Nat Hazards (2017) 88:55–70 DOI 10.1007/s11069-017-2856-9 ORIGINAL PAPER
Enhancing the effectiveness of flood road gauges with color coding Fang Jing1 • Li-Zhuang Yang1 • Ya-Li Peng1 • Ying Wang1 • Xiaochu Zhang1,2,3 • Da-Ren Zhang1
Received: 10 March 2017 / Accepted: 3 April 2017 / Published online: 5 May 2017 Ó Springer Science+Business Media Dordrecht 2017
Abstract Recent years have witnessed many drowning tragedies on roads, even where flood gauges were equipped to warn drivers about the depth of the flood water. The effectiveness of traditional flood gauges using the digit representation has been questioned. According to the attention–knowledge–compliance three-stage model, we propose that color coding can improve the effect of flood gauges. The present study compared the performance of a prototypical color flood gauge with two types of digit flood gauges in four experiments. Our results revealed a general advantage of color flood gauges over digit flood gauges in all testing conditions (with or without reflection and dynamic or static observation), which was manifested in both a sample of people lacking driving experience and a sample of experienced drivers. Specifically, the mean accuracy of color gauge conditions in the four experiments increased by at least 16%, and the mean response time to the color gauge was shortened by at least 0.8 s compared with the digit gauge. The
Electronic supplementary material The online version of this article (doi:10.1007/s11069-017-2856-9) contains supplementary material, which is available to authorized users. Fang Jing and Li-Zhuang Yang have contributed equally to this work. & Da-Ren Zhang
[email protected] Fang Jing
[email protected] Li-Zhuang Yang
[email protected] Xiaochu Zhang
[email protected] 1
CAS Key Laboratory of Brain Function and Disease and School of Life Sciences, University of Science and Technology of China, Huangshan Road 443, Hefei 230027, Anhui, China
2
School of Humanities and Social Science, University of Science and Technology of China, Hefei, Anhui, China
3
Center of Medical Physics and Technology, Hefei Science Center, CAS, Hefei, Anhui, China
123
56
Nat Hazards (2017) 88:55–70
substantial benefit of color gauges over digit gauges suggests that color coding can be used to improve the effectiveness of flood gauges on flood-prone roads. Because of its low cost, the color coding flood gauge may be useful in the developing world to warn against extreme rainfall. Keywords Flood gauge Flooded road Warning sign Traffic safety Color coding
1 Introduction Flooding is a common natural disaster. Because of global climate change, extreme rainfall is expected to increase with global warming (Donat et al. 2016), which may challenge regions where the drainage infrastructure is poor, especially in developing countries where rapid urbanization accompanies incompatible city design (Hapuarachchi et al. 2011; Singh and Kumar 2013; Webster 2013; Yin et al. 2015). Although great progress has been made in the technique of flood warning and forecasting with high spatial–time resolution (Braud et al. 2016), the rate of vehicle-related drowning tragedies on flooded roads is still high (Morss et al. 2016; Yin et al. 2016). Recent years have witnessed many drowning accidents around the world, including America (DailyMail 2016), China (ChinaDaily 2013; Yin et al. 2015), India (Singh and Kumar 2013) and Australia (Hamilton et al. 2016). The reasons are twofold. First, flood forecasting is a difficult job given that a flood is a rapidly evolving threats and is highly dependent on specific locations. Second, some people explain the flood forecasting as less severe based on inappropriate beliefs or low risk perception (Bodoque et al. 2016; Grothmann and Reusswig 2006; Hamilton et al. 2016; Kievik and Gutteling 2011; Ludy and Kondolf 2012; Morss et al. 2016; Parker et al. 2007; Pearson and Hamilton 2014), or have insufficient trust in forecasting systems or experts (Lin et al. 2008; Shamir et al. 2013). Even the ‘‘7.21’’ flood tragedy in Beijing did not change public risk perception for a long time (Su et al. 2015). Therefore, it is optimal to give real-time flood warnings on flood-prone roads to aid drivers in making safe decisions and avoiding drowning accidents. Two kinds of equipment can be used: an automatic monitoring system and the manual observation equipment, such as flood gauges. The advantage of the automatic monitoring system is its automaticity in detecting the water level and its feedback through large LED screens. However, its disadvantage is its vulnerability to issues such as the interruption of power supplies during thunderstorms and its high cost for both establishment and maintenance, which cannot be applied to all flood-prone locations, especially in the developing world. Thus, it is valuable to study how to improve the effectiveness of reliable and low-cost warning systems, such as flood gauges. Flood gauges using digit representations (named digit gauges) are widely used at floodprone locations to act as the warning sign to alert drivers the water depth. For example, the national standard of flood gauges on roads in the USA is W8-19, which uses black digits (MUTCD 2009). Digit gauges are also widely used in China but without an established national standard. It is noteworthy that vehicle-related drowning still occurs, even where the digit gauges are equipped. For example, a digit gauge had been installed before the accidents in Beijing and Guangzhou (ChinaDaily 2013; Zhang et al. 2013; Zheng et al. 2013). Readers can also refer to Note S1 and Fig. S1 in the supplementary material for
123
Nat Hazards (2017) 88:55–70
57
