1 Introduction

In the summer of 2014, attention turned towards the countryside as fire spread over an area of 65 square kilometers of forest in the county of Västmanland in central Sweden. The media closely followed the efforts and experiences of the fire and rescue services (FRS) while this disaster unfolded. It eventually became the largest such disaster in Sweden in recent decades, costing more than one billion Swedish crowns [1]. This example may not be representative of the entire country, but it does illustrate the importance of having a well-coordinated FRS system that is capable of reaching the most remote areas.

While access to emergency services has become an integral part of the United Nations’ Sustainable Development Goals [2], ensuring equitable access to emergency services is still a challenge, not least in the most sparsely populated areas of the globe. Unlike equal access for everyone, equitable access entails that resources are available where and when, and for whom, they are needed [3]. In practice, this means that planning that aims at increasing equality of access to resources is concerned with making resources available for as many as possible, while planning striving for increased equity prioritize those who are more likely to need them—even if it comes at the expense of making resources available for fewer persons in total.

Relatively little is known about the nature of socio-spatial disparities in emergency services, particularly in the Swedish context. International literature shows that the organization of FRS [e.g. 4, 5], remoteness [e.g. 6] and road infrastructure [e.g. 7] influence disparities between estimated and real response times. Such disparities may obscure potential inequities in access to FRS from planners, who are rendered unaware of where resources are needed. It is therefore important to investigate how estimated response times correspond with real response times. If response times are systematically underestimated in some places or depending on some factors, planners need to be aware of this in order to avoid unexpectedly long response times, which might entail increased property loss, suffering and fatalities.

Therefore, the aim of this study is to analyze real and estimated response times of the FRS in Sweden and assess how these correspond, to show where estimations are inaccurately reflecting real response times. For planners, such knowledge can help allocate resources by controlling for potential estimation errors. In extension, this could make the allocation of resources more equitable and efficient. The structure of this article is as follows. First, the relevant literature on FRS response times as an equity issue is presented. The case study and methodology are then outlined, followed by results and discussion. The article concludes with implications for both research and practice.

1.1 Theoretical Background and Previous Studies

1.1.1 Equity in Emergency Services

Our key understanding is that equity is both a process and an outcome. Equity is a way for people to be involved in deciding what they need to achieve as well as the determined outcomes, whether those outcomes are support, resources, or decision-making control. While equality gives everyone the right to access FRS if there is an emergency, equity ensures there are FRS when people need them [8]. Thus, equity is concerned with the unequal treatment of unequals, where the needs of individuals or groups determine how resources are allocated [3].

The key objective when localizing emergency service resources may be thought of as the ability to provide the best possible service to the public [9]. Achieving equitable service is, in a sense, a problem of distributing resources in a way that specifically benefits those who need them. Within strategic planning, the main focus has generally been on choosing sites for stations (ibid.) and vehicles [9, 10]. However, equitable access to services may also be generated inadvertently by large scale infrastructure planning—for example by reductions in service provision. For example, in Rio de Janeiro, Brazil, the transportation network was expanded ahead of the Olympics in 2016. Due to subsequent cuts to service provision, the new infrastructure did not increase accessibility to services as initially planned. Instead, accessibility levels were reduced, penalizing poor population groups in particular [11]. It is important, therefore, to recognize that equitable accessibility to emergency services is not planned separately from other infrastructure projects.

1.1.2 Factors Affecting Response Times

Reducing FRS response times is important as incurred structural damage increase with time. For fires in building, each minute of delay has been shown to increase costs by approximately 2500 € [12], and longer response times are associated with higher risk of fatalities [13]. However, the pace of structural damage may vary by context. In non-residential buildings, for example, the materials inside the building influence the area burned [14]. Likewise, fire spread depends on the type of vegetation. Wildfires are more likely to spread in shrublands and pine forests compared to annual crops and evergreen oak woodlands, although this ‘selectivity’ decreases as fire size increases [15]. Rural households have also been argued to be more vulnerable to wildfires [16]. Considering wildfires are likely to become more common in the future as the climate changes [17], this vulnerability may be exacerbated in rural areas.

