Making Climate Forecast Information More Useful
Skillful climate forecasts are valuable to society to the extent that they provide knowledge that can be used to cope better with climate variations. This chapter examines what forecasts might offer to improve the outcomes of weather-sensitive activities and what is known about how individuals and organizations are likely to interpret and use forecast information. We first consider what kinds of climate forecast knowledge might prove valuable. We then examine the limited available information about how coping systems have actually responded to skillful seasonal-to-interannual climate forecasts, supplementing this with other sources of insight, including basic knowledge about human information processing and knowledge about human use of information in situations that may be relevant by analogy. This examination yields a set of hypotheses about the characteristics that make forecast messages and information systems useful.
Useful Information that Climate Forecasts Might Provide
Chapter 3 shows the variety of ways in which individuals and organizations cope with variable climates. Climate forecasting is intended to help them cope better, but not all forecast information will necessarily be useful toward this goal. Forecast information can have value only if people can change their actions in beneficial ways based on the content of the information. As the following examples show, different kinds of
forecast information are useful, depending on the climate-sensitive sector, the region, and the coping strategies used.
In agriculture, a forecast is useful to the extent that it permits more advantageous ex ante actions, such as altered choice of crop species and cultivars and timing of tillage (Mjelde et al., 1988) or altered composition or allocation of herds (Stafford Smith and Foran, 1992; Ellis and Swift, 1988). For example, a skillful forecast may allow a farmer to diversify less and to match cropping decisions more closely to expected climatic events. A farmer who can anticipate that rainfall is likely to be unusually ample can grow seeds that are sensitive to water availability to improve profits; conversely, a farmer who knows that there is a high probability that rainfall will be unusually low can conserve on inputs, use less water-sensitive inputs, or refrain from application of any unfruitful inputs at all. Forecasts of growing season length or degree-days may be useful in similar ways. However, forecasts are helpful only if they arrive before planting or stocking decisions are made and if the producer is capable of responding. Some responses, such as changing livestock species, may require resources available only to the most successful producers.
Regional conditions affect the usefulness of forecasts. In South Asia, where models of El Niño/Southern Oscillation (ENSO) allow for fairly skillful predictions of average temperature and precipitation several months in advance, it might seem that climate forecasts would be broadly useful to farmers. But this may not be so. Forecasts can benefit the 10 to 15 percent of farmers in the semiarid areas who would lose money by planting in bad-climate years (Rosenzweig and Binswanger, 1993): they could decide not to farm. But the majority of farmers, who can expect to profit even in a dry year, might not benefit from the forecasts. The reason is that no farming practices can be undertaken prior to the onset of the monsoon, so that even if a long-range forecast of the monsoon onset could be made, it would provide no benefit. A prediction of the magnitude of the monsoon may also provide no benefit to farmers whose practices would be the same regardless of its magnitude.
Institutional factors may affect the value of forecasts. In the United States, the usefulness of a climate forecast may depend in complex ways on whether a farmer is covered by crop insurance. Some analysts (e.g., Gardner et al., 1984) argue that federally subsidized crop insurance imposes a ''moral hazard'' by encouraging farmers to take imprudent risks, for example, by being less diversified and more dependent on dryland practices in regions of marginal climate than their uninsured counterparts. Insurance also decreases the incentive for farmers to change their practices on the basis of a climate forecast, since they are covered against disasters.
In water management, distinct kinds of forecast information are use-
ful depending on the decision and its context. For example, water managers in the western United States typically base streamflow forecasts on existing hydrologic conditions (e.g., current water content of the snowpack) and historic records of high, normal, and low precipitation during the remainder of the forecast period. This procedure gives managers a rough indication of the upper and lower bounds and most likely inflow conditions for the system. Climate forecasts can improve decisions based on this procedure if they provide more accurate expectations about rainfall at the watershed level. However, the value of such forecasts is likely to hinge on whether adequate representations of forecast accuracy and uncertainty are provided. Forecasts may also help water project managers inform irrigators, whose water entitlements are calculated as a share of the available supply, of impending shortfalls early in the season so they can make adjustments. Although such advice may be helpful, if it is based on a poor-quality forecast or on unskilled interpretation of the forecast, water users may take inappropriate actions, such as fallowing unnecessarily or incurring unneeded expenses for wells or water purchases to protect perennial crops. An example involving such outcomes is discussed in the next section.
The usefulness of forecasts also depends on the state of preexisting water management institutions. For example, it may be supposed that the prior appropriation system in the western United States is rigid, leaving water users with little discretion to make adjustments that take forecast information into account. However, by providing a clear link between water availability and use rights, the senior priority rule allows water users and managers to calculate the probability of obtaining water under any particular right given the predicted climatic conditions and to make appropriate investments or water purchases to achieve desired levels of reliability (Hutchins, 1971; Trelease, 1977). Forecasts that give additional lead time might also allow more efficient adjustments by enabling irrigation districts and individual irrigators to plan more effectively for fallowing, crop switching, or other methods of water use reduction and for improved operation of water banks.
The insurance industry and its clients might benefit from forecasts that accurately estimate the probabilities of hurricanes, floods, droughts, or wildfires striking policy holders in particular areas. For example, insurers and reinsurers could calculate premiums based on risk rather than history. However, this would be an improvement only if available predictions are sufficiently accurate and if insurance regulators allow the change. The usefulness of forecasts to insurers is also constrained by difficulties transforming the kinds of information forecasts provide into forms used in insurance firms' procedures of risk analysis (Golnaraghi, 1997). Crop insurers might use climate forecasts to decide how much
reinsurance to purchase. This information is needed with enough lead timewhich can be several monthsto sign with the federal government for reinsurance. Forecasts as little as one month in advance may be sufficient to help insurers provide farmers with risk-management services (Golnaraghi, 1997).