detailed information. The effectiveness of the widely used digit gauge as a warning sign for flooding has been questioned. It should be noted that most digit flood gauges on roads are similar to the design of digit gauges used in rivers and reservoirs. This kind of design is suitable for a technician to read a precise depth level by slow static observation. However, drivers should correctly recognize the warning signal and adopt optimal actions within a few seconds in a fast-moving vehicle. Requirements for the design of road gauges should be similar to those for traffic warning signs. According to the ergonomic principles on the design of traffic warnings (Ben-Bassat and Shinar 2006b; Chan and Ng 2010), especially the attention–knowledge– compliance three-stage model (named the AKC model) (Laughery and Wogalter 2014), an effective warning sign should attract attention with salient features first, then elicit necessary knowledge quickly to promote safe decision making and successfully enable compliance behavior. Color coding is an effective strategy for warning design in all three stages of the AKC model (Laughery and Wogalter 2014), which has been widely used in the design of traffic signs (MUTCD 2009), dashboards of automobiles and clinical equipment (Wilbanks and Langford 2014). We propose that color coding can also be used to improve the efficiency of flood gauges. The risk of driving into flooded roads depends on the depth of water (Liu et al. 2016; Morss et al. 2016; Wu et al. 2016; Yin et al. 2016). For example, water depth lower than 20 cm allows for slow passage, but driving when the water depth is higher than 27 cm is forbidden according to the Beijing Traffic Administration Bureau (see supplementary material, Note S2). The specific depth criteria may vary in different countries and regions (Morss et al. 2016; Yin et al. 2016). Thus, according to the perceived risk level from the gauge, drivers only need to make a quick and categorical decision: turning around, or moving on. The design of flood gauges should speed these decision processes. Color coding, such as red, green and yellow, can efficiently label different categories (Trudel et al. 2015; Wickens et al. 2012) and be implicitly mapped with actions such as ‘‘stop’’ or ‘‘go’’ (Or and Wang 2014; Pravossoudovitch et al. 2014). Thus, we designed a prototypical color flood gauge (named color gauge) using red, yellow and green to indicate a danger, a potential risk or a safe situation, respectively, for driving on the flooded road (Fig. 1). The novelty of the color gauge is that the level of risk can be decoded directly from the color. It is a signal of danger when the gauge is submerged and only the red part can be perceived, and drivers should turn around. It is a signal of potential risk when the gauge is submerged and the yellow part is perceived, and drivers should be more cautious in making a decision. It is safe when the green part of the gauge can be perceived and drivers can pass slowly. As color can be processed automatically (Treisman and Gormican 1988; Wolfe and Horowitz 2004) and implicitly mapped to safety actions (Or and Wang 2014), we hypothesized that the flood gauge using color coding is superior to the digit flood gauge. However, to the best of our knowledge, no study has been performed on the effectiveness of using three-color coding in flood road gauges. To test this hypothesis, we performed four experiments, which manipulated different conditions to simulate the viewing conditions on rainy days. First, there were two viewing distances in the laboratory: 4 and 8 m, which correspond to 50 and 100 m on the road according to the computation of visual angle. Second, the reflection of gauge in the surface water was manipulated. There were two variants: with reflection or not. During the heavy rain, the water surface will be strongly interfered that no reflection can be observed. When the rain stops, the water surface is so calm and mirror-like that a clear reflection of the gauge will emerge on the surface. The last manipulation was to simulate the moving viewing conditions. Specifically, Experiment 1 compared the color flood gauge and the
123
58
Nat Hazards (2017) 88:55–70
Fig. 1 Stimuli of the color and digit flood gauges used in the study. a From left to right: the white digit flood gauge used in Experiment 1, the black digit flood gauge used in Experiments 2–4 and the color gauge used in Experiments 1–4. b Flood gauges dipped into eight levels of water depth but without reflection. c Flood gauges dipped into eight levels of depth with a reflection (see supplementary material Note S3). Only the lower part of each gauge stimuli was presented in b and c for the convenience of presentation. The water depth between 10 and 18 cm is safe for driving; 20–28 cm indicates a potential risk; 30 cm or above indicates a danger. d The visual angle of a gauge (with reflection) increased gradually in the dynamic viewing condition in Experiments 3 and 4 to simulate a driver approaching to the gauge
digit flood gauge with white digits similar to the gauge used in Beijing; Experiments 2, 3 and 4 compared the color gauge with black digits gauge similar to the gauge (W8-19) used in the USA (see Fig. 1; supplementary material).
2 Method 2.1 Experiment 1 2.1.1 Participants Participants were at least 18 years old, with a binocular uncorrected or corrected visual acuity of 5.0 or above and no color vision deficits. After the experiment, they were paid 30 RMB for their participation. The Research Ethics Committee of the University of Science and Technology of China approved this study.
123
Nat Hazards (2017) 88:55–70
59
Sixteen participants (4 females, mean age 20.4 years) attended the experiment. Three of them had obtained driver’s licenses: one had no driving experience in the past 3 years; one had less than 100 driving hours in the past 2 years; and one had approximately 500 driving hours in the past 2 years. None of them had participated in similar experiments.
