A hazard map is a common device for providing people with information about the environmental hazards to which they are exposed. Hazard maps differ in the temporal contexts that they address—chronic hazards that will happen sometime in the future vs. acute hazards that are imminent threats. Maps of chronic hazards might be used by specialists such as land use planners, structural engineers, and emergency managers to identify geographic areas that can expect an event of a specified probability (e.g., the area inundated by a 100-year flood) or of a specific magnitude (e.g., the area expected to be affected by a Category 5 hurricane storm surge). They can use these maps to guide community hazard mitigation and emergency preparedness actions that are implemented long before an acute threat arises. In addition, authorities can use these maps to promote household emergency management actions.

Maps of acute hazards can be used by authorities to identify protective action recommendations (PARs) for the risk area population. That is, these maps can be used to guide decisions about who is at risk, what protective action the risk area population should take, and when they should implement those protective actions (Cova et al., 2017). For example, coastal emergency managers use the hazard maps in HURREVAC to guide their decisions about PARs when their jurisdictions are threatened by approaching hurricanes. In turn, authorities can use acute hazard maps to communicate hazard warnings that promote appropriate household protective actions. The objective of these warnings is not only to increase compliance with PARs, but also to decrease shadow evacuation so those who are not at risk do not impede the evacuation of those who are at risk (Yin et al., 2016; Lindell et al., 2019a).

When constructing hazard maps, one might think that hazard analysts only need to draw lines in the right places on a base map and give those maps to risk area populations in order to increase their risk perceptions and, thus, their emergency preparedness for chronic threats and protective actions for acute threats. Unfortunately, that is not the case because studies dating back at least to Handmer (1980) have documented people's misunderstanding of hazard maps. Consequently, risk communicators ranging from warning meteorologists to local emergency managers need to understand the limits of people's comprehension of different types of hazard maps.

At the outset, it is important to understand that hazard maps are one of three means of communicating hazard information—spatial, numeric, and textual. Moreover, the effects of this hazard information can be understood in terms of the Protective Action Decision Model (Lindell and Perry, 2004, 2012; Lindell, 2018), which posits that a hazard map is a form of warning message that is primarily directed toward the warning receiver's perceptions of the threat. Once a hazard map elicits a high threat perception, the receiver is motivated to search for additional information (if the information source is not highly credible, the threat is ambiguous, or the logistics of response are uncertain) and to initiate protective action. Thus, map elements that promote these processes will be more effective in eliciting appropriate protective actions. The following sections will elaborate on these ideas by discussing indicators of map comprehension, cognitive processes in map comprehension, and cognitive abilities affecting map comprehension. The article will then review of research on people's comprehension of, and behavioral response to, different types of hazard maps, and conclude with a summary of research findings and recommendations for future research.

One obvious reason why the participants in the Wu et al. (2015b) study issued tardy PARs is that, even though the hurricane tracking displays provided information about where to initiate PARs, they lacked any graphical guides for deciding when to initiate them. As noted earlier, this issue was addressed by Cova et al. (2017), who suggested that emergency managers establish warning triggers to define the times that PARs should be initiated. Warning triggers address three primary questions: which locations or population segments are at risk, what is the best PAR for each location or population segment, and when should the PARs be initiated? The first question is fundamentally geographic and involves careful identification of population segments that are likely to be adversely affected (Sorensen et al., 1992; Hsu and Peeta, 2014). The second question involves selecting the best available PARs for different target groups (e.g., evacuate or shelter-in-place). The third question requires assessing the amount of time that target groups will need to complete their PARs before hazard impact (Sorensen, 1991; Lindell and Prater, 2007).

The first step, identification of locations at risk is illustrated in Figure 1, which shows a hurricane approaching a coastal jurisdiction. The figure's right-hand side has a small circle depicting the hurricane eye, which is surrounded by a larger circle representing the radius of Tropical Storm force wind. The hurricane track is pointed toward a coastal Risk Area, which is defined by the hurricane center's forecast point of landfall, the distance to either side which is defined by the radius of hurricane force wind that will cause wind and surge damage, and the inland distance that there will be wind and surge damage, which is defined by hurricane intensity. There are Caution Areas adjacent to the Risk Area along the coast because of the uncertainty about the storm track and storm size, as well as inland from the Risk Area because of the uncertainty about the storm intensity.