Climate forecasts might give public health systems an unprecedented degree of early warning of the likelihood of epidemics, based on climatic or ecological analysis before disease organisms appear. ENSO forecasts, to the extent they can be linked to conditions conducive to disease outbreaks (Epstein et al., 1995), may facilitate early public health interventions. Taking advantage of the forecasts would require a sufficient level of knowledge to link climate parameters to ecological events affecting disease organisms, an adequate surveillance system, and appropriate training and communication systems for health early warning. Given these advances, public health responses might include immunizations, neighborhood clean-ups, and pesticide applications. For example, a combination of climate forecasting and remote sensing imagery can help in preparing for outbreaks of eastern equine encephalitis by determining where and when temporary pools of standing water are likely to appear and how long they may last. With such information, it is possible to take preventive action to control the population of infected Aedes vexans mosquitoes with larvicide applications. Because maturation of larvae to adults occurs in about seven days, accurate information on standing pools of water after a rain is necessary within two days, to allow time for dip sampling and application of larvicide (Epstein et al., 1993a).
In the energy industry, improved forecast skill might help gas companies with inventory management and with anticipating price fluctuations. Hydro-dependent utilities might benefit from seasonal forecasts of precipitation and runoff, and utilities with seasonal demand profiles might benefit from seasonal forecasts of heating or cooling degree-days; their specific information needs and lead times are unknown.
These examples illustrate that the usefulness of climate forecast information depends on the match between various attributes of the information and the needs and capabilities of individuals and organizations who may be affected and on the ability of these users to get the information processed to fit their needs. Among the attributes of climate information that are frequently important are lead time, the particular climatic parameters being forecast, the spatial and temporal resolution of the forecast, and its accuracy. These are discussed in more detail at the end of the chapter.
Whatever new information forecasts provide, some actors may benefit more than others because they are in better positions to take advantage of the information. For example, some individuals may have more
savings or better access to credit that allows them to take better advantage of a forecast of favorable climatic conditions. Some may own a specialized resource, such as a senior water right or a piece of farmland whose value varies with climatic conditions. Some actors may gain advantage in contractual negotiations if they receive and correctly interpret forecast information earlier than others. Such distributional consequences are shaped by actors' situations and by the institutions that shape them (e.g., water law, insurance regulations), by the availability of insurance and credit, and by the design of disaster preparedness and relief programs. It is possible that, in some sectors or regions, the overall benefits of climate forecasts may be distributed so that some groups gain greatly while others do not benefit at all, or even find themselves worse off. If such outcomes arose, they might greatly dampen enthusiasm for climate forecasting.
Responses To Past Climate Predictions
A useful source of information on how weather-sensitive sectors and actors may respond to climate forecasts in the future is their response to past climate forecasts. Unfortunately, skillful forecasts are very recent, so there has been relatively little opportunity to learn from experience. A few case studies have been done of situations in which affected groups have acted on climate or hydrological forecasts on the time scale of months. Three are described below, with the tentative lessons that seem to flow from them. This body of research is far too limited to treat these lessons as more than hypotheses. However, they are valuable because they show responses to actual climate forecasts. Systematic studies based on responses to forecasts of the 1997-1998 El Niño could add greatly to understanding.
Drought Forecasts in the Yakima Valley
Glantz (1982) examined the case of an erroneous forecast of drought in the Yakima valley of Washington state in 1977. Irrigation in the Yakima valley supports some high-value crops, including orchards and mint. In February 1977, the Bureau of Reclamation forecast that water available for summer irrigation in the valley would be less than 50 percent of normal. On the basis of this forecast, they told senior water rights holders that they would receive 90 percent of their allocations and more junior water rights holders that they would only receive 6 percent of their normal allocationsinsufficient to protect their perennial crops and orchards from drought. Farmers responded by drilling deep wells at costs of $25,000 to $250,000 per farmer; deciding not to plant a crop but to fallow;
leasing or selling water to those with perennial crops at up to four times the normal price; transplanting valuable crops to regions with senior water rights; and weather modification activities costing $400,000.
As the season advanced, the bureau revised its forecast, and by May, long after most of these adjustments had been made, it announced that junior rights holders would, in fact, receive 50 percent of their allocations. By the end of the summer, it was clear that water supplies had been almost 83 percent of normal and that junior rights holders had received 70 percent of normal allocationsmore than enough to protect crops and orchards against drought damage without dramatic adjustments. Farmers were sufficiently angry about having spent large sums on unnecessary adjustments in response to the bureau's erroneous forecast that they sued the bureau for more than $20 million in compensationa suit that never went to trial.
Glantz discusses several specific problems with the bureau's forecast, including estimation errors in the original prediction (they had failed to include return flow), poor communication of uncertainties, and lack of openness about errors in the forecast. Long-standing institutional water rights arrangements also created a very difficult situation for junior rights holders faced with a drought forecast. Several lessons can be drawn from the Yakima study. The most striking is that responses based on acceptance of erroneous forecasts can have serious economic, distributive, and legal consequences. The case also suggests the need to check forecasts very carefully for errors before releasing them, to clearly communicate uncertainties and the message that forecasts evolve during a season, and to consider how institutional frameworks can redistribute the impacts of a forecast as well as the event.
