Changing your perspective: the impact of different visualisation methods on seismic hazard maps

A colour map (or colour scheme) is the visual translation of data values. For example, in cases where we have geographical data that we can only present in two dimensions, a colour map can act as a third axis. The topography of a region (the height above or below sea-level) can be shown through mapping the colours to height. This practice is widely applied in academia, but often the choice of colour map is done in a passive way, for example, by using the default colour values of a software program. However, when analysing data for the first time, visualisation methods such as the choice of colour map play an important role in influencing how a user, whether scientist or member of the public, interprets results (Crameri et al. 2020). In Fig. 1, we present an original image (Fig. 1a) that becomes distorted with a choice of colour map (Fig. 1b). In this example, the rainbow-style jet colour map highlights some points over others in the converted image (Fig. 1b). This can be seen in Terry Fox’s t-shirt being highlighted, as well as his face being difficult to recognise. A reason behind this is that the jet (rainbow) colour transition (e.g., the changes between the different colours in the palette) is not equally distributed, which smothers some data points and accentuates others (Fig. 1b). Conversely, the colour map in Fig. 1c has an even transition between the colours and returns the original image well. This even transition creates a perceptually uniform colour map, with the colours being perceptually ordered in accordance to brightness. In this paper, we will discuss how this difference in colour visualisation shown in Fig. 1 could potentially cause miscommunication in data interpretation (Crameri et al. 2020).

In addition to data distortion, rainbow and jet colour schemes cause issues in the field of accessibility and inclusivity. The uneven colour gradient, and sub-selections thereof such as red–green colours, can create difficulties in determining unique values for those with certain colour vision deficiencies (CVDs), as shown in Fig. 2. Given that large random population surveys indicate that potentially 1 in 12 men (and a much smaller number for women) have some form of CVD (Birch 2012), the application of rainbow/jet style colour maps could be exclusionary to millions of Canadians.

Despite the deficiencies in rainbow/jet being well known (e.g., Light and Barltein 2004; Hagemeier-Klose and Wagner 2009; Hawkins 2015; Kovesi 2015; Stauffer et al. 2015; Moreland 2016; Thyng et al. 2016; Crameri et al. 2020; Zeller and Rogers 2020; Kaspar and Crameri 2022), the Earth Science community still commonly uses such colour schemes to present data. Of the many different disciplines that comprise Earth Sciences, the subject of geophysics often applies rainbow/jet colour maps to its data analysis. Figures 3a–3c show recently published images of geoscience data from the Canadian Journal of Earth Sciences and other scientific journals. Although several academic journals have recently put policies in place to exclude rainbow/jet style colour maps to enhance accessibility and inclusion (IPCC 2022; Elsevier 2024; Tektonica 2024), such policies, and their compliance, are not universal. Furthermore, public-facing data such as weather and hazard warnings are commonly presented using uneven colour maps (Fig. 3d). Given the widespread importance for the public to be able to rapidly and accurately access data such as earthquake hazard, then such public-facing information is required to be as inclusive as possible.

An earthquake can be a devastating natural disaster, causing casualties and economic loss due to abrupt changes in ground motion. Although we cannot prevent natural earthquakes from happening, we could mitigate the impact of an earthquake by understanding potential ground motion and how this may impact infrastructure (e.g., potential damage). A number of countries create seismic hazard maps that show a parameter related to ground motion based on previous earthquake history, regional rock type, and probability analysis. Recent examples of publicly available seismic hazard maps from USA, Europe, and Canada are shown in Fig. 4.

Despite previous work in the region of communicating hazard through maps (Hagemeier-Klose and Wagner 2009; Thompson et al. 2015; Miran et al. 2017), and the perception of seismic hazard specifically (Gaspar-Escribano and Iturrioz 2011; Marti et al. 2019; Fyfe and Molnar 2020; Schneider et al. 2023; Dallo et al. 2024a, 2024b), all the hazard maps in Fig. 4 are presented using an uneven colour scheme.

In this paper, we focus on the impact that applying a rainbow/jet colour map has on the interpretation and perception of the hazard. The application of rainbow/jet colour maps is still widespread, particularly for seismic hazard, but here, we use the Canadian Seismic Hazard Map (CanadaSHM6, produced by Natural Resources Canada (NRCan)) as an example. This map was chosen for its hazard-related content but also the original plotting choices, including the rainbow-like colour map and the irregular spacing of tick marks on the colour legend (Fig. 4c). In the “Materials and Methods” section, we give an overview of the data and our visualisation methods, and then present the impact of changing the colour map in the “Results” section. We highlight the potential next steps in the “Discussion”. The aim of this paper is to raise the discussion of colour map choice specifically in the area of Canadian Earth Science, where inaccessible and exclusionary practices are still prevalent.

