Climate Forecast Maps as a Communication and Decision-Support Tool: An Empirical Test with Prospective Policy Makers

Article excerpt

Introduction

Since antiquity, people have been using maps to represent space. By doing so, one can "look at" and think about the represented space while not actually standing within that space. On a map, one can record and organize information about the represented space at a reduced scale, thus making it possible to think about regions impossible to view in their entirety from any single vantage point. And, importantly for this study, one can communicate information about the represented space to another human being.

In this paper, we examine communication issues concerning climate forecasts, using currently issued forecast maps targeted to a broad spectrum of decision makers, including agricultural and environmental policy makers. Climate forecast maps differ from the more studied weather maps (Monmonier 2000), in that the forecast looks months to seasons into the future, rather than hours to days, and thus incorporates a greater degree of uncertainty. Climate has long been of major concern due to its potential impact on people's activities. Researchers have been attempting to offer seasonal or inter-annual climate forecasts several months ahead of time, so that people can make important decisions that are sensitive to future climate conditions (e.g., grain production, water management, natural disaster control). Climate prediction skills have advanced in the last few decades, especially since the success of an experimental, model-derived prediction of El Nino in the late 1980s (Cane et al. 1986; see also Goddard et al. 2001 for a review of the current state of prediction efforts).

Current climate prediction models are generally based on the El Nino-Southern Oscillation (ENSO), which refers to shifts in sea surface temperatures in the equatorial Pacific and related shifts in barometric pressure gradients and wind patterns in the tropical Pacific. Researchers have shown that ENSO activity has a large global impact on inter-annual climate variability and, importantly for this study, that it is highly correlated with agricultural production. For example, Cane et al. (1994) showed that more than 60 percent of variance in maize yield in Zimbabwe was explained by an index of ENSO, and that model-derived predictions of ENSO provided fairly accurate forecasts of maize yield. Other researchers have also discussed potential benefits of climate forecasts to agriculture, at various parts of the world and at various scales (e.g., Hammer et al. 2001; Hansen 2002; Jones et al. 2000).

Such potential benefits of climate forecasts, however, cannot be fully realized unless people understand and use climate forecasts appropriately. Some researchers discussed the difficulty that people have in interpreting and applying climate forecasts in practice and argued for the necessity of systematically examining communication issues about climate forecasts so that people can take advantage of their potential benefits (e.g., Nicholls 1999; Pfaff et al. 1999). We consider two possible sources of difficulty in understanding climate forecasts: (a) the probabilistic nature of climate forecasts and (b) the complex presentation formats of current forecast products.

First, in climate prediction, it is not possible to offer a deterministic forecast; instead, forecasts are currently given as probabilities of predicted precipitation and temperature for specific regions falling into the upper, middle, and lower one-third in the past years' database for the regions. There is an extensive literature on human understanding of probabilities and decision-making under uncertainty (e.g., Kahneman et al. 1982), which shows that people do have difficulty understanding probabilities. For instance, people make probability judgments on the basis of heuristics, such as "representativeness" and "availability," which lead to serious biases (Tversky and Kahneman 1974). And people (not only laypersons but also professionals such as physicians) tend to make errors in statistical reasoning, particularly neglecting base rates in Bayesian inference problems (Eddy 1982; Tversky and Kahneman 1982). …