ENSO-based Forecasts in Northeast Brazil, 1991-1992 and 1996
Droughts sometimes associated with El Niño have often caused serious agricultural losses and human suffering in northeast Brazil, a region where there is widespread poverty and vulnerability to climatic variations. In addition, the cold phase of ENSO, La Niña, is associated with abundant rainfall over the region, sometimes leading to floods that also disrupt the region's economy. Researchers in climate modeling have used the onset of El Niño to forecast drought in the region up to 6 months in advance and, more recently, have learned that droughts in northeastern Brazil are even more strongly correlated with Atlantic sea surface temperature. Therefore, accurate prediction of ENSO and Atlantic sea surface temperature has the potential to improve well-being in the region by providing policy makers with information on anticipated climate variations.
In 1983, when no preparations were made for El Niño, yields of cotton, rice, beans, and corn were less than 50 percent of normal. In the state of Ceara, corn yields fell from 0.54 to 0.12 tons per hectare. The government spent $1.8 billion in short-term relief, which included employing 3 million people in public works to construct irrigation systems and reservoirs and trucking in drinking water. By contrast, the state government of Ceara responded vigorously to a forecast of the 1991-1992 El Niño, which was released by the state's Foundation for Meteorological and Hydrological Resources (FUNCEME) (Golnaraghi and Kaul, 1995). The government instituted several policies, including guiding farmers on what and when to plant (distributing seeds more resistant to water stress and maintaining a strict planting calendar); controlling water consumption in Fortaleza (Ceara's capital city); and rushing the construction of a new dam on the Pacajus River. Policy implementation included the organization of a grassroots campaign in which the governor himself traveled through the state's countryside to vouch for the reliability of FUNCEME's forecast and the benefits that could stem from its application.
One way to estimate the value of the forecast is by comparing agricultural output in 1987 and 1992. During the 1987 El Niño episode, 30 percent less rainfall resulted in output of approximately 15.5 percent of the region's mean output; in 1992, when rainfall was 27 percent below normal, agricultural output in Ceara was approximately 82 percent of the region's mean. These data suggest that the application of a seasonal forecast greatly benefited agricultural output.
However, a recent study of the social implications of seasonal forecasting in northeast Brazil (Lemos et al., 1998) suggests that the picture is much more complex. For example, agricultural subsidies were much more easily available in 1992 than in 1987 and would have boosted agricultural production even in the absence of a seasonal climate forecast. Also, the link between ENSO and regional climate is rather weak, with ENSO accounting for only about 10 percent of the rainfall variation over northeast Brazil (Hastenrath and Heller, 1977). Drought and high rainfall in northeast Brazil may also be associated with other phenomena, such as Atlantic sea surface temperatures and the movement of the intertropical convergence zone. In addition, many small and subsistence farmers have little flexibility in responding to forecasts (Lemos et al., 1998).
The credibility of seasonal forecasts in northeast Brazil was reduced in 1996 when FUNCEME's seasonal forecast of higher than normal rainfall proved inaccurate. As a result, policy makers were very cautious about issuing a forecast of the 1997-1998 El Niño, and there was considerable skepticism among the public. Forecasters delayed issuing a forecast in 1997 and farmers were reluctant to change their strategies; the consequences are not yet fully known. Another cause of resistance to seasonal
forecasts in northeast Brazil is that the prediction of a drought raises a set of unpleasant expectations for many in the region. Past governments typically responded to droughts with large-scale relief efforts that included infrastructure projects and emergency food and work projects and that sent relief funds to certain powerful interests and created a sense of dependency in the population. Many policy makers are concerned about drought forecasts because they do not want, nor can they afford, to perpetuate this drought "industry" (Magalhaes and Magee, 1994).
The case of northeast Brazil provides several lessons about the value of seasonal forecasting in a region where drought can have devastating impacts. It demonstrates the ease with which forecasters can lose their nerve, and the public its trust, as a result of an inaccurate forecast such as occurred in 1996, and the implications for subsequent forecasting efforts. It also shows that some farmers are unable to use seasonal forecasts because they do not have the resources or flexibility to respond. Another important insight is that it is important to include economic and political factors such as subsidies in assessing the effects of a prediction for agriculture, in order not to overestimate forecast value and to consider local history in making assumptions about how a forecast will be received.
The Credibility of Famine Early Warning Systems
Seasonal climate forecast information is also used in famine early warning systems. Since the 1970s, the U.S. government has used climate information to anticipate the onset of famine, to target people at risk, to reduce response time, and to estimate food and other relief requirements, especially in Africa (Walker, 1989; Hutchinson, 1998). The U.S. Agency for International Development has had a warning system for Sub-Saharan Africa since 1981, initially based on information about rainfall, vegetation, and crop yields. The key indicator has been a vegetation index, derived from the AVHRR (Advanced Very High Resolution Radiometer) satellite of the National Oceanic and Atmospheric Administration, which provides information about the progress of the rainy season through monitoring the productivity of natural pasture and large-scale agriculture. Forecasts of seasonal agricultural production are made based on past relationships between early season rainfall and yields. The famine early warning systems can be considered a form of seasonal forecasting because they anticipate conditions up to 6 months in advance, through a combination of qualitative assessment and crop predictions.
By the mid-1980s it was obvious that biophysical information needed to be linked to socioeconomic information in order to provide useful famine warning because famine is created as much by social, economic and political conditions as by drought. Thus, the system now couples a wide
variety of biophysical and satellite measurements with information on health and nutrition, agricultural inputs and markets, and indicators of socioeconomic stress such as livestock and jewelry sales. These indicators are combined into country reports (e.g., for Ethiopia or Mali), which are published and distributed on a regular basis as the growing season progresses and used to plan any relief efforts. Local governments and nongovernmental organizations receive the reports as well as U.S. government and international agencies.