For this study, existing data were visualised with the widely used proprietary software ArcGIS Pro (ESRI). As we are interested in presenting data that have significant public interest, we are using NRCan’s Seismic Hazard Map (Fig. 4c; accessed 2 April 2023). First, we describe what a seismic hazard map represents and provide an overview of how such data are created. Finally, we describe the overall process for re-creating the presented colour map, though additional methodological details are presented in the “Results” section for ease of reading.

Since 1897, the Government of Canada has been monitoring earthquakes through the Geological Survey of Canada. The first Canadian Seismic Hazard Map was prepared in 1953 and showed how likely it is that a location will experience damaging ground motion caused by earthquakes (Adams 2010; Adams 2019). A map from the most updated model, the 6th Generation Hazard Model and Seismic Hazard Maps (CanadaSHM6) (Adams et al. 2019), is shown in Fig. 4c and highlights the potential ground motion related to earthquakes propagating across Canada.

When calculating the Seismic Hazard for CanadaSHM6, Canada is divided into eastern and western regions due to the differences in tectonic structure, wave propagation rate, and earthquake history (Kolaj et al. 2019). Western Canada is exposed to higher amounts of seismic events than eastern Canada because a part of the western region is located along the Pacific “Ring of Fire” (Fig. 5). On average, western Canada (as defined in Fig. 5) experiences around 1000 earthquakes a year (Government of Canada 2019a). Eastern Canada is located within a more tectonically stable region of the continental North American plate (Fig. 5). However, earthquakes also happen in eastern Canada (Fig. 5), where approximately 450 earthquakes occur yearly (Government of Canada 2019b), including earthquakes related to ancient tectonic rift margins, such as the western Quebec and St. Lawrence seismic zones in southeastern Canada (Kolaj et al. 2020c).

The Cornell–McGuire methodology and Probabilistic Seismic Hazard Assessment (PSHA) were used to create CanadaSHM6 (Adams et al. 2019). The Cornell–McGuire methodology (McGuire 1993) uses historical seismicity and the regional tectonic setting to estimate the Seismic Hazard values (Pavlenko and Kijko 2019). PSHA (Richter 1958; Weichert 1980) considers all possible regional seismic sources and calculates the various ground motion levels with a specific probability within a specific time interval (Haase et al. 2011; Pavlenko and Kijko 2019). Within PSHA, the ground motion models and source model together provide future predictions of the probabilities of ground motions (Kolaj et al. 2020a, 2020b). To present predictions of future ground motion, the open-source application OpenQuake (GEM 2022) has been used in CanadaSHM6 (Allen et al. 2020). A detailed description of the methodology used to calculate the seismic hazard values can be found in Kolaj et al. (2020a, 2020b).

Site condition, describing the interaction of seismic waves with local rock/soils at the surface, is also essential in hazard calculation. In seismology, the P-wave, the primary wave (or compressional wave), is the first wave that creates vertical vibration, followed by the S-wave (shear wave), which arrives later and slower, creating horizontal and vertical vibration (Yoshida 2015). Shear waves are important in earthquake-resistant design as they are greater in amplitude and more destructive compared to the P-wave (Yoshida 2015). The site condition parameter Vs30 is defined as the time-averaged shear wave velocity in the upper 30 m of the crust and is used to understand the rock/soil conditions and how seismic waves propagate.

There are five different site classes for ground conditions across Canada (A, B, C, D, E), and the values of each class represent different ranges of Vs30 values, going from fast to slow (Table 1) (Kolaj et al. 2019). Among all the levels, Site Class C (Vs30 = 360–760 m/s) is the condition most used to visualise the Canadian Seismic Hazard data.