2.1.2 Task, materials and procedure There were two types of gauges: the color gauge and digit gauge. Both gauges had eight different depths indicating different risk levels, respectively (safe depth: 10, 15, 18 cm; potential risk depth: 20, 25, 28 cm; danger depth: 30, 35 cm; see Fig. 1; Note S2 in supplementary materials). The presentation order of those eight levels of gauge pictures was randomly manipulated in different viewing conditions. The task adopted across the four experiments was a risk-level judgment task, in which participants reported orally whether the picture of gauge indicated a signal of safeness, potential risk or danger. The color gauge indicates the depths of water in red, yellow and green, respectively. Green refers to the safe depth (\20 cm); yellow to the potential risk depth (20–30 cm); and red to the dangerous depth (30 cm and more). See Fig. 1 for an illustration. The white digit gauge was used in Experiment 1. The height of all digits, except the ‘‘5’’ of ‘‘1.5’’ was 14 cm, which was similar to that of W8-19 (approximately 15 cm, MUTCD 2009). The RGB codes of colors used in this study were: red (255,0,0), yellow (255,255,0), green (0,255,0), blue (0,0,255), white (255,255,255) and black (0,0,0). Both the color and digit gauges had two versions: gauge without reflection and gauge with reflection (Fig. 1b, c; and Note S3 in the supplementary material). In Experiment 1, there were two sections, and each section contained two inter-condition-independent variables: gauge type (color and digit) and reflection (with and without). Thus, there were four sub-conditions, where four different gauges (color with reflection, digit with reflection, color without reflection and digit without reflection, Fig. 1) were presented. The orders of the conditions were counterbalanced across subjects. In each condition, each of the eight different depth gauges was displayed four times (a total of 32 trials), and the presentation order of different depth gauge was pseudo-randomized. In a single trial, a fixation point was displayed on a black background for 2 s. Then, a gauge was presented until the subject’s response or after a maximum of 5 s. Subjects were instructed to orally report the risk level (‘‘safe,’’ ‘‘potential risk,’’ ‘‘danger’’) as accurately as possible. The subject’s oral response triggered a voice key within milliseconds (Serial Response Box, Psychology Software Tools, Inc.), which recorded the response time, defined as the timing difference between the onset of gauge stimuli and the onset of the oral response. Then, the experimenter pressed a button to trigger the next trial. Verbal reports of responses were also recorded and scored after the experiment. The presentation of stimuli and timing was controlled via MATLAB and Psychtoolbox-3 (Kleiner et al. 2007). The experiment was conducted in a laboratory, and the observing distance was 8 m in the first section. To simulate the distance of 100 m on the road, the sizes of all gauge figures displayed on the monitor were 8% those of the real gauge. For example, a digit (14 cm) was displayed with a height of 1.12 cm. In the second section, the observing distance was 4 m (corresponding to 50 m in the field). The whole experiment lasted for approximately 50 min.
123
60
Nat Hazards (2017) 88:55–70
2.2 Experiment 2 2.2.1 Participants Twelve participants (6 females, mean age 23.3 years) attended the experiment. Three of them had obtained driver’s licenses but lacked driving experience (\2 h within a year). None of them had attended similar experiments.
2.2.2 Task, materials and procedure The methods of Experiment 2 were the same as in Experiment 1, except that the digit gauge was replaced with the black digit on the yellow background, similar to the gauge W8-19 (see Fig. 1).
2.3 Experiment 3 2.3.1 Participants Twelve participants (5 females, mean age 22.8 years) attended the experiment. Two of them had obtained driver’s licenses but lacked driving experience (\10 h). None of them had attended similar experiments.
2.3.2 Task, materials and procedure The methods of Experiment 3 were basically the same as Experiment 2 except another factor of movement was also manipulated. The visual angle of the gauge stimuli dynamically increased to simulate driving closer to the gauge in the moving condition, while the visual angle of gauge stimuli did not change as in Experiment 1 and Experiment 2 in the static condition. In the moving condition, the first frame of the gauge stimuli was identical to the static one, but gradually enlarged with the speed of 48 ms/slide to simulate a situation where a driver was approaching the gauge at the speed of 10 m/s (i.e., 36 km/h). The enlarged figure disappeared when the subject made a response. In Experiment 3, there were two sections, and each section contained two independent variables: gauge type (color and digit), and movement (moving and static), so there were four conditions (color moving, digit moving, color static and digit static). The orders of the conditions were counterbalanced across subjects. In the first section, both the color and digit gauge without reflection were used; the gauges with reflection were used in the second section. The observing distance was 6 m in the two sections.
2.4 Experiment 4 2.4.1 Participants Twelve participants (1 female, mean age 26.4 years) with driving experience (mean driving years: M = 3.50, SD = 1.31; total driving time per week: M = 3.61 h, SD = 2.91 h) attended the experiment. None of them had attended similar experiments.
123
Nat Hazards (2017) 88:55–70
61
2.4.2 Task, materials and procedure The task, procedure and materials were as same as Experiment 3 except that experienced drivers were recruited.
3 Results Mean response accuracy and mean response times (RTs) for correct responses for each condition were computed for each subject. See Fig. 2 for detailed information (or see supplementary material Table S1 and Table S2 for descriptive statistics). Then, a separate analysis of variance (ANOVA) with all manipulated factors was performed for each experiment. To compare the advantages of the color gauge in separate conditions, we obtained an advantage score by subtracting accuracy (or RT) in the digit gauge condition from accuracy (or RT) in the color gauge condition and then performed pairwise t tests. Details of t test results for accuracy, RT, and details of ANOVA are shown in Tables S1, S2 and S3, respectively, in the supplementary material.