Several lessons can be drawn from the experience with famine early warning systems for the new developments in seasonal forecasting. These include the importance of combining environmental and social information to provide accurate assessments of agricultural production and other social impacts and the value of including local decision makers and nongovernmental organizations in the development and distribution of forecasts.
Indirect Sources of Insight into Responses to Climate Forecasts
Although climate forecasts have been widely available in the United States for more than three decades from government, academic, and private sources, little is known about how they are used. Because of the limited amount of direct knowledge about responses to climate forecasts, a considerable portion of the knowledge relevant to providing people with improved climate forecast information is indirect. Some of this is in the form of general knowledge of how people think about weather and climate; some consists of knowledge about how human beings as individuals and in organizations acquire and process new information generally; some comes from knowledge about how people use information in possibly analogous situations.
Beliefs About Weather and Climate
Until recently, nonspecialists' beliefs about weather, climate, and climate changes and variations have been of interest mainly to academic anthropologists. Research on ethnometeorology, perceptions of weather, and hundreds of other topics in nonwestern societies can be examined through the web site of the Human Relations Area Files at Yale University (http://www.yale.edu/hraf/home.htm). Many traditional societies, including those in ENSO-sensitive areas, have long-standing and complex theories about weather and climate, some of which they use for forecasting deviations from seasonal averages (e.g., Antunez de Mayolo, 1981; Ramnath, 1988; Bharara and Seeland, 1994; Pepin, 1996; Eakin, 1998).
Cultures that are highly dependent on variable climate-ecosystem relationships tend to observe these relationships closely, so their skill in forecasting may have increased over time. Many elements of traditional forecasting methods are in fact explainable by modern scientific principles (Pepin, 1996); however, there has been little if any investigation of how much skill these forecasting systems provide. The persistence of folk theories of climate does not establish their predictive value: some of them, particularly those tied closely to religious rituals, may serve mainly to allay anxiety among people utterly dependent on unpredictable and variable climatic events (Wilken, 1987).
Whatever their level of skill, the existence of traditional climate forecasts has implications for the coping strategies people use and for their acceptance of information from modern climate forecasts (e.g., Oguntoyinbo and Richards, 1978). On the positive side, traditional forecasting indicates the receptivity of certain social groups to the concept of climate forecasting and presumably also their appreciation of the fact that seasonal forecasts are imperfect. In addition, the traditional forecasts probably identify the climatic parameters that are most relevant to their users' subsistence decisions. On the negative side, adherents of traditional forecasting systems may resist new systems, even if they are more skillful, and once modern forecasting systems are adopted, any value the traditional explanatory systems may have for purposes other than climate forecasting (e.g., forecasting crop diseases) may be discredited or lost.
There has been little research in Western societies on beliefs about seasonal-to-interannual climate variability. However, research on beliefs about climate change suggests that people tend to assimilate new information about climate into cognitive structures or mental models that they use for conceptually related mattersother environmental problems affecting the atmosphere. For example, nonspecialists frequently confuse climate change and stratospheric ozone depletion; there is also a widespread belief that "air pollution" (which for many people is associated with phenmena like smog, ozone alerts, and acid rain) is a cause of climate change (Kempton, 1991; Löfstedt, 1992, 1995).
Weber (1997) found a strong effect of mental models on perceptions of climate change and variability among cash-crop farmers in the U.S. Midwest. Their beliefs about climate change had more effect than length of personal experience on their ability to detect recent increases in maximum July average temperatures in their locality and in the variability of those temperatures. Farmers with longer experience were slightly less likely to notice the recent warming, but a much more reliable predictor was whether or not the farmers believed in global warming. The majority of believers in global warming correctly detected and classified the temperature increase, which fit their mental models, whereas the majority of
disbelievers incorrectly remembered no change in maximum July temperatures. The disbelievers were more accurate, however, in detecting increased variability in recent temperatures, probably because they interpreted recent increases in average high temperatures as reflecting variability rather than a trend.
Human Information Processing and Climate Information
Recent advances in cognitive psychology regarding information processing provide insight that can be applied to human beliefs about weather and climate and can put the above findings into a conceptual framework. This research has established that people are not passive recipients of information that they accumulate and store for future reference; rather, they attend to and encode information selectively. Also, people often construct beliefs when needed for a situation, rather than simply recalling them from memory (Payne et al., 1992). Such construction has been shown to be based on mental models of the phenomenon under question, which usually involve causal connections between variables in the mental model but often omit relevant variables and their relationships (Bostrom et al., 1994). Understanding the mental models people might use to assimilate climate forecast information may therefore help with the task of making this information intelligible to the potential users.
A historical example concerning interannual climate variations illustrates how human beings assimilate climatic information into preexisting mental modelsand the shortcomings of this cognitive strategy (from Kupperman, 1982). It also shows that predictions based on preexisting mental models often survive a long series of disconfirming empirical evidence.
English settlers who arrived in North America in the early colonial period operated under the assumption that climate was a function of latitude. Newfoundland, which is south of London, was thus expected to have a moderate climate, and Virginia was expected to have the climate of southern Spain. Despite high death rates due to weather that was consistently much colder than expected, the resulting failure of settlements, and pressure from investors disappointed by the colonies' inability to produce the rich commodities associated with hot climates, colonists clung with persistence to their expectations about the local climate based on latitude. Reluctant to accept the different climatic conditions as a new fact in need of explanation, they instead generated ever more complex rationalizations and alternative explanations for these persistent deviations from their expectations. Samuel de Champlain, for example, took a single mild winter in 1610 as indication that his mild climate expectations
were justified after all, suggesting that the severe winters he had experienced during each of the six preceding years must have been what would nowadays be called statistical outliers.