The second step, identifying the appropriate PAR for each location, involves choosing among evacuation, shelter in-place, and allowing people to continue normal activities. Evacuation is expensive and disruptive so it should be recommended only when other actions provide insufficient protection. In turn, the level of protection depends upon three fundamental criteria—time, distance, and shielding. Evacuation addresses the first two criteria by reducing the amount of time that people are exposed to a hazard and it increases distance from the hazard source. By contrast, sheltering in-place provides shielding from a hazard by reducing the infiltration of hazardous materials into a structure (e.g., volcanic ash) or by resisting dangerous temperatures (e.g., wildfires) or pressures (e.g., tornadic wind). Evacuation is common for hurricanes striking the United States because this hazard provides sufficient forewarning to clear a risk area before hazard impact and few single family homes in coastal areas provide adequate protection from wind and surge. However, tornadoes provide insufficient time to evacuate safely, so authorities typically recommend shelter in-place (see Lindell et al., 2019a, Chapter 3, for further discussion).

The third step, determination of when to issue PARs, is based on the principle that protective actions should be completed before the arrival of hazardous conditions. As Figure 2 indicates, hurricane tracking involves an initial detection stage, after which the storm continues to be monitored until it makes landfall. As the storm progresses, meteorologists make projections about landfall location, wind speed, storm surge, and rainfall that, in turn, serve as decision information for emergency managers. These emergency managers use the decision information about expected storm impacts to establish the warning triggers that address which locations or population segments are at risk, what is the best PAR for each location or population segment, and when the PARs should be initiated.

A graphical guide for deciding when to initiate PARs can be constructed by recognizing that the time required for a resident or transient household to clear the risk area after incident initiation can be defined as a function of four Evacuation Time Estimate (ETE) components,

where t_T is a household's total clearance time, t_d is the authorities' warning decision time, t_w is the household's warning receipt time, t_p is the household's evacuation preparation time, and t_e is the household's evacuation travel time (Lindell et al., 2019a, section 3.2). As Cova et al. (2017) noted, authorities who have a predefined trigger for initiating an evacuation can eliminate t_d, leaving household warning receipt time, household evacuation preparation time, and household evacuation travel time. Warning receipt time and evacuation preparation time are unique values for each household but can be aggregated into distributions that show the percentage of households over time who have received a warning or completed evacuation preparations. In turn, the ETE can be converted to an evacuation decision arc by recognizing that the critical distance of the hazard from a reference point (e.g., the nearest location of the threatened population) equals rate (the forward speed of the hurricane) multiplied by time (the ETE). The left side of Figure 3 shows the hurricane Risk Area from Figure 1 with three arcs. A conventional evacuation decision arc is defined by the most likely values of the ETE for that Risk Area and the current value of the hurricane's forward speed (the speed at which the hurricane center is moving over the water). However, both the Risk Area's ETE and the hurricane's forward speed are quantities whose uncertainty can be represented by a minimum probable radius, most probable radius, and maximum probable radius for the evacuation decision arc. Using this display, authorities can trigger an evacuation once the hazard reaches the evacuation decision arc (Lindell et al., 2019a, section 3.3.2). In the case of a hurricane, evacuation should begin when the radius of Tropical Storm force wind touches the evacuation decision arc that local authorities have selected. Tropical Storm force wind is a common evacuation trigger because this is the wind speed at which an evacuation could be stalled if high profile vehicles such as buses and recreational vehicles are overturned by high wind striking them from the side.

There is strong empirical support for some conclusions about people's comprehension of hazard maps. First, people's risk judgments can be characterized by a proximity heuristic (Teigen, 2005), which generates a perceived risk gradient that is a monotone decreasing function of distance from a hazard source (Lindell and Earle, 1983). However, there are differences among hazards regarding the application of this heuristic because there are corresponding differences among hazards regarding the location of the hazard source with respect to contour lines drawn on a hazard map. For wildfires, the hazard source is at the fire perimeter and the principal risk area is downwind of the burn area. For hurricanes, the hazard source is within the radius of hurricane wind at the apex of the uncertainty cone and the principal risk area is along the forecast storm track. Finally, the hazard source for tornadoes is located inside the warning polygon that defines the risk area. Consequently, tornado polygons produce a centroid effect that is not found for wildfires or hurricanes. By contrast, responses to the latter two types of hazard displays exhibit a transect effect, which is a special case of the proximity effect. Instead of proximity to the current location of the hazard source, the transect effect is an outcome of proximity to the future location of the hazard source.