After taking into consideration historical earthquake locations, local and regional tectonics, rock condition, and probabilities of rock failure, the output parameters of a seismic hazard map include peak ground acceleration (PGA, unit: g) and peak ground velocity (PGV, unit: m/s), and spectral acceleration (Sa, unit: g). Gravity is commonly used in the seismic hazard models and is sometimes used as percentage g; the acceleration of gravity is defined as 9.80665 m/s² (U.S. Geological Survey 2023). PGV is the greatest speed reached by the ground movement, and PGA is the peak acceleration experienced by a particle on the ground (U.S. Geological Survey 2023). Spectral accelerations (Sa(T)) are used to understand the expected motion experienced by a building measured by a particle on a massless rope on top of a building and having the same natural period (T) of vibration with the building as the ground shaking (U.S. Geological Survey 2023).

The different periods/frequencies of vibration will impact different building types. The natural frequency of a building is the frequency that the building moves with the force, which is determined by the weight, stiffness, and height of the building (Eriksson 2018). For example, short buildings with less than seven storeys will be damaged by short-period spectral acceleration, such as a time period of 0.2 s, and buildings with over seven storeys, will be damaged by low frequency, such as 10 s. Spectral acceleration (Sa(T)) is used to understand the maximum building motion at period T. Periods of 0.2, 0.5, 1.0, 2.0, 5.0, and 10.0 s were used in the CanadaSHM6 to understand the maximum building motions (Adams et al. 2019), with 0.2 s being the most commonly applied in the seismic hazard maps.

A key factor in seismic hazard maps is the probability level attached to the ground motion calculations. The probability level used in CanadaSHM6 (and shown in Fig. 4c) is a 2% chance of exceedance in 50 years, which is commonly the basic probability level for seismic safety regulations and design standard for building codes (Wang and Ormsbee 2005). This is equivalent to an annual probability of 0.04% that a location will exceed a given ground motion value (Kolaj et al. 2020a). This probability tolerance has been used by Canada’s Seismic Hazard Maps since 2005 (Kolaj et al. 2020b).

For this project, NRCan provided the completed dataset for CanadaSHM6 in CSV format, containing seismic hazard data relating to 34 733 locations across onshore and offshore Canada. No new data were introduced during this project. Moreover, NRCan also provided the colour scheme (with the legend interval) to support the re-visualising of the seismic hazard map used in this study. Therefore, an exact colour scheme has been used to generate the government published map presented here.

The dataset NRCan provided contains Sa(0.2) across Canada with X450 Site Designation and a 2% chance of exceedance in 50 years (where X450 represents the Vs30 value equal to 450 m/s): X450 is the condition most used in CanadaSHM6 for nationwide maps, while a spectral acceleration period of 0.2 s describes potential shaking to low-storey buildings, which is commonly used to simplify seismic hazard map for public consumption (Government of Canada 2019d).

The above review of how seismic hazard maps can be created (here, specifically for the CanadaSHM6) is important as it highlights the intricate and complex nature of the data. Despite these multifaceted methods and datasets, the resulting interpretation by the public needs to be as clear and concise as possible. Seismic Hazard Maps are used in multiple socio-economic and industrial areas, such as for emergency response (Shedlock et al. 2000) as well as developing the National Building Code to design and construct buildings as earthquake-proof as possible. Buildings that can withstand an earthquake will minimise casualties and economic loss. Furthermore, the insurance industry also uses seismic hazard maps to protect people and recover financially from earthquake damage (King et al. 2014). Given its importance, seismic hazard maps should be universally readable for all users, professionals, and the public.

The research here aims to highlight the inaccessibility of presenting hazard data with rainbow/jet colour maps, and to understand the impact of changing visualisation methods on data interpretation (and indeed how this might impact industries such as insurance and construction). To accomplish this, we introduce two methods of analysis in the form of k-means clustering-based image segmentation and a colour comparison.

By applying colour to our data, we are creating boundaries relating to the seismic hazard. For example, we can identify with the colour boundaries area of high seismic hazard and low seismic hazard. We apply image segmentation (e.g., Dhanachandra et al. 2015) to see whether this data science technique can identify any difference in the generation of such boundaries due to the choice of colour map. In image segmentation, pixels that have similar attributes are grouped together—in this case, into a total of three different clusters representing low, medium, and high values. Using the package imsegkmeans in Matlab R 2024a, we create a pixel-wise mask for objects in an image, which gives us a more comprehensive and granular understanding of the object, and then comparing these cluster patterns provides evidence on how these images could be perceived differently. If the cluster patterns are the same between colour maps, then there is no perceived difference in the high and low boundaries. However, if there are changes in the patterns, then the colour map is creating different boundaries between the data.