This example and the research conclusions it illustrates suggest some hypotheses about how people may respond to climate forecasts. Such forecasts tell people that their expectations about the future climate, whether based on traditional forecasting methods or on the historical average of past conditions, should be revised. They thus provide people with new informationoften, information inconsistent with their current beliefs. The new information will probably be understood better and accepted more fully if recipients can interpret it within a causal model of climate change or variability that they understand and with which they agree. This conclusion in turn suggests that, to encourage use of information from ENSO-based forecasts, users should first be educated about the ENSO mechanism and how it affects local climate, and those who deliver information should learn about how their audiences think about climate. The research also suggests, however, that nonexperts who learn to use a mental model of ENSO may treat ENSO-based forecasts as having more certainty attached to them than the scientific evidence warrants. It will probably also be important for nonexperts to adequately assimilate the difficult concept that forecasts are probabilistic and uncertain (see below).
Climate forecasts are based on covariations among variables (in a simple example, tropical sea surface temperature and precipitation in southern California). The cognitive literature on conditions that facilitate people's detection and understanding of covariation should help in the design of educational materials that could accompany climate forecasts. For a wide range of species, including humans, associative learning of the relationship between two variables occurs only to the extent that the relationship is necessary to predict the outcome variable (e.g., Rescorla and Wagner, 1972). The surprise of an unpredicted outcome is a great motivator for learning to occur. Thus, it may be expected that learning about the relationship of ENSO to local climate will be best accomplished soon after extreme climatic events, such as those associated with the 1997-1998 El Niño.
The cognitive literature suggests that individual's mental models and prior expectations, especially if they make for a plausible, causal story, strongly influence the ways they perceive and interpret events, as illustrated in the historical example given above. People tend to focus on observations that conform to their beliefs (Mynatt et al., 1977; Wason, 1960); to under- or overestimate actual statistical relationships depending on their prior expectations (Nisbett and Ross, 1980), even perceiving and reporting covariation according to their expectations in sets of data in which no covariation exists (Chapman and Chapman, 1967); and to seek
out less situational information if they hold strong preconceptions about a given relationship (Alloy and Tabachnik, 1984). People also encode and retrieve correlated information much more effectively when the correlations can be explained on the basis of prior expectations (Bower and Masling, 1978). Although objective evidence of covariation clearly plays some role in its detection (Wright and Murphy, 1984), the evidence is overwhelming that perceptions of relationships are dominated by prior expectations. These findings suggest that, if people learn a mental model that includes relevant predictor and outcome variables and facilitates correct expectations about relationships, they will find it easier to understand the information in climate forecasts, will be more likely to trust and use the forecasts, and will update their mental models more appropriately on the basis of observations and climatic information.
Two key attributes of climate forecast information that are likely to affect how individuals interpret it are that forecasts are probabilistic and uncertain. Probabilistic information is difficult to assimilate because people do not naturally think probabilistically (Gigerenzer and Hoffrage, 1995). Evolution may have favored development of the ability to count events but not the ability to divide those counts by population totals (and thus to estimate probabilities). Consequently, people do not estimate probabilities well (Kahneman and Tversky, 1972; Bar-Hillel, 1980). For example, a sample of Nigerian farmers could supply qualitative information about weather (they had elaborate teleological explanations about the causes of rainfall and drought that involved God and punishment for breaking religious codes) but not quantitative or probabilistic information (Oguntoyinbo and Richards, 1978). In making medical diagnoses, doctors typically reason deterministically unless they are taught probabilistic reasoning in courses on medical decision making (Elstein et al., 1990). When information is provided to the public in a probabilistic way, as for example in weather forecasts, probability is usually described with verbal expressions (e.g., slight chance, almost certain)a format recipients prefer (Wallsten et al., 1986; Weber, 1994). Although words are considerably less precise than numbers, it is possible that they offer reasonably good quality information when the probabilities themselves are imprecise.
Uncertainty in climate forecasts, due to poor input information, imperfect climate models, and the inherent unpredictability of many situations, means that forecasts carry the risk of being wrong. Thus, the research literature on decision making under risk and uncertainty is relevant. Uncertainty about the precise likelihood of events and of outcomes associated with those events has been shown to be aversive to decision makers, in the sense that people will avoid such decision options or are willing to pay a premium to reduce such uncertainty or ambiguity (Ellsberg, 1961). Greater precision in predictions seems to lead to greater
comfort with the decision, as does greater perceived personal competence in the decision domain (Heath and Tversky, 1991). Although much is known about people's preferences when they are given a choice between an option in which the probability of events is precisely specified and one in which the probability is ambiguous, vague, or uncertain, we know little about how choice is affected when all risky options are described either with great uncertainty or with great precision. Understanding how people make such choices is relevant to designing methods of conveying the information in climate forecasts, which will have uncertainty attached.
The fact that some forecasts will inevitably be wrong raises questions about how people will interpret forecast information after a forecast failure. Research evidence offers some suggestions. One is that erroneous forecasts may destroy trust in the organizations that provide the forecasts. Research by Slovic and collaborators shows that trust in institutions and in information sources affects people's perceptions of technological and environmental risks and that such trust is easier to destroy than to build (Slovic, 1993; Slovic et al., 1991). This finding suggests an asymmetry in the consequences of forecasts that may arise after they generate false alarms or, perhaps, false reassurances. Overconfident predictions and forecasts not borne out by actual events are likely to have an especially strong influence on the future use of forecast information, perhaps doing damage to the acceptance of forecasts that future, carefully qualified forecasts cannot repair.
