomings. The future rates of change in each variable has to be assumed, often just loosely based on past trends. And all the rates of change are, by the nature of the model, exponential, when in fact some real-world trends are linear. Crop yields, for instance, very reliably grow in linear fashion, which means that the percentage increase in a given year declines over time. Assuming exponential rates will therefore overestimate future yield growth, often by a wide margin.

All forecasts are subject to a difficult tradeoff. On the one hand, including many variables and building sophisticated models makes forecasts more representative of how the world works, in all its complexity. Yet by bundling lots of assumptions and variables together, these models often end up as black boxes, where the impacts of any given policy are hidden or inscrutable. Simpler models can be more transparent and accessible, but might not be very accurate at characterizing trends in, for example, environmental impacts. And looking at correlations doesn’t necessarily shed much light on how and why things are changing in one direction or another.

This offers some lessons both for producers and consumers of forecasts. Instead of just searching for clear pictures of the future, we should start paying more attention to the mechanics and processes that will build that future.

What are the technologies most ready to replace fossil fuels and mitigate carbon emissions? What are the practices that most efficiently and sustainably improve agricultural yields? What types of policies have been shown to quicken or slow urbanization historically? These questions, and many others, can’t yield simple and eye-catching numbers. But they can focus priorities and offer actionable policy lessons.


Comments