3.1 Experiment 1 3.1.1 Accuracy An omnibus ANOVA with gauge type, view distance and reflection revealed a significant main effect of gauge type [F (1, 15) = 190.352, p \ .001, partial g2 = 0.927], which suggested a general advantage (26.5%) of the color gauge (M = 95.31%, SEM = 0.78%) compared with the digit gauge (M = 68.72%, SEM = 1.52%). See Fig. 2 for an
Fig. 2 Comparison of the color gauge with the digit gauge in accuracy and response times (RT). The color gauge outperformed digit gauge in all four experiments (all ps \ 0.001). The bars with shadow indicate the observing condition that the gauge was in reflection; the bars without shadow indicate the condition of gauge without reflection. Ref reflection condition, NoR no-reflection condition
123
62
Nat Hazards (2017) 88:55–70
illustration. Additionally, the interaction between type and distance was significant [F (1, 15) = 9.294, p = .008, partial g2 = 0.383], and post hoc t test showed that the advantage of the color gauge at longer distances (8 m, M = 31.90%, SEM = 2.04%) was larger than that at 4 m (M = 21.29%, SEM = 3.05%) [t = 3.049, df = 15, p = .008]. There was also a significant interaction between gauge type and reflection [F (1, 15) = 11.335, p = .004, partial g2 = 0.430]. The utility of the digit gauge was seriously undermined when viewed with reflection (M = 63.81%, SEM = 2.54%) than when viewed without reflection (M = 73.64%, SEM = 2.15%) [t = -2.732, df = 15, p = .015]. In contrast, the utility of the color gauge improved when viewed with reflection (M = 96.88%, SEM = 0.82%) compared with without reflection (M = 93.75%, SEM = 1.08%) [t = 2.716, df = 15, p = .016]. Thus, the advantage of the color gauge in the with reflection condition (M = 33.07%, SEM = 2.98%) was larger than that in the without reflection condition (M = 20.12%, SEM = 2.45%) [t = 3.367, df = 15, p = .004]. More details of the ANOVA results are shown in Table S3 in the supplementary material.
3.1.2 Response times The omnibus ANOVA suggested a significant main effect of gauge type [F (1, 15) = 82.378, p \ .001, partial g2 = 0.846], which suggested a response time advantage (963 ms) of the color gauge (M = 960 ms, SEM = 37 ms) over the digit gauge (M = 1923 ms, SEM = 127 ms). Additionally, the interaction between gauge type and reflection was also significant [F (1, 15) = 78.460, p \ .001, partial g2 = 0.840], suggesting a differential advantage of the color gauge. Specifically, the speed advantage in the reflection condition (M = 1204 ms, SEM = 118 ms) was larger than in the no-reflection condition (M = 721 ms, SEM = 100 ms) [t = 8.858, df = 15, p \ .001]. Other interactions were not significant (all ps [ .05). In summary, the results of Experiment 1 demonstrated that participants performed better when viewing the color gauge than when viewing the digit gauge both in accuracy and response times.
3.2 Experiment 2 3.2.1 Accuracy A significant main effect of gauge type was revealed by the three-factor omnibus ANOVA [F (1, 11) = 129.531, p \ .001, partial g2 = 0.922], indicating an advantage (24.2%) of the color gauge (M = 96.36%, SEM = 0.81%) over the digit gauge (M = 72.16%, SEM = 2.67%). Additionally, the interaction between type and distance was also significant [F (1, 11) = 32.051, p \ .001, partial g2 = 0.744], indicating the differential advantage of the color gauge over the digit gauge in different view distances. The superiority of color digits over digits became larger in the long viewing distances (8 m, M = 34.20%, SEM = 2.11%) compared with the short viewing distance (4 m, M = 14.19%, SEM = 3.29%) [t = 5.661, df = 11, p \ .001]. Other interactions were not significant (p [ .05, Table S3 for the details)
123
Nat Hazards (2017) 88:55–70
63
3.2.2 Response times The main effect of gauge type was significant [F (1, 11) = 116.670, p \ .001, partial g2 = 0.914], reflecting a general speed advantage (834 ms) of the color gauge (M = 1005 ms, SEM = 50 ms) over the digit gauge (M = 1839 ms, SEM = 91 ms). All other interactions involving gauge type were not significant (all ps [ .05). In summary, the results of Experiment 2 demonstrated again the superiority of the color gauge over the digit gauge.
3.3 Experiment 3 3.3.1 Accuracy The omnibus ANOVA on accuracy revealed a significant main effect of gauge type [F (1, 11) = 55.317, p \ .001, partial g2 = 0.834], indicating a general advantage (23.4%) of the color gauge (M = 95.18%, SEM = 0.90%) over the digit gauge (M = 71.77%, SEM = 2.64%). All interactions were not significant (all p [ .05), which suggested that the color gauge is superior to the digit gauge in all testing conditions on accuracy measures, e.g., compared with the digit gauge, the color gauge significantly enhanced accuracy by 20.18% and 26.65% in the moving and static conditions, respectively (ps \ .001).
3.3.2 Response times The main effect of gauge type was significant [F (1, 11) = 67.064, p \ .001, partial g2 = 0.859], suggesting a general speed advantage (985 ms) of color gauge (M = 897 ms, SEM = 27 ms) over the digit gauge (M = 1882 ms, SEM = 128 ms). The interaction between type and movement condition was also significant [F (1, 11) = 23.309, p = .001, partial g2 = 0.679]. Specifically, the advantage of the color gauge over the digit gauge was greater in the moving condition (M = 1176 ms, SEM = 146 ms) than in the static condition (M = 792 ms, SEM = 104 ms) [t = 4.828, df = 11, p = .001]. All other interactions involving the gauge type were not significant (all ps [ .05). In summary, the results of Experiment 3 demonstrated that the color gauge advantage was manifested, even in the moving condition.
3.4 Experiment 4 3.4.1 Accuracy The omnibus ANOVA on accuracy revealed a significant main effect of gauge type [F (1, 11) = 45.122, p \ .001, partial g2 = 0.804], indicating a general advantage (16.23%) of the color gauge (M = 95.70%, SEM = 0.89%) over the digit gauge (M = 79.47%, SEM = 2.58%). Additionally, a significant interaction between type and movement was also found [F (1, 11) = 21.268, p = .001, partial g2 = .659]. Performance in the digit gauge condition was vulnerable to the moving/static manipulation, while the color gauge condition was not affected as revealed by pairwise t tests [moving vs static in the digit gauge condition: t = 5.373, df = 11, p \ .001; color gauge condition: t = 0.527, df = 11, p = .609].