The cognitive literature may offer further guidance on how to frame forecast information and recommend actions based on the forecasts. For example, in a wide range of situations, the disutility that people experience at the loss of some commodity (e.g., a crop) is far greater than the utility experienced when the same commodity is gained (e.g., Kahneman et al., 1991). Also, people feel worse when they experience a bad outcome as the result of an action (e.g., a crop failure because they changed to a different variety of seed corn in response to a climate forecast) than when it results from inaction (e.g., a crop failure because they did not change their seed corn) (Baron and Ritov, 1994). Weber (1994) has summarized the effects of such phenomena on perceptions and beliefs about uncertain events. The precise implications of these general human tendencies in interpreting outcomes for the delivery of climate forecasting information need to be determined by further research.
Another general human tendency with relevance to the use of climate forecasts is to stop the search for responses to a situation once a single solution has been found, even if taking other, additional actions would provide additional benefits. For example, Weber (1997) has documented that, even though American farmers can cope with climate change and variability in many disparate ways (e.g., use of futures markets, changes
in production practices, lobbying for government action), few farmers, even if they believe in the need for adaptation, employ more than one class of adaptive responses, even when other, potentially complementary, strategies are available. Other decision makers, including experts, have similar tendencies. For instance, radiologists often halt their search for abnormalities in radiographs after finding one lesion, leaving additional lesions undetected (Berman et al., 1991). A single solution seems to provide sufficient assurance that a problem has been dealt with, and the resulting peace of mind seems to prevent the generation of additional solutions or adaptations. These results suggest the potential value of providing the users of climate forecasts with checklists or other external aids that identify a full complement of interventions that would allow them to benefit from the forecasts.
Most of the systematized knowledge about how individuals form and change their beliefs about the environment is based on studies with respondents in the United States. However, culture influences a wide range of psychological processes, including some that are likely to affect the use of forecast information. For example, Yates and collaborators (1989, 1996) reported cultural differences in the use and interpretation of the probability scale, and Weber and Hsee (1998) have documented cultural differences in risk perception and risk preference. There is some evidence of cultural differences in tolerance of ambiguity (Hofstede, 1980), although little about cultural differences in people's preference for precision versus ambiguity. Thus, there are reasons to expect that culture may influence the ways in which people interpret, understand, or use climate forecasts. There is little basis in theory or data, however, for predicting the magnitude of these effects or for characterizing them. Cross-cultural comparisons of perceptions and beliefs about climate variability and climate forecasts would change this situation. The kind of research that would be desirable is exemplified by the work of White (1974), which provides more than 20 case studies of responses to natural hazards in a variety of countries, each conducted by local investigators using a common methodology and assessment protocol.
Organizational Responses to New Information
Climate forecasts present organizations with the challenge of processing and acting on new information. However, many organizations that could benefit from climate forecast information have not established routines or responsibilities for processing this kind of information. It is difficult to anticipate how well organizations will assimilate and use the new and uncertain information in climate forecasts. Much of the research on organizations suggests that they react to information and set priorities in
relatively haphazard ways, as suggested by the ''garbage can'' model of organizational behavior (March and Olsen, 1986). In addition, motivational obstacles such as defensiveness often prevent professionals in organizations from learning from and adapting to new information (Argyris, 1991), and phenomena of group dynamics, such as "groupthink" (Janis, 1972), sometimes prevent groups in organizations from taking advantage of the heterogeneous perspectives from which their members view new information. The experience of the insurance industry before and after Hurricane Andrew shows that organizations may fail to respond appropriately to uncertainty and new information. The surprise and upheaval occasioned by the magnitude of losses from Andrew suggest that firms learned by getting hurt rather than by developing and using better predictive models. In recent years, however, many firms have been changing their organizational structures to provide for more timely information flow in light of the need for speedy responses to new information, e.g., matrix structures, joint ventures, strategic alliances, etc. (Nadler et al., 1992). Such efforts may help some organizations manage the new kinds of information climate forecasts provide.
Organizational responses, like those of individuals, depend on specifics about the information recipient and its context. Thus, generalizations may not be helpful. For example, there is evidence that different institutional arrangements are differentially effective as risk reduction strategies, depending on the type and timing of information (Frey and Eichenberger, 1989). In some situations, markets provide efficient solutions; in others, they do not.
The impact of new information seems to depend on prior knowledge and expectations in the receiving organizations. For example, new information has far greater impact on financial markets when there is greater preexisting uncertainty about expected profitability and future cash flows (Brous and Kini, 1992). Market reactions to new information also depend critically on expectations of the target company's financial performance prior to the disclosure of the new information (e.g., Kasznik and Lev, 1995; Datta and Dhillon, 1993).
Research on responses to new information by organizations responsible for disaster warning are of particular relevance for climate forecasts because these organizations sometimes provide warnings of climatic events. The performance of organizations responsible for detecting disasters and managing disaster-related information depends on how well they cope with a variety of challenges. They must reach an agreed interpretation of the available information despite disagreements among individuals; specify tasks, roles, responsibilities, and relationships between tasks; develop appropriate lines of communication; ensure that individuals know their roles; and provide sufficient resources to act. Many of
these challenges are repeated at the inter-organizational level, where organizations have tasks, roles, and so forth. Organizations and the entire system must also meet the challenge of maintaining the vigilance and flexibility needed to respond quickly to new and surprising developments. Organizations that respond well to information about disasters tend to have had considerable time and experience addressing these challenges. To the extent that delivering climate forecast information requires the involvement of new organizations or requires existing organizations to behave differently, a period of learning is likely to be required before effective response can be counted on. Practice has been an effective way of facilitating learning with disaster warning organizations (Drabek, 1986; Mileti and Sorenson, 1997).