In the “Results” section, we will present a re-created seismic hazard map and then apply different visualisation methods to investigate the effects of a rainbow/jet colour map. Finally, the importance of these results and their potential impact will be addressed in the “Discussion” section.

This section presents the results of the following changes: re-creating the published map; changing intervals to log scale; applying equal intervals to the colour bar; applying even colour gradient; and applying CVD filters to the colour map.

The first step of the study was to recreate a seismic hazard map with similar plotting methods and compare the output with the CanadaSHM6 map that is publicly available (Fig. 6a). Figure 6b was created in ArcGIS Pro 3.0.0 using NRCan’s CanadaSHM6 dataset and legend interval. As a result, the pattern of the seismic hazard across the country is similar between the two maps. For example, in both maps, the hazard values near Victoria and Montreal are significantly higher than in Central Canada, and there exists a high seismic hazard in the Canadian polar region (Fig. 6).

To obtain this re-created map from the original data using ArcGIS Pro, the Inverse Distance Weighted tool was used to produce a continuous map from individual data points (Fig. 6b). The Inverse Distance Weighted Interpolation methods (Spatial Analyst Tools) estimate cell values for the missing areas by averaging the sample data points in the surrounding processing cell. The closer the sample data points, the more weight will be assigned. The individual data points were interpolated to the raster file that covers the entire region of Canada (including onshore and offshore regions), then clipped to the same boundary as the other Seismic Hazard Maps of Canada. The original map by NRCan (Fig. 6a) contains contour lines based on the data values (though which values or increments is not clear), as well as a smoother change between the data values as compared to Fig. 6b. This may be due to different smoothing methods used during the interpolation of the data, which is not recorded in the public domain. This research will focus on the general patterns, and the smoothing methodology will not influence the result.

A key part to this study is not only the uneven colour map provided (e.g., rainbow/jet) but the unequal intervals between the axis ticks. At first glance, the unequal intervals appear to be in log scale (e.g., log10), although this is not explicitly stated in the literature surrounding the seismic hazard map (e.g., Adams et al. 2019). To test this hypothesis, we applied a log scale to the supplied data in Fig. 7. The pattern generated on our log scale map (Fig. 7) is similar to the public CanadaSHM6 (Fig. 6a), but not exactly the same. The values between 0 and 0.07 g were divided into five intervals in Fig. 7b (log value from −2.45 to −1.15). In contrast, the original map categorises all those values into less than two continuous colour intervals (red arrows, Fig. 7a). Therefore, the original colour map (Figs. 6a and 7a) does not have a log scale applied (Fig. 7b).

After concluding that the data were not presented in log scale, we next applied an equal interval across the colour map in Fig. 8. This was done by taking the maximum and minimum values provided in Fig. 6a (0–5 g) and dividing into 11 different equal intervals. The hazard perception for most of Canada has changed significantly after updating the legend to equal intervals. Most of the places in Canada show a light purple colour on the map, meaning most areas are below 1 g spectral acceleration values. The colour visualisation of hazard for the most populated regions, as indicated by the cities plotted, has reduced significantly as compared to the original map. The only place with a red colour (above 3.5 g spectral acceleration) is on and adjacent to Graham Island, the island north of Vancouver Island in British Columbia (Fig. 8).

The results presented so far have used a rainbow-like colour map, in Fig. 9, we present the dataset with an even colour gradient. In this case, we apply the “batlow” colour map (Crameri 2018), which is perceptually uniform (even changes between the colours) and perceptually ordered (colours in order of brightness), as well as being CVD friendly (Fig. 2). It has several hues and is considered an alternative to rainbow-like colour maps. In Fig. 9a, we apply batlow with an even colour interval classification (c.f., Fig. 8) and then in Fig. 9b with a log scale interval.

When comparing to the equal interval colour map batlow (Fig. 9) with the rainbow variation originally used (Fig. 8), the transitions between high and low values is smooth with the batlow colour scheme (Fig. 9a). The darkest areas are those with lower spectral accelerations values, and the colour becomes lighter as the value increases (Fig. 9a). In contrast, the value transition for the rainbow colour scheme is not as straightforward. For example, the circled areas in the batlow map showing “medium” levels of spectral acceleration in Nunavut and the Rockies (Fig. 9a) are not as easily identifiable in the rainbow colour map (Fig. 8). The rainbow colour map creates only a subtle difference to the surrounding area, with the circled region appearing washed out (Fig. 8).