123
64
Nat Hazards (2017) 88:55–70
3.4.2 Response times The ANOVA results on response times also showed the general color advantage, manifested by a significant main effect of gauge type [F (1, 11) = 63.584, p \ .001, partial g2 = 0.853], indicating a general advantage (1106 ms) of the color gauge (M = 1206 ms, SEM = 103) over the digit gauge (M = 2312 ms, SEM = 186). Additionally, the color advantage was modulated by movement [type 9 movement: F (1, 11) = 8.991, p = .012, partial g2 = 0.450] and reflection [type 9 reflection: F (1, 11) = 12.673, p = .004, partial g2 = 0.535]. The color advantage was more obviously manifested in the moving condition (M = 1293 ms, SEM = 189 ms) than in the static condition (M = 918 ms, SEM = 101 ms) [t = 2.999, df = 11, p = .012]. The color advantage was more obviously manifested in the reflection condition (M = 1295 ms, SEM = 159 ms) than in the condition without reflection (M = 915 ms, SEM = 138 ms) [t = 3.560, df = 11, p = .004].
3.4.3 Comparison of Experiment 3 and Experiment 4 Experienced drivers in Experiment 4 showed a weaker color advantage on the measure of accuracy but stronger color advantage on the measure of response times, which raised the concern of whether driving experience is a modulating factor for the usefulness of color coding. Corrected RT, computed as response times divided by accuracy, is believed to be insensitive to the speed–accuracy trade-off (McIntyre and Gugerty 2014). To test whether driving experience was a modulating factor, we used the corrected RT as the dependent variable to perform the mixed-factors ANOVA analysis. The results showed that neither the main effect of driving experience nor interactions involving driving experience as a factor reached significance (all ps [ .19). However, the main effect of gauge type was significant [F(1, 23) = 169.817, p \ .0001, partial g2 = 0.939], suggesting that the color gauge advantage also holds true in both skilled drivers (Experiment 4) and people without driving experience or with little driving experience (Experiment 3) given the speed accuracy trade-off. In summary, the results of Experiment 4 verified that the color gauge advantage (accuracy: 16%, RT: 1106 ms) also applied to the sample of experienced drivers. Additionally, experienced drivers have slower responses and higher accuracies, indicating that they may have adopted a cautious strategy in our task.
3.5 The color advantage for each subject We next examined whether the color gauge advantage was consistently replicated individually. The color gauge advantage, defined as the difference between the accuracy (or RT) of the color gauge and the accuracy (or RT) of the digit gauge, was computed individually on both accuracy and RT. A total of 52 participants’ advantages in accuracy and RT were calculated and displayed using frequency tables and scatter plots (see Fig. 3). It is clear that all participants performed better in the color gauge condition than the digit gauge condition.
123
Nat Hazards (2017) 88:55–70
65
Fig. 3 Frequency tables and scatter plot of individual color gauge advantages. a Frequency table of color advantage on accuracy (defined as the difference between mean accuracy on color gauge condition and the mean accuracy on digit gauge condition). b Frequency table of digit advantage on RTs (defined as the difference between mean RTs of color gauge condition and mean RTs of digit gauge condition). c A scatter plot of individual performance on accuracies of color gauge condition and digit gauge condition. d A scatter plot of individual performance on RTs of color gauge condition and digit gauge condition
4 Discussion 4.1 The advantages of the color gauge in different conditions The four experiments demonstrated that the color flood gauge outperformed the digit flood gauge (including both white and black digit versions) in all different testing conditions. Specifically, compared with the digit gauge, the color gauge improved accuracy by 26.5, 24.2, 23.4 and 16.2% in Experiments 1, 2, 3 and 4, respectively, and reduced the response times by 963, 834, 985 and 1106 ms, respectively. The effect size of the color advantage in the four experiments was substantially large (the partial g2 of the main effect of gauge type was at least 0.8) indicating that the color gauge substantially outperforms the digit gauge. The color gauge advantage was also manifested in both people lacking driving experience (Experiment 1, 2 and 3) and experienced drivers (Experiment 4). Moreover, the color gauge advantage was consistently replicated in every subject (n = 52), as shown by individual analysis. Color coding has many advantages according to the three-stage AKC model (Laughery and Wogalter 2014). First, color is a salient feature that can be automatically processed. Numerous studies on attention and perception have shown that color is a very salient feature that efficiently captures attention (Treisman and Gormican 1988; Wolfe and
123
66
Nat Hazards (2017) 88:55–70
Horowitz 2004). Second, it is acknowledged that the meaning of red, yellow and green is a universally understood and simple method for communicating risk information (Dowding et al. 2015). A cross-occupational and cross-culture study also showed a consistent color– concept association (Or and Wang 2014). Ergonomic studies also revealed that red is strongly associated with danger or hazard (Pravossoudovitch et al. 2014). Third, the color is linked with direct and unambiguous behaviors, such as green with ‘‘go’’ and red with ‘‘stop’’ (Or and Wang 2014). Red is especially useful in the design of warning signs, as psychologically, red can motivate one to avoid rather than approach (Mehta and Zhu 2009). Because of the excellent ergonomic properties mentioned above, color coding has been widely used in different situations such as standardized traffic sign systems (MUTCD 2009), the universal traffic light system, dashboard design on motor vehicles and clinical equipment (Wilbanks and Langford 2014) and labeling on the health level of foods (Trudel et al. 2015). The present study demonstrated that color coding can substantially improve the effectiveness of flood gauges, which further verified the soundness of the AKC model as a theoretical framework of warning sign design. As the risk level is represented with three well-known traffic light colors in the color gauge, drivers can efficiently make a safety decision. However, the digit on the digit gauge needs to be interpreted. Thus, the indirect mapping with digit and risk level requires deliberate processing instead of automatic processing, which means longer response times. As a slower safety decision increases the risk of traffic tragedies on flooded roads, it is valuable to expedite the decision-making