By Liz Allen
Just how well can we model the impacts that climate change will have on agriculture in the Pacific Northwest? Simply put, there will always be uncertainty about exactly how the climate of the future will differ from historical patterns and what those changes will mean for farmers in the region. How accurately models can project future conditions is a big and complex topic. Discussions of uncertainty can quickly veer into fairly esoteric scientific and philosophical territory, leading to questions such as: What constitutes scientific proof? How can we account for inherent randomness within systems when studying the future? And, how useful are model projections when there is uncertainty associated with nearly every input variable? In practice, however, modeling is tremendously useful because it allows researchers and agricultural decision-makers to put bounds on uncertainty. Thus, modeling helps make plans for the future in spite of uncertainty.Making decisions in the face of multiple unknowns is nothing new. Farmers constantly make business and land management decisions in the context of uncertainty. For example: Will heavy spring rain cause muddy conditions that force a grower to sow wheat later in the season? Will they have to contend with crop damage due to Russian wheat aphids this year? What effect will aphids have on yields? And just how much more life is left in their farm equipment before it needs to be replaced? Questions about future climate change impacts are also critically important for producers, and clear communication about model uncertainty is needed so that agricultural professionals can apply information from models to help them understand climate-related risks as they make real-world decisions.
In order to understand climate change projections, we first need to recognize that uncertainty comes in several different flavors, some that we can do something about, and others that we cannot. The first, epistemic uncertainty, is incomplete knowledge about a topic that we can learn more about by conducting further research (thus, it is reducible uncertainty). For example, we may not know how aphid populations respond to increased winter precipitation, but we can collect data in fields and in laboratory experiments to quantify—and better understand—the link between the aphids’ behavior and precipitation. This differs from stochastic uncertainty, which is the result of inherent variability within a system. Not clear? Try this: We cannot say how much winter precipitation there will be, exactly, in 2034, because some aspects of oceanic and atmospheric circulation patterns random. However, as we develop increasingly sophisticated and detailed models of the global climate system we can give a smaller range of precipitation levels that are possible in 2034, and what amounts are most likely based upon the trends we observe at meteorological stations. A third variety of uncertainty, sometimes called fundamental uncertainty, can be described as ignorance-about-ignorance. For example, suppose a new virus emerges in 2034 and decimates aphid populations. A model cannot predict an event like this because it involves variables—that new virus—that fall outside the boundaries of what the model developers were considering when they created the model.
Once sources of uncertainty are defined, those uncertainties can be explicitly visualized and discussed in presentations of model outputs. For example, in the 2011 Columbia River Basin Supply and Demand Forecast, scientists projected 2030s monthly water supply within the Yakima River Basin and communicated several sources of uncertainty. First, these projections account for epistemic uncertainty that is due to differences among various regional climate models, which each project different monthly temperature and precipitation trends. The range of regional climate model projections is represented by a green cloud. At the same time, stochastic uncertainty also exists due to year-to-year variability in streamflow. Monthly water supply is modeled for a year with average streamflow, as well as a particularly wet year (with annual streamflow higher than 80% of the years on record) and a particularly dry year (with annual streamflow lower than 20% of the years on record). Describing and quantifying these dimensions of uncertainty in the Supply and Demand Forecast enhanced the usability of model results for decision-makers. A Yakima Basin grower could look at these projections and better understand likely changes in early spring water supply within the next two decades. This understanding may inform long-term decisions about cropping strategies and infrastructure investments. Considering climate models alongside personal goals and risk tolerance supports informed planning in spite of the multiple types of uncertainty that still remain.
If you’re interested in digging deeper into the discussion of sources of uncertainty and how those uncertainties are communicated, check out the fact sheet titled Modeling Environmental Change. This resource uses examples from the 2011 Columbia River Basin Supply and Demand Forecast to guide readers in interpreting environmental model results. The fact sheet could be a useful reference to have on hand when the updated 2016 Columbia River Supply and Demand Forecast is released in November.