The transect effect found in laboratory experiments is consistent with the results of survey research showing that the rate of sheltering in-place in response to a tornado warning was more strongly correlated with proximity to the forecast track than being in the warning polygon (Nagele and Trainor, 2012). Similarly, Miran et al. (2018) found an effect of proximity to the forecast track on protective action during the May 2013 Oklahoma City tornadoes. Finally, Krocak et al. (2020) found that the probability of a false-positive tornado warning perception as a function of distance from the nearest tornado warning followed an exponential decay function in which the probability declined from approximately p = 1.0 within the warning polygon to approximately p = 0.45 at 100 miles.

The transect effect is so strong that it substantially weakens the edge effect that the NWS intends for tornado polygons (which is different from the hurricane uncertainty cone that is only a 67% confidence interval). The weak edge effect found in tornado and hurricane display experiments is consistent with findings from the many field studies that have reported evacuation shadow—people's evacuation from areas outside the evacuation zones that authorities have officially designated (see Lindell et al., 2019a, Chapter 4). In the case of the tornado polygon, the conflict between the NWS's intended response (there is no need for protective action outside the polygon) and people's typical response is quite likely a result of cascading conservatism that arises from the absence of a clear verbal or numeric definition of the meteorologist's subjective probability of a tornado strike outside the polygon's edges (Sorensen and Mileti, 1987). Since viewers do not know whether the tornado polygon represents a 50, 75, or 99.99% confidence interval, they are more likely to treat it as the former than the latter.

Display effects are difficult to summarize succinctly because different displays produce distinctive types of inferential errors. For example, it is now known that few people are susceptible to error of thinking that a hurricane cannot possibly track outside the uncertainty cone as Broad et al. (2007) warned. However, there does seem to be a significant proportion of viewers who mistake the uncertainty cone's increasing cross-section for an increase in hurricane size (Ruginski et al., 2016; Padilla et al., 2017). This is a potentially important finding, but it is unclear what implications, if any, this has for viewers' protective action decision making. One possibility is that perception of an increasing hurricane size would increase spontaneous evacuation by viewers in areas on either side of a designated evacuation zone (e.g., the Caution Areas on either side of the Risk Area in Figure 1. This possibility should be tested in future research that asks people to report their expected protective actions. Of course, there may be some discrepancies between people's expected protective actions and those that they take in an actual event because, as the PADM indicates, protective actions are not determined by risk perceptions alone; such actions are also affected by perceptions other stakeholders and of the alternative protective actions (Lindell and Perry, 1992, 2004, 2012; Lindell, 2018). Indeed, as Maidl and Buchecker (2015) found, simply distributing static/non-interactive hazard maps to a risk area population has only small impacts on hazard awareness, discussion with others, and expectations of preparedness—let alone actual preparedness. This is undoubtedly a result of the many variables other than hazard map content that influence people's protective actions.

Another issue needing further research is the characterization of map comprehension components. MacPherson-Krutsky et al. (2020) characterized these as basic and advanced skills, but Cao et al. (2016) characterized map effectiveness dimensions as accuracy of understanding, risk perception, response time, and preference and ease of understanding. There is also a need for further research on interactive maps because an increasing number of features, such as the ability to display or suppress different data layers, increases the complexity of hazard map operation. Thus, research will be needed that can identify the most readily understandable procedures for conveying to novice users how to select and deselect different map features (Cao et al., 2017; MacPherson-Krutsky et al., 2020).

There is also a need to examine the degree to which the magnitude of transect effects in both tornado and hurricane studies are influenced by the range of points over which viewers are asked to judge hazard impacts affects the size of the transect effect. Specifically, Wu et al. (2014) obtained a much larger transect effect than Ruginski et al. (2016) almost certainly because Wu and his colleagues asked for p_s judgments in a 360 degree circle whereas Ruginski and his colleagues asked for expected damage judgments along a 650 km/400 mi line (see also Boone et al., 2018; Liu L. et al., 2019). This suggests that the range of response options offered to novice map viewers acts as a subtle cue to the range of hazard impacts. A related issue concerns the degree to which the findings from hazard map experiments generalize to risk area residents because most tornado and hurricane displays have obtained p_s and damage judgments at experimenter-selected locations on a map. However, many risk area residents cannot identify their locations on a map (Zhang et al., 2004; Arlikatti et al., 2006), so there is a need for hazard map displays to identify the location of the hazard and the viewer (Liu B. F. et al., 2017).