process as much as we can. Our results demonstrated that color coding can substantially reduce response times. The measure of response times in our study is similar to the measure of perception response time (PRT) or brake reaction time (BRT), which is a key factor for traffic safety (MUTCD 2009). As no study on flood gauges has adopted the measure of response times and accuracies, we refer to the literature reporting PRT or BRT to estimate the practical value of our findings. A meta-analysis shows that the largest decrements in performance by cell phones were in BRT, with an increase of 0.25–0.50 s (Caird et al. 2008). Additionally, BRT was shortened by 0.26 s with the aid of an in-vehicle crash warning system (Bao et al. 2012).The less than half second lag has made some governments enact laws to restrict cell phone usage during driving (Ibrahim et al. 2011). Given the data above, the gain of response time of the color gauge is at least 0.8 s with an accuracy advantage of at least 16%, which may motivate policy makers to reconsider the design of flood gauges. Color gauges are inexpensive, so they should be easily applied in all flood-prone locations, particularly in the developing world. The advantage of color coding is especially manifested in distinguishing a potential risk from a danger. A water level of 28 and 30 cm (Fig. 1) indicated a potential risk and a danger, respectively. To distinguish them is important, but it is difficult to recognize whether the digit gauge is 28 or 30 from a distant location due to the small visual angle. However, the color representation is less vulnerable to distance than the digit representation. Our results showed a 35% benefit at the long viewing distance (8 m, corresponding the field distance 100 m); the accuracies of digit gauges were approximately about 60% and the accuracies of color gauges were at least 95% (see Note S4 and Table S4 for details). We speculate that the manifested color advantage at ranges of 28 and 30 cm may result from a color contrast between red and yellow color and a high luminance contrast between the yellow area and black area on the 28-cm color gauge (Fig. 1b, c), but no such contrasts on the 30-cm gauge. It is noteworthy that the high luminance contrast would be very helpful for people with color vision deficiencies (Owsley and McGwin 2010).
123
Nat Hazards (2017) 88:55–70
67
The advantage of color gauges is also manifested in the reflection condition. Flood gauges sometimes dip into the flood water with a mirrored reflection in the water surface, which undermines drivers’ ability to obtain depth information from the digit gauge. Our results verified that performance in the digit gauge condition was seriously weakened. In contrast to the digit gauge, the performance in the color gauge condition was enhanced. The reason may be that the size of the key color area in the reflection condition was doubled compared with that in the without reflection condition. For example, as mentioned above, the key color area (i.e., the distinguishing feature) between the 28- and 30-cm color gauge is the yellow area (Fig. 1c), so that the yellow viewing area in the reflection condition was doubled as the mirrored part shares the same color and the same size as the upper part. Thus, the yellow signal should be more salient and more easily identifiable in the reflection condition than in the without reflection condition. However, the reflections of the digit and part of the digit are upside down and may produce crowding effects that may interfere with the recognition of the digit gauge. The color advantage we found may also apply to the dynamic viewing situation while driving. Experiment 3 demonstrates that the color gauge enhanced accuracy by 20% and shortened RT by 1.1 s compared with the digit gauge in the moving condition, which suggests a substantial advantage of the color gauge over the digit gauge in the dynamic viewing condition. The color gauge advantages in the measure of response times and accuracies were also manifested in our experienced driver sample (see Experiment 4). Interestingly, a comparison between Experiment 3 and Experiment 4 (experienced drivers) revealed that experienced drivers have a higher accuracy for the digit gauge than people without driving experience and longer response times than people lacking driving experience, which suggests that a different speed–accuracy trade-off strategy might be adopted by experienced drivers. To minimize the impact of strategies, we compared the results of Experiment 3 and Experiment 4 on the measure of corrected RT, a measure considering the speed–accuracy trade-off (McIntyre and Gugerty 2014). Our results indicate that driving experience is not a modulating factor for the color advantage with the speed–accuracy trade-off considered, which provides further evidence that our findings can be generalized to the population of drivers.
4.2 Limitations and future directions Laboratory studies, including static experiments, are meaningful to evaluate the effectiveness of different types of signs (Ben-Bassat and Shinar 2006; Shinar and Vogelzang 2013; Ting et al. 2008). Despite this, more experiments, that are closer to real-life situations, such as driving simulator and field experiments, are necessary in the future. Whether the color gauge advantage can also be manifested in the population of older drivers needs further research. Previous studies have shown that older drivers are less efficient in the comprehension of traffic signs (Ben-Bassat and Shinar 2015; Liu and Jhuang 2012), in risk detection (Clarke et al. 2007) and in other cognitive processing abilities (Hashim et al. 2014). Thus, the age factor should be investigated in future studies. One potential future direction is to combine the color gauge with a warning sign for more effectiveness. We focused on the contrast between the color gauge and digit gauge because the presentation of water-depth information is the most relevant factor influencing drivers’ risk perception. In the US standard (MUTCD 2009), there is another sign, i.e., the warning sign W8-18, the ‘‘ROAD MAY FLOOD.’’ The function of the warning sign is to alert drivers of the potential risk of making necessary preparations. Thus, a better design and installation of the gauge combined with a warning sign to prevent drowning should be
123
68
Nat Hazards (2017) 88:55–70
examined in the future, e.g., what is the suitable distance between the warning sign and the gauge, and whether it is helpful to add distance on the sign, as a spatial or distance cue, and whether it is better to add words to the warning sign to alert drivers to the flood gauge. Studies using a driving simulator or field experiment to answer the above questions are needed.