The log scale map was also visualised using the batlow colour scheme (Fig. 9b). As a result, the perception of the log scale of CanadaSHM6 using the batlow colour scheme will show the same pattern as the original CanadaSHM6 colour scheme (Fig. 7a). Using the batlow colour scheme to visualise, the CanadaSHM6 does not identify any new features that were not shown using the original colour scheme. Both Figs. 7a and 9b highlighted the high-risk areas. However, the regions that transition from low to high risk can be easily identified in the batlow colour scheme compared to the original colour scheme (e.g., region between British Columbia and Alberta).

To better understand the impact on visualisation by implementing the change in colour map, we have applied a clustering technique to images that use the original colour map (e.g., rainbow/jet) and with a uniform colour map (e.g., batlow). This allows for a more direct analysis of the differences between the visual representation of the data. Figure 10 shows the k-means clustering-based image segmentation for three clusters (e.g., Dhanachandra et al. 2015) on a number of different images of the data. Three clusters are chosen to represent regions of low, medium, and high hazard based on the colours presented by the image. Comparing the visual clusters identified by this technique in Figs. 10a and 10b, there are differences in cluster patterns of the data between even colour scale plots (Figs. 8 and 9a). In particular, there are more designated areas of high and medium values in the rainbow/jet colour map (Fig. 10a) than in the batlow scheme, which clusters most values as low (Fig. 10b).

In Figs. 10c and 10d, we compare the cluster pattern between log scale representations of the data with rainbow/jet and batlow. By implementing a segmentation of three clusters to the images, there are again differences between the cluster patterns of low, medium, and high data values. For example, the batlow colour map in Fig. 10d produces a pattern of high spectral acceleration in the region of Quebec and Ontario (Fig. 10d), which is not well defined using a non-uniform colour map (Fig. 10c). The impact on the cluster patterns through the choice of colour map shows less defined areas and more diffuse zones. The important point here is that the choice of colour map, and the gradients between the colours, can create different automated clusters of data (and not the data values itself).

Given that the rainbow/jet colour scheme is not a perceptually uniform and ordered colour scheme, it is known to not be user-friendly for CVD users (Crameri et al. 2020). To assess the seismic hazard map for universal accessibility, the Colour Vision Simulator and greyscale tool within ArcGIS Pro was used to test the data against common CVDs (Fig. 11). The black and white map (8-bit greyscale to simulate total colour blindness, monochromacy) shows the visualisation issue with the jet colour scheme in that the values around 0.1, 0.2, and 0.95 g are similar grey colours, which are difficult to distinguish (Fig. 10a, highlighted by box and arrows). We performed a colour comparison on the colours by taking the values of red, green, and blue for each set (a number out of 255) and then calculating the similarity between them (see eq. 1). For 0.1 and 0.2 g, the similarity is 98% (Fig. 10a). Given these colours are so similar, the only way to understand the difference in greyscale is to check the data behind the map and read the values for each data point.

A similar issue with distinguishing values appears in the CVD scenarios of tritanopia (blue-blind vision, Fig. 11b), and deuteranopia (green-blind vision, Fig. 11c). In Fig. 11b, the spectral acceleration values of 0.05 and 0.5 g are shown in the same colour within the colour map, as well as values around 0.12, 0.17, and 0.3 g. When applying the red, green, and blue values to eq. 1, our colour comparison between 0.07 and 0.45 g is 95% (Fig. 11b). Figure 11c shows 0.07 and 0.15 g as well as 0.2 and 0.9 g to be matching in colour. Furthermore, the areas between 0.4 and 0.8 g are represented in similar colours without an obvious transition between high and low values. Our colour comparison between 0.26 and 0.88 g is 87% using eq. 1 (Fig. 11c).

Results from our testing show the impact of changing the visualisation methodology on the perception of the dataset. In this case, we use a seismic hazard map as our dataset, as it is both scientifically complex and of high societal value. However, the dataset itself is not the focus of this paper, rather it is the visualisation of information through methods that are not accessible and inclusive (e.g., rainbow/jet colour maps). Fundamentally, colour maps that do not have even gradients between the colours can distort data and become inaccessible to a proportion of the population. To highlight this point, we can represent the colour map as a position axis (Fig. 12) where it becomes clear that the colour changes across the map are not even.