Future research should also examine the possibility that training can correct misconceptions about hazard map displays by identifying commonly held erroneous beliefs and seeking to correct them (e.g., Whitney et al., 2004). In particular, it is important to determine if training can suppress the centroid effect in p_s judgments about tornado polygons. For example, researchers might try to see if warning people about the centroid effect during simulated TV broadcasts of tornado polygons would accomplish this aim. However, such warnings might have limited effect because Boone et al. (2018) reported that training had limited success in correcting the misconception that the expanding uncertainty cone indicates increasing hurricane size. Research on hazard map training should also examine viewers' metacognition because the accuracy of people's insight into their map comprehension skills is likely to affect their training motivation (Gully and Chen, 2010).

Another direction of future research is to expand the types of displays that are studied. Hurricane display research conducted to date has emphasized alternatives or supplements to the forecast track, such as the uncertainty cone and the ensemble display. However, there has been no research on the forecast wind swath, which is defined by the locations that could be subjected to hurricane force wind rather the locations through which the hurricane center could pass (which is the definition of the uncertainty cone). Because the wind swath provides a better representation of the locations at risk, it is likely to elicit better compliance with evacuation PARs (Mileti and Sorensen, 1990; Lindell, 2018). Research on hurricane displays might also follow the lead of tornado polygon studies in examining the effect of presenting hurricane radar images, which provide graphic displays of storm size and intensity. This information could have the positive effect of increasing evacuation expectations in risk areas advised to evacuate, but might also have the negative effect of increasing shadow evacuation from inland areas that are not advised to evacuate.

Another suggestion derived from tornado polygon research would be to see if providing impact-based warnings to hurricane risk areas would increase compliance with evacuation PARs in the same way that these warnings increase shelter in-place for tornadoes (Casteel, 2016, 2018). Such studies will need to carefully examine different verbal, numeric, and visual ways of describing impacts because the wording of a Hurricane Ike warning predicting “certain death” for those who failed to evacuation seems to have had no effect on evacuation rates (Wei et al., 2014).

The Cao et al. (2017) display of expected fire front arrival times suggests that displaying Tropical Storm force wind arrival times at different locations might also be useful. This information would be consistent with warning researchers' recommendations to provide information about the expected time of impact so viewers know when to begin evacuation preparations (Mileti and Sorensen, 1990; Lindell, 2018). In addition, hurricane and wildfires displays could be improved by adding evacuation decision arcs that inform viewers when they should leave their homes to ensure that they reach safety before hazardous conditions arrive (Lindell and Prater, 2007; Cova et al., 2017). Such displays could overcome the deficiencies in PAR selection observed in Wu et al. (2015b).

Cao et al. (2017) advocated providing dynamic/interactive wildfire hazard maps viewers with interactive controls so viewers can select the elements that are most important to them. For example, they recommended additional spatial information (e.g., fire location, containment status, current and forecast wind conditions, and fire spread prediction) and textual information (e.g., well-known landmarks, names of jurisdiction subject to PARs, roads that have been closed, and names and addresses of community shelters). However, this calls attention to the need for research into intuitive interface design that relies on viewers' immediate understanding of features such as locating their home on the map, manipulating map layers, accessing non-spatial information, and adjusting the map scale.

In addition, researchers should identify ways to assist risk area residents in focusing their attention on the display elements that the experts judge to be most important. For example, researchers might follow the example of Drews et al. (2014, 2015) in asking expert Incident Commanders to identify the critical variables that should receive the most attention when selecting PARs. One possibility might be to design interfaces so the most important variables are default options. This would allow novice viewers to use the displays with minimal or no training, yet allow more advanced viewers to access additional information. In addition, novice decision makers could be provided with simplified instruction manuals and train on multiple scenarios until they reach proficiency. Unfortunately, this type of training is not feasible for risk area residents and even less feasible for tourists who are visiting the area for only a short duration. Instead, these population segments need well-designed static/non-interactive hazard maps, as well alternatives to maps such as prominent markers at hazard zone boundaries in the community.

Finally, future research on hazard maps should also examine the effects of hazard zone residence, map experience, cognitive variables such as spatial ability, and demographic variables. For example, these studies should examine the extent to which the findings from experiments using convenience samples generalize to risk area residents that have greater levels of hazard knowledge. In addition, researchers should examine people's responses to understudied hazards such as wildfires because most of the articles on wildfire visualization describes specific wildfire hazard maps (e.g., Liu D. et al., 2019) or wildfire decision support systems (e.g., Pence and Zimmerman, 2011). There is also a need to study floods and hazardous materials releases to see if they elicit responses that differ from those of the more extensively studied hazards reviewed here.

ML reviewed the literature and prepared the manuscript.

The author declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.