5 Conclusions The current experiments demonstrated that the performances for the color flood gauge were better than the digit flood gauge in all conditions (gauge with or without reflection, and moving or static presentation/observation). The mean color accuracy of the first three experiments was higher than the digit gauge by at least 20%, and the mean color RT was shortened by 0.8 s. The Experiment 4 showed similar color advantages for experienced subjects as well. The results suggest that using color coding can enhance the effectiveness of the traditional digit flood gauge. Acknowledgements This work was supported by grants from the National Natural Science Foundation of China (Grant Nos. 31500917, 30870764, 31471071), China Postdoctoral Science Foundation (Grant No. 2016M592051) and Hefei Science Center, CAS ‘‘User with Potential’’ (Grant No. 2015HSC-UP017).
References Bao S, LeBlanc DJ, Sayer JR, Flannagan C (2012) Heavy-truck drivers’ following behavior with intervention of an integrated, in-vehicle crash warning system: a field evaluation. Hum Factors 54:687–697. doi:10.1177/0018720812439412 Ben-Bassat T, Shinar D (2006) Ergonomic guidelines for traffic sign design increase sign comprehension. Hum Factors 48:182–195. doi:10.1518/001872006776412298 Ben-Bassat T, Shinar D (2015) The effect of context and drivers’ age on highway traffic signs comprehension. Transp Res F Traffic 33:117–127. doi:10.1016/j.trf.2015.07.009 Bodoque JM, Amerigo M, Diez-Herrero A, Garcia JA, Cortes B, Ballesteros-Canovas JA, Olcina J (2016) Improvement of resilience of urban areas by integrating social perception in flash-flood risk management. J Hydrol 541:665–676. doi:10.1016/j.jhydrol.2016.02.005 Braud I, Borga M, Gourley J, Hurlimann M, Zappa M, Gallart F (2016) Flash floods, hydro-geomorphic response and risk management. J Hydrol 541:1–5. doi:10.1016/j.jhydrol.2016.08.005 Caird JK, Willness CR, Steel P, Scialfa C (2008) A meta-analysis of the effects of cell phones on driver performance. Accid Anal Prev 40:1282–1293. doi:10.1016/j.aap.2008.01.009 Chan AHS, Ng AWY (2010) Investigation of guessability of industrial safety signs: effects of prospectiveuser factors and cognitive sign features. Int J Ind Ergon 40:689–697. doi:10.1016/j.ergon.2010.05.002 ChinaDaily (2013) Beijing takes lesson from fatal downpour. http://usa.chinadaily.com.cn/china/2013-07/ 20/content_16805895.htm Clarke DD, Ward P, Bartle C, Truman W (2007) The role of motorcyclist and other driver behaviour in two types of serious accident in the UK. Accid Anal Prev 39:974–981. doi:10.1016/j.aap.2007.01.002 DailyMail (2016) Chilling video of Houston flood victim’s final moments shows her driving into submerged underpass and frantically trying to escape sinking SUV. http://www.dailymail.co.uk/news/article3551725/Chilling-video-Houston-flood-victim-s-final-moments-shows-driving-submerged-underpassfrantically-trying-escape-sinking-SUV.html Donat MG, Lowry AL, Alexander LV, O’Gorman PA, Maher N (2016) More extreme precipitation in the world’s dry and wet regions. Nat Clim Chang 6:508. doi:10.1038/Nclimate2941 Dowding D et al (2015) Dashboards for improving patient care: review of the literature. Int J Med Inform 84:87–100. doi:10.1016/j.ijmedinf.2014.10.001 Grothmann T, Reusswig F (2006) People at risk of flooding: why some residents take precautionary action while others do not. Nat Hazards 38:101–120. doi:10.1007/s11069-005-8604-6
123
Nat Hazards (2017) 88:55–70
69
Hamilton K, Peden AE, Pearson M, Hagger MS (2016) Stop there’s water on the road! Identifying key beliefs guiding people’s willingness to drive through flooded waterways. Saf Sci 89:308–314. doi:10. 1016/j.ssci.2016.07.004 Hapuarachchi HAP, Wang QJ, Pagano TC (2011) A review of advances in flash flood forecasting. Hydrol Process 25:2771–2784. doi:10.1002/hyp.8040 Hashim MJ, Alkaabi MSKM, Bharwani S (2014) Interpretation of way-finding healthcare symbols by a multicultural population: navigation signage design for global health. Appl Ergon 45:503–509. doi:10. 1016/j.apergo.2013.07.002 Ibrahim JK, Anderson ED, Burris SC, Wagenaar AC (2011) State laws restricting driver use of mobile communications devices: distracted-driving provisions, 1992–2010. Am J Prev Med 41:238–239 Kievik M, Gutteling JM (2011) Yes, we can: motivate Dutch citizens to engage in self-protective behavior with regard to flood risks. Nat Hazards 59:1475–1490. doi:10.1007/s11069-011-9845-1 Kleiner M, Brainard D, Pelli D (2007) What’s new in Psychtoolbox-3? Perception 36:14 Laughery KR, Wogalter MS (2014) A three-stage model summarizes product warning and environmental sign research. Saf Sci 61:3–10. doi:10.1016/j.ssci.2011.02.012 Lin SY, Shaw DG, Ho MC (2008) Why are flood and landslide victims less willing to take mitigation measures than the public? Nat Hazards 44:305–314. doi:10.1007/s11069-007-9136-z Liu YC, Jhuang JW (2012) Effects of in-vehicle warning information displays with or without spatial compatibility on driving behaviors and response performance. Appl Ergon 43:679–686. doi:10.1016/j. apergo.2011.10.005 Liu J, Shi ZW, Wang D (2016) Measuring and mapping the flood vulnerability based on land-use patterns: a case study of Beijing, China. Nat Hazards 83:1545–1565. doi:10.1007/s11069-016-2375-0 Ludy J, Kondolf GM (2012) Flood risk perception in lands ‘‘protected’’ by 100-year levees. Nat Hazards 61:829–842. doi:10.1007/s11069-011-0072-6 McIntyre SE, Gugerty L (2014) Applying visual attention theory to transportation safety research and design: evaluation of alternative automobile rear lighting systems. Accid Anal Prev 67:40–48. doi:10. 