Given that distance would never be represented with a distorted position axis such as the one provided in Fig. 12b, then scientific data should not be distorted through a decision on a colour map. As shown in Fig. 11, the uneven change between the colours can also make the data not universally accessible, as readers with CVDs are unable to fully read the information (as shown by our colour comparison value, C, eq. 1). Furthermore, by implementing k-means clustering image segmentation, we generate a qualitative view of the difference the choice of colour map makes in creating boundaries between data points (Fig. 10). For instance, by segmenting our image into three clusters, we can generate different patterns (boundaries) between the high and low data.

Focussing on the Canadian seismic hazard map as presented in Fig. 6a, we identify two key areas of concern. The first is that the intervals between the colours are uneven as they increase from 0 to 5 g. The data are presented in a scale that is similar to a log scale but not exactly. Log scale maps are commonly used when dealing with a large scale and many data points at the tail of the data range. Thus, the log scale could theoretically be applied to the dataset, but the legend of the CanadaSHM6 in Fig. 6a does not show the values in the log scale, and the related publications do not describe the maps using the log scale to represent the data (Petersen et al. 2019; Salsabili et al. 2019). The reproducibility of such a graphic or dataset is thus compromised.

A second area of concern is that this important dataset is being visualised with a rainbow-type colour map. Although this map is publicly facing (including that it is freely accessible and from a government source), it is subsequently used to create a simplified map of “relative hazard” as shown in Fig. 13. Such a derivative is highly visible to the public and likely re-used by mainstream media or re-shared on social media platforms, especially in times of earthquake activity. As mentioned in the “Methods” section, the processes involved in creating a seismic hazard map are intricate and complex. However, the information for the public needs to be clear and concise. Figure 13 depicts the dataset distilled into its core message: where are the areas of highest seismic hazard risk? Through analysing the relative hazard data values, we can create Table 2, which shows what spectral acceleration values relate to a high and low seismic hazard. When comparing the original dataset with the jet colour map (Fig. 6a), it is clear that the red colour represents the high hazard (Fig. 13) as it is a value of over 0.85 g. However, the uneven and non-log scale of the colour map intervals makes this interpretation cumbersome.

To highlight the importance of these public-facing maps, the insurance industry uses seismic hazard data to set rates for earthquake damage protection. It is important to understand how insurance companies use seismic hazard values to explain insurance rates for their clients, and how this may be impacted by visualisation methods. Using the Co-Operators as an example (as they are one of the largest insurance companies across Canada), their website directs clients to the simplified relative hazard map, as shown in Fig. 13, to explain the risk (Co-operators 2003; last accessed 15 November 2022). The website indicates that the probability of strong shaking is more than 30 times greater in the regions of the highest hazard than in the area of the lowest hazard (e.g., Table 2).

However, the simplified relative hazard map (Fig. 13) and the original seismic hazard data presented in Fig. 6a pose the same potential issue: the colour map highlighting red in Fig. 6a or Fig. 13 gives the perception that all of the red region is of a similar value. Yet, there is a significant range of values that encompass those red values—in particular from 0.8 g up to 5.0 g in Fig. 6a. A key issue is that in this visualisation method used in Fig. 6a indicates the seismic hazard risk of Vancouver, BC to be similar as Tofino, BC, as they are all in the red region of the colour map (Fig. 13b), despite Tofino having a spectral acceleration 170% higher than Vancouver (Table 3). However, the summary Fig. 13 perceives this hazard to be the same through simplifying the relative hazard. In this case, Graham Island in BC has the same relative hazard as Vancouver (high, Fig. 13b), despite being 412% higher (Table 3). To counter this simplification, Fig. 9 presents the full range of data with an even gradient and equal interval and although not as dramatic as Fig. 6a in showing seismic hazard, the data are not distorted.

This paper does not intend to take aim against the methodology of how seismic hazard maps are created, but rather to highlight the potential pitfalls and potentially unforeseen consequences of presenting data with unscientific colour choices. In terms of seismic hazard maps, the use of misleading colour maps and unequal intervals are not a singular issue for Canada, as highlighted by other countries maps in Fig. 4. Furthermore, despite the efforts of numerous studies and campaigns (e.g., Light and Barltein 2004; Hawkins 2015; Kovesi 2015; Stauffer et al. 2015; Moreland 2016; Thyng et al. 2016; Crameri et al. 2020; Zeller and Rogers 2020; Kaspar and Crameri 2022; Dallo et al. 2024a, 2024b), rainbow/jet colour maps and uneven intervals are prevalent not only in academic research but in important public-facing information (Fig. 5). What is clear from examples such as Figs. 10 and 11, the choice of your colour map will impact the perception of your data.