Accurate predictions of response times are important for an effective and equitable allocation of resources, especially in rural areas as resources are more scarce there compared to urban areas [18]. Although response times tend to be shorter in urban areas [19], travel speeds to rural fires tend to be higher as speed limits are higher and the road network less complex in rural contexts [18]. Unpredictable factors that are hard to incorporate in calculations, e.g. sudden traffic jams at road crossings, also influence real response times and may generate a mismatch between estimated and real response times [7]. As a result, modelled response times tend to be lower than real response times. For example, Patel et al. [6] found that pre-hospital times for ambulances, i.e. the time elapsed from when an ambulance is requested by a patient, to when the patient arrives at the hospital, were higher in reality—especially in rural areas. Some suggestions to improve the accuracy of estimations include spatial models and spatial survival analysis [20].

The way emergency services are organized has an effect on such services’ performance and quality [see e.g. 21–23]. In many countries, FRS have recently undergone organizational changes, in part with the intention to become more financially efficient in the light of rising service costs. Studies on whether or not collaborations with other public sectors enhance the performance and efficiency of service delivery show mixed results [23]. In Sweden, collaborations between public actors and sectors have increased in the last decades. In practice this has led to using smaller rescue team units, introducing first responders, and increased numbers of formalized cooperative organizations between neighboring municipalities. Despite these changes, Holmgren and Weinholt [22] argue that cost efficiency of FRS in Sweden varies regionally, and found no evidence of improved efficiency on the national level from formalized cooperation. Likewise, Jensen [21], interviewing emergency managers in the United States, found that organizational directives on the national level may not be equally suitable for different regions. Local contexts such as geographic distances and available resources, and specific cultures, require unique structures and responses that are contextually appropriate.

Collaboration between agencies, primarily through information sharing, can increase the capabilities of FRS [4]. Information sharing is a key also to effective FRS response, and unclear roles or lack of cooperation, or unwillingness to share information, are recognized as major barriers to collaboration across agencies [5]. Cooperation between various public actors and agencies is also important for the effect of implemented fire safety interventions, e.g. risk assessments and home visits to certain groups [24]. On the one hand, the unfixed and adaptable nature of the intervention programs allowed for intervention and adjustment to meet contextual challenges. On the other hand, uncertainty in how to implement interventions and how to know ‘best practice’, as well as unclear division of costs for upholding cooperation, is a source of frustration for municipalities working with them [24].

Using technology can help overcome such barriers. For example, deploying drones to fires has been shown to improve context awareness for fire fighters, and to create a feeling of assurance for people in the vicinity [25]. Part-time fire fighters are furthermore often used in areas where the number of incidents is too few to motivate full-time services [22]. Volunteers, too, play an important role in emergency response to fires by filling in when needed and by acting as first responders [21].

1.1.3 Socio-spatial Disparities in Emergency Services

Response times of FRS have been found to be longer in new residential areas outside cities [26]. This is concerning because fire incident rates, potential property losses and human casualties tend to be higher in such areas [26]. Socio-spatial characteristics of areas, too, influence both the risk of fires occurring and the damage they incur. For example, domestic fires in Helsinki have been found to be more prevalent in densely populated areas with low income levels and old buildings [27]. Similar findings are presented by Shai [28], whose results indicate that low income earners and people living in old houses, and particularly the combination of the two, are significantly correlated to fire injuries. The occurrence of fires, and the costs incurred by them, have also been argued to be highest in low-income neighborhoods [29]. In England, educational attainment has been related to an increased risk of casualties in residential fires [30].

The risk of fires thus varies between places, both on the urban–rural scale but also between areas within cities. If the aim of FRS is to reduce the risk of fatalities and the damage incurred by fires and other accidents, the aforementioned studies suggest that planning FRS must be based on where these events are likely to occur, which hinges on both a spatial and socioeconomic factors.