1016/j.aap.2014.02.007 Mehta R, Zhu R (2009) Blue or red? Exploring the effect of color on cognitive task performances. Science 323:1226–1229. doi:10.1126/science.1169144 Morss RE, Mulder KJ, Lazo JK, Demuth JL (2016) How do people perceive, understand, and anticipate responding to flash flood risks and warnings? Results from a public survey in Boulder, Colorado, USA. J Hydrol 541:649–664. doi:10.1016/j.jhydrol.2015.11.047 MUTCD (2009) The manual on uniform traffic control devices U.S. Department Transportation Federal Highway Administrator Or CKL, Wang HHL (2014) Color-concept associations: a cross-occupational and -cultural study and comparison. Color Res Appl 39:630–635. doi:10.1002/col.21832 Owsley C, McGwin G Jr (2010) Vision and driving. Vis Res 50:2348–2361. doi:10.1016/j.visres.2010.05. 021 Parker D, Tapsell S, McCarthy S (2007) Enhancing the human benefits of flood warnings. Nat Hazards 43:397–414. doi:10.1007/s11069-007-9137-y Pearson M, Hamilton K (2014) Investigating driver willingness to drive through flooded waterways. Accid Anal Prev 72:382–390. doi:10.1016/j.aap.2014.07.018 Pravossoudovitch K, Cury F, Young SG, Elliot AJ (2014) Is red the colour of danger? Testing an implicit red-danger association. Ergonomics 57:503–510. doi:10.1080/00140139.2014.889220 Shamir E, Georgakakos KP, Spencer C, Modrick TM, Murphy MJ, Jubach R (2013) Evaluation of real-time flash flood forecasts for Haiti during the passage of Hurricane Tomas, November 4-6, 2010. Nat Hazards 67:459–482. doi:10.1007/s11069-013-0573-6 Shinar D, Vogelzang M (2013) Comprehension of traffic signs with symbolic versus text displays. Transp Res F Traffic Psychol Behav 18:72–82. doi:10.1016/j.trf.2012.12.012 Singh O, Kumar M (2013) Flood events, fatalities and damages in India from 1978 to 2006. Nat Hazards 69:1815–1834. doi:10.1007/s11069-013-0781-0 Su Y, Zhao F, Tan LZ (2015) Whether a large disaster could change public concern and risk perception: a case study of the 7/21 extraordinary rainstorm disaster in Beijing in 2012. Nat Hazards 78:555–567. doi:10.1007/s11069-015-1730-x Ting PH, Hwang JR, Fung CP, Doong JL, Jeng MC (2008) Rectification of legibility distance in a driving simulator. Appl Ergon 39:379–384. doi:10.1016/j.apergo.2007.08.002 Treisman A, Gormican S (1988) Feature analysis in early vision evidence from search asymmetries. Psycho Rev 95:15–48. doi:10.1037//0033-295x.95.1.15 Trudel R, Murray KB, Kim S, Chen S (2015) The impact of traffic light color-coding on food health perceptions and choice. J Exp Psychol Appl 21:255–275. doi:10.1037/xap0000049
123
70
Nat Hazards (2017) 88:55–70
Webster PJ (2013) Improve weather forecasts for the developing world. Nature 493:17–19 Wickens CD, Hollands JG, Banbury S, Parasuraman R (2012) Engineering psychology and human performance, 4th edn. Pearson, Cambridge. Wilbanks BA, Langford PA (2014) A review of dashboards for data analytics in nursing. CIN Comput Inform Nurs 32:545–549. doi:10.1097/Cin.0000000000000106 Wolfe JM, Horowitz TS (2004) What attributes guide the deployment of visual attention and how do they do it. Nat Rev Neurosci 5:495–501. doi:10.1038/nrn1411 Wu XH, Zhou L, Gao G, Guo J, Ji ZH (2016) Urban flood depth-economic loss curves and their amendment based on resilience: evidence from Lizhong Town in Lixia River and Houbai Town in Jurong River of China. Nat Hazards 82:1981–2000. doi:10.1007/s11069-016-2281-5 Yin J, Ye MW, Yin Z, Xu SY (2015) A review of advances in urban flood risk analysis over China. Stoch Environ Res Risk A 29:1063–1070. doi:10.1007/s00477-014-0939-7 Yin J, Yu DP, Yin Z, Liu M, He Q (2016) Evaluating the impact and risk of pluvial flash flood on intraurban road network: a case study in the city center of Shanghai, China. J Hydrol 537:138–145. doi:10. 1016/j.jhydrol.2016.03.037 Zhang DL, Lin YH, Zhao P, Yu XD, Wang SQ, Kang HW, Ding YH (2013) The Beijing extreme rainfall of 21 July 2012: ‘‘Right results’’ but for wrong reasons. Geophys Res Lett. doi:10.1002/grl.50304 Zheng ZP, Qi SZ, Xu YT (2013) Questionable frequent occurrence of urban flood hazards in modern cities of China. Nat Hazards 65:1009–1010. doi:10.1007/s11069-012-0397-9
123