In addition, the choice of colour map with regards to the background of the image should be taken into consideration, as the contrast between background and foreground information allows data to be prominent (Crameri et al. 2020). For instance, if the figure has light colours as the highest data value in the colour map, then the background of the image should be dark to provide the strongest contrast to the data. Given that most seismic hazard maps are produced with a white or light background (e.g., Fig. 4), then a potential high data value could be a dark red with low values being a lighter colour (e.g., inverted Glasgow/Bilboa from Crameri 2018 or inverted thermal/solar from Thyng et al. 2016).

A number of previous studies have conducted surveys and interviews on the perception of risk given particular colour map choices (Thompson et al. 2015; Miran et al. 2017; Marti et al. 2019; Fyfe and Molnar 2020; Schneider et al. 2023; Dallo et al. 2024a, 2024b). For seismic hazard in Vancouver, Fyfe and Molnar (2020) applied an iterative consultation approach to identify areas of effective hazard communication. Dallo et al. (2024b) outlined a framework for wider consultation, focussing on the Swiss public’s perception of seismic hazard. The work found that the Swiss population prefers a blue–brown–red colour scale and that different social groups perceive the usefulness and relevance of the earthquake risk map differently (e.g., homeowners find it more useful). Given the current use of inaccessible colour maps in seismic hazard (Fig. 4) and the number of previous studies related to visualisation available (Hagemeier-Klose and Wagner 2009; Gaspar-Escribano and Iturrioz 2011; Hawkins 2015; Kovesi 2015; Stauffer et al. 2015; Moreland 2016; Thyng et al. 2016; Crameri et al. 2020; Zeller and Rogers 2020), a next step in this research would be a wider consultation on best practices for Canadian hazard visualisation.

In addition, it will be valuable to understand more concretely how seismic hazard maps are used to calculate earthquake insurance (as highlighted by Dallo et al. 2024b). In knowing how such maps are implemented, it will be important to highlight how visualisation methods could impact public insurance costs. We also propose potential new seismic hazard maps using even gradients between colours and equal intervals (e.g., Fig. 9a) or with a log scale that can capture more coherently the wide range of values within the dataset.

In conclusion, we present an overview of how the choice of colour map can impact the perception of data. Specifically, we take an important public resource in seismic hazard data and show that applying rainbow/jet colour maps are not inclusive and accessible to the wide population (Fig. 11). Furthermore, we highlight that the choice of interval between colours can also impact how hazardous an area is represented and perceived (Fig. 9a). The work here is timely, as the use of rainbow/jet in all fields of Earth Science is still prevalent. We, as a community of Canadian geoscientists, have a choice to be more inclusive in many aspects of Earth Science (e.g., Stokes et al. 2019). However, a first step is to be more active, rather than passive, in the choice of visualisation methods when it comes to presenting our data in publications, at conferences, and to the public.

We would like to thank NRCan for providing the data to conduct this study and in particular Michal Kolaj for initial discussion and support.

Data are publicly available from NRCan https://earthquakescanada.nrcan.gc.ca/hazard-alea/interpolat/index-en.php.

Conceptualization: YY, FC, GES, PJH

Data curation: YY, PJH

Formal analysis: YY, FC, GES, PJH

Funding acquisition: PJH

Investigation: YY, FC, GES, PJH

Methodology: YY, FC, GES, PJH

Project administration: YY, PJH

Software: YY

Supervision: PJH

Validation: YY, FC, GES, PJH

Visualization: YY, FC, GES, PJH

Writing – original draft: YY

Writing – review & editing: YY, FC, GES, PJH

We also acknowledge support from a Natural Sciences and Engineering Research Council of Canada Discovery Grants for PJH (RGPIN-2022-03084). This research was enabled in part by support provided by Compute Ontario (computeontario.ca) and the Digital Research Alliance of Canada (alliancecan.ca) for an allocation to PJH. GES acknowledges support from the Research Council of Norway through its Centers of Excellence funding scheme, Project Number 223272, and through its Young Research Talent scheme for “POLARIS—Evolution of the Arctic in Deep Time”, Project Number 326238.