Insights from Analogous Types of Information Transmission
Much can be learned about the way individuals and organizations respond to climate forecasts by studying analogssituations that strongly parallel the use of climate forecasts. According to Jamieson (1988), the instructive potential of an analog lies not in how closely the form of the analog physically resembles the intended target, but rather in how closely the characteristic processes regulating the form of the analog resemble the intended target. Thus, if the essential characteristics can be defined for the situations in which climate forecast information is received, people can be expected to respond to climate forecasts the way they respond in other situations sharing those characteristics.
Climate forecast information has at least the following distinctive characteristics: (1) it is intended not only to inform, but to benefit the recipients; (2) it is based on scientific techniques few of the recipients can understand; (3) it provides generalized forecasts of events several months in the future; (4) the forecasts are probabilistic rather than deterministic; (5) the probabilities offered are themselves uncertain; (6) the information is of a kind that may not fit easily into the recipient's mental models; (7) the credibility of those offering the predictions is hard to determinethey have a very limited track record; and (8) the relevance of the predictions for the recipients' decisions may not be obvious unless someone interprets the forecast in light of the recipient's information needs.
Which other situations of information transmission are analogous to these? Several situations that share many of the above characteristics have been the focus of considerable research on the conditions for effective information transmission. In each of these situations, scientifically based information is offered to individuals in the belief that they can improve their well-being by acting on it and with the intent to influence them to act accordingly.
One such analog is in the public health field, in which information has been used in efforts to promote numerous kinds of healthy behaviors, including cessation of cigarette smoking, change in diet to reduce fat and add fiber, and reduction of behaviors that increase the risk of infection by the human immunodeficiency virus (e.g., Green, 1984; Green et al., 1986; Becker and Rosenstock, 1989; Green and Kreuter, 1990; Aggleton et al., 1994). Another analog exists in energy and environmental policy, in which information has been an important element of efforts to promote energy conservation, recycling, and other so-called proenvironmental behaviors by individuals and households and hundreds of empirical studies have been examined to learn their lessons (e.g., National Research Council, 1984; Katzev and Johnson, 1987; Lutzenhiser, 1993; Gardner and Stern, 1996).
A third analog is in the area of disaster warning, in which information is used, for example, to induce people to construct tornado shelters, raise levees, and protect their lives and property from oncoming storms (Mileti and Sorenson, 1987, 1990; Mileti et al., 1992). Contemporary disaster warning systems based on improved capabilities in prediction and forecasting have dramatically reduced the loss of life and injuries from all hazards in the United States, including climatic hazards. A fourth analog, commonly called risk communication, involves the design and distribution of messages about public health, safety, and environmental hazards that are designed to generate levels of concern and behavior change considered appropriate by those designing the messages (e.g., National Research Council, 1989). Because of certain issues raised by risk communication research, we return to this topic only at the end of this section.
The "green revolution" in agriculture, which developed knowledge and technology as well as spreading information, attempted to induce farmers to adopt new seeds and cultural practices in order to dramatically increase grain production. It shares some of the distinctive features of climate forecasting and is particularly interesting because it induced farmers to do things they may also do in response to climate forecasts. The experience of the green revolution may therefore also yield hypotheses worth systematic examination in the context of climate forecasting. This experience is summarized in Box 4-1.
General Principles for Designing Information Programs
Each of these analogs shares most of the distinctive characteristics of climate forecasts information listed above. In each field, there have been numerous studies of the effectiveness of information and of the systems for delivering it, and reliable concepts and methods have been developed for conducting studies to assess how scientific information is interpreted
and used. Although the behaviors to be changed are very different in each case, there is notable consistency and complementarity among the major lessons researchers have drawn from efforts to use scientific information to change behavior. Here are some of the lessons, stated as general principles:
Alternative Models for Designing Information Programs
The above general principles have coalesced into somewhat different approaches to disseminating information in different fields. In the health promotion field, a community-based approach to health has developed that pursues what a study by the Institute of Medicine has called a "willing compliance" model of change, as distinguished from an "authoritarian" model (Institute of Medicine, 1997:68; see also Evans and Stoddart,
1994; Patrick and Wickizer, 1995). This approach seeks to develop community coalitions that involve the affected groups, to use these coalitions to identify health priorities, and to maintain these coalitions throughout the process of implementing community-based efforts to bring about change. Similarly, in the field of proenvironmental behavior, there has been increasing recognition that community-based programs that employ a variety of behavioral change strategies, including participatory approaches, are among the most promising strategies available (Gardner and Stern, 1996). In the field of disaster warning, however, a more authoritarian model of persuasive communication, relying on scientists to gather information and government agencies and private-sector organizations to disseminate it, has proven useful.
The divergence in the lessons drawn from research in apparently analogous fields may be reconciled by examining the various purposes and contexts of information programs. A National Research Council (1989) study that examined risk communication about a broad range of health, safety, and environmental hazards distinguished two distinct and sometimes conflicting purposes for providing information about hazards: to inform audiences and to influence them. The study pointed out that the appropriateness (especially for government) of using messages to influence people to respond in particular ways to hazards is judged quite differently depending on which behaviors are being promoted, the degree of scientific consensus about the information being delivered, the compatibility of government influence with individual autonomy and related values, and the influence techniques employed in designing the message (National Research Council, 1989:80-93).