1.2 Contributions

This study has several contributions. More than assessing the spatial patterns of FRS accessibility, we investigate how real and estimated response times relate. We also assess how potential disparities between them varies with difference types of events and in different locations. This is important because estimations may lead to ineffective and inequitable allocation of FRS resources, e.g. stations and vehicles. From an equity perspective, assuring efficient use of resources is particularly important for rural populations as resources are more scarce in rural areas [18]. The study is novel in the sense that we assess how disparities between estimated and real response times is affected by cooperation agreements, motivated by previous research that show mixed findings regarding how cooperative FRS organization impact response times [23]. It also builds on previous Swedish research on individualized interventions to increase fire safety in Sweden, where the decentralized structure of Swedish FRS has been argued to be a barrier to implementation because of difficulties related to information sharing and, relatedly, costs [24].

2 The Study Context

Sweden is one of the largest countries in Europe by area, but with the fourth lowest population density [31]. The country has also has a high level of urbanization: in 2016, 86% of its population was classified as urban-dwelling [32], compared to 73% in Europe [33] and 54% globally [32]. Most people are concentrated in the south. Sweden has 290 municipalities with an average of 32,000 inhabitants, but 140 of those municipalities have a population of less than 15,000 inhabitants and 40 municipalities have more than 70,000 inhabitants. Of Sweden’s nine million residents, about two million live in rural municipalities.

In this paper, the municipalities were categorized as urban, densely populated and rural based on an official Swedish system from The Swedish Association of Local Authorities and Regions (SALAR). The system is based on several criteria, including the size of urban areas, the distance to urban areas and commuting patterns [34]. Municipalities were classified as urban if they had a population larger than 200,000 living in the municipality’s largest city or if at least 40% of the total population commute to a city with at least 200,000 inhabitants; as densely populated if they had a population between 50,000 and 200,000 with 40,000 or more living in the municipality’s largest city and as rural if the largest city in the municipality had a population below 40,000 or if at least 30% of the population commute to work in a densely populated municipality.

Providing an equitable supply of FRS to Swedish citizens can be a challenge as many municipalities are sparsely populated. In Sweden, the organization of FRS is a municipal responsibility [35] and municipalities are autonomous in choosing how to organize them. The response to fires is required to begin within an acceptable time, although the law does not specify or define what this is [22]. Some municipalities provide FRS on their own, although it is common to establish official cooperation agreements across municipal borders. These cooperation schemes are configured by the involved municipalities and are adapted to the specific context. The study area presented in Figure 1 shows types of municipalities (left) as well as all current municipal cooperation groups (right).

Figure 1
figure 1

Sweden, separated by types of municipalities (left), population density (middle) and municipalities in formal cooperation with other municipalities, separated by colors to indicate different organizations (right) (Color figure online)

Full size image

3 Materials and Methods

3.1 Data and Data Preparation

Event reports for municipal FRS were obtained from the Swedish Civil Contingencies Agency [Swe: Myndigheten för Samhällsskydd och Beredskap (MSB)]. All reports for 1 year, 2018, were considered, but false alarms, events lacking position coordinates, and events where the response time was unrealistic (lower than 1.5 min or higher than 60 min) were removed. Furthermore, events on islands and other areas not reachable by a road vehicle were removed. This reduced the total number of analyzed events from 133702 to 61420.

The data in the event reports include, among many things, the event type and time stamps for:

  • Reception of alert, Tstart

  • First vehicle out, Tout

  • First vehicle arrived, Tarrive

The data also include categorization of events, indicating, for example, if an event occurred in a building or if it was a traffic incident. Information about fire station locations and type were obtained from an outdated file from MSB, which was manually updated to mirror the state of autumn 2019 by checking official information on the FRS’ websites. Stations were classified according to Table 1. The call-out time is the time from when the station receives the event notification (the alert), to when the first vehicle starts travelling towards the event. For stations of mixed types (e.g. a full-time station with additional part-time crew, or part-time station with a first response unit), the type with the shortest call-out time is used.

