Uncertainties in Projections of Future Effects of Climate on Salmon

This paper will discuss the need for scientists to explicitly consider both physical and biological uncertainties when making forecasts of effects of climate change on salmon. Too often, researchers on this topic only take into account uncertainties generated by using different scenarios of climatic driving variables, which are produced by different global climate models (GCMs).  Furthermore, few researchers take into account the effects of natural variability and observation error on uncertainty in parameter estimates of the oceanographic and biological response models, making it difficult to discern true underlying relationships.  Recently, structural uncertainty about such relationships among components of aquatic systems has been recognized as a larger source of uncertainty than natural variation and observation error. However, few scientists who work on links between climate and fish explore the influence on fish responses to climate of structural uncertainty in functional forms of relationships among physical oceanographic processes and/or biological responses. Instead, most analyses use only the best point estimates of parameters and relationships among system components.  Such limited analyses of uncertainties can drastically underestimate the range of plausible responses of salmon populations to climatic change and may also underestimate risks.  I will draw upon the methods of advanced statistics, risk assessment, risk communication, and risk management to show how to deal with five major sources of uncertainty (1) natural variability at both short and long periods and at various spatial scales, (2) observation error, (3) structural complexity, (4) outcome uncertainty (i.e., implementation uncertainty), and (5) inadequate communication among scientists, decision makers, and stakeholders. The first two sources of uncertainty can be at least partially dealt with by (a) using state-space and/or errors-in-variables models that explicitly estimate parameters for each of natural variation and observation error, and (b) fitting hierarchical models to reflect spatial covariation among nearby populations arising from shared environmental drivers. One way to deal with structural uncertainty is to conduct sensitivity analyses using stochastic simulation models that are run across hundreds of hypotheses (i.e., assumed structural relationships). The fourth source of uncertainty reflects distributions of outcomes due to management regulations not having the expected effect. When this uncertainty is combined with the previous three, closed-loop simulations (management strategy evaluations) can be run to determine the best management option. Finally, this paper will describe some methods for dealing with communication about uncertainty and risk among scientists, decision makers, and stakeholders, a topic that has received too little research.