In this light, the differential emphasis on "participatory" and "authoritarian" models of communication in different fields most likely reflects the content and history of the fields. In certain environmental risk areas, such as management of potentially carcinogenic chemicals and radioactive materials, a high degree of uncertainty and controversy has surrounded scientific information; it is not always evident which actions are most appropriate for reducing risk; and there is a long history of accusations, sometimes substantiated, that corporations and government agencies have misinformed the public. Highly participatory approaches are necessary in these areas to allow for discussions of how to proceed under uncertainty and to address the problem of mistrust of official sources of information. In other areas, such as earthquake engineering, commercial aviation, hurricane warning, and infectious disease, the historical legacy of risk communication includes greater consensus on how to reduce risks, a much higher level of trust in expert sources of information, and a greater willingness to accept authoritarian styles of communication.
A more recent National Research Council study reached similar conclusions. It found that developing a useful understanding of risks depends ''on incorporating the perspectives of the interested and affected parties from the earliest phases of the effort to understand the risks'' in order to meet "the challenges of asking the right questions, making the appropriate assumptions, and finding the right ways to summarize information" (National Research Council, 1996b:3). This study proposed a participatory strategy for developing most kinds of environmental information, recognizing, however, that the most effective degree and type of participation is situation specific.
What kind of model is most appropriate for delivering climate forecast information? Without much of a base in empirical knowledge, it is necessary to hypothesize on the basis of analogy. In our judgment, the context of climate forecast information at its current stage of development is more similar to that of information about health promotion, energy conservation, and hazardous substances than to that of short-term disaster warnings. There is too much uncertainty and potential controversy about what the available scientific information implies for human response to use an authoritarian approach aimed at influencing people. Even an authoritarian style of informing people seems inappropriate because of the large gaps in knowledge about which information would be decision relevant for which recipients. Consequently, we believe much can be gained by using participatory approaches that rely heavily on the involvement of communities of potential forecast users both for developing climate forecast information and for designing information delivery systems. Such approaches are likely to provide climate scientists with useful and timely information about the attributes of forecasts that will make them useful for the intended recipients, to build understanding among the recipients of what forecasts can and cannot do, and to develop an appropriate level of trust in forecast information.
The evidence from analogs suggests that in the future, if and when the accuracy and importance of climate forecasts is convincingly demonstrated to users and forecasts are prepared and presented in ways that meet users' information needs, something more like the disaster warning model of information delivery may prove effective. However, inappropriately high levels of expressed confidence in forecasts, acrimonious controversy about forecasting models, and forecasts that deliver information recipients perceive as irrelevant are all likely to delay the coming of such a future. These judgments, of course, are preliminary and should be tested by empirical research.
An example from Mexico illustrates one way a participatory approach to developing forecast information might proceed. Dialogues between climate forecasters and forecast users have led to an effort to seek a com-
promise between climatological research agendas and the needs of users. As a result, Mexican climatologists have begun to work on correlating the frequency of "black" frosts with El Niño years despite the extreme difficulty of predicting frost hazards. They have done this largely because consultations with farmers have shown that frost is one of the greatest risks they face.
Some scientists fear that early consultations with forecast users may unduly raise their expectations. It may take several years for scientists to produce predictions with the necessary detail and accuracy to be useful to a particular sector and, in the meantime, users may lose confidence or interest in the process or use unreliable information to make costly decisions. Explaining the limitations and challenges of the predictive research may be critical to maintaining user confidence. For example, it may be impossible to predict a midseason dry spell or the date of a significant frost. Explaining why the science is more uncertain about certain topics than others and conveying uncertainty in terms understandable to specific user groups can help recipients make more appropriate use of available information and participate constructively in forecast development. There is a lack of systematic knowledge at this time concerning how to convey the state of prediction science to particular types of forecast users in a helpful way.
The limited research on responses to actual climate forecasts and larger bodies of knowledge on information use generally and in partially analogous situations have yielded some promising findings and hypotheses, as well as developing a set of methods for assessing the ways scientific information is used and the ways information delivery systems function. Although the general findings need further validation as applied to climate forecast information, they suggest ways to go about organizing and distributing such information so it can be used effectively within social coping systems. They also suggest directions for research on how to make climate forecasts more valuable.
In sum, climate forecasts are useful only in relation to the actions people can take, given forecast information, to improve their outcomes. Many factors specific to forecasts and to the recipients' decision situations affect the potential usefulness of forecast information. To improve the usefulness of climate forecasts, it is important to identify the decision-relevant attributes of forecast information for particular activities and actors and to encourage forecasters to provide information with those attributes when possible. It is also important to consider what the recipients of climate forecasts are likely to do in practice, given the coping strategies they actually use, their ability to modify those strategies in response to forecast information, the normal routines of their activities, their usual practice in dealing with new information that is offered to them as helpful, their level of trust in the forecast and its source, and other realities of their situations. Available evidence suggests that the benefits from improved information typically go disproportionately to the wealthy and better educated. Closing the gap between the potential value of climate forecast information and its actual value will depend on developing focused knowledge about which forecast information is potentially useful for which recipients, about how these recipients process the information, and about the characteristics of effective information delivery systems and messages for meeting the needs of particular types of recipients. It may also depend on improved understanding of how to design information systems that effectively reach marginalized and vulnerable groups.