Table 1 Descriptions of Station Types and Associated Call-Out Times
Full size table

The Swedish National Road Database [Swe: Nationella Vägdatabasen (NVDB)] was the basis for the road network, utilizing the full Swedish network with all links containing speed limit data, summing up to 2041416 links, covering more than 708,000,000 m of road.

3.2 Response Time Calculations

The calculations of the estimated response times were done using the Network Analysis extension in ESRI ArcGIS 10.2. The Closest Facility tool, which is based on Dijkstra's algorithm, was used to get the fastest response time to each event. Two different types of response times were calculated:

  1. 1.

    Real response times (Rreal): For each event, the historical response time was calculated as the difference between the time of arrival for the first vehicle and the time the station received the event notification, i.e. Rreal = Tarrive − Tstart

  2. 2.

    Estimated response time (Rplan): For each event, the estimated response time was calculated as the fastest path from a station to the event, considering the call-out time (which was set according to the station type).

3.3 Regression Modelling

A model of average difference (real response time—estimated response time) was run over all analyzed events aggregated per municipality (N = 290). Separate models were run for all events and also for three broad categories of events: nature-related, building-related and traffic-related events. These categories were chosen because response times have been shown to vary due to road network complexity [18, 19]. Sudden events typical in urban areas, such as traffic jams at road crossings, have furthermore been argued to generate a mismatch between estimated and real response times [6, 7]. Moreover, in the network analysis, the emergency is reached by arriving to the closest point to it on the road network. In cases where fires are located in the forest, away from the road network, real response times should be longer than estimated response times as the extra time spent to reach the actual scene, off the road network, is not accounted for in the network analysis.

The explanatory variables were chosen primarily based on previous literature, and secondarily on our hypotheses and on the performance of the models. First, related to the aforementioned complexity of the road network, we included a variable indicating the share of the ground covered by forests, hypothesizing that larger forest coverage should explain greater differences between estimated and real response times. Population density was also added to the model, as this variable has previously been identified as related to higher response times to fires [27]. As we are also interested in assessing the effect of collaboration between municipalities, we also added a dummy variable indicating whether a municipality had any established collaboration agreements or not. Related to this, we also added a variable indicating the investment in FRS to understand if larger investments can explain variations in performance. Dummy variables indicating municipality types were added as control variables.

The modelling strategy started with running regular ordinary least square models. As spatial autocorrelation was present in the models, indicated by positive Moran’s I statistics, we also ran spatial lag and spatial error models to control for potential spatial interactions in the data, which could be caused by clustering of similar values in neighboring areas (lag) or by omitted explanatory variables (error).

4 Results

4.1 Comparing Estimated and Real Response Times

A histogram over the real and the estimated response times is presented in Figure 2. With 60-s bins in the histogram, it is evident that there exist longer response times in reality than what has been planned for. For both types of times, the bin with the largest number of events is the ‘360 s’ bin; for the real response times, 10.6% of the events are included in this bin, while the equivalent for the estimated times is 14.0%.

Figure 2
figure 2

A histogram over the real and the estimated response times

Full size image

Looking further into the difference between the two types of response times, Figure 3 shows a histogram over the time difference for each event, calculated as Rreal − Rplan. The bins with the most events are the’60 s’ and ‘120 s’ bins, with 15.1% of the events in each bin. In a majority (81.3%) of events, the real response time is longer than the estimated, but for over 30% of the events, the real is less than 2 min longer than the estimated response time. Note that the spike at the ‘3600 s’ bin is due to that bin containing all events with a difference between 1380 and 3600 s, while the rest of the bins are 60 s wide.

Figure 3
figure 3

Histogram over the difference between real time and estimated response time

Full size image

4.2 Spatial Patterns of Real and Estimated Response Times

Table 2 shows descriptive statistics for the events, split by the type of municipality they occurred in. The overall median real response time was 486.0 s, compared to the median for estimated response times of 398.4 s; a median difference of 83.6 s. Response times were overestimated, on average, by more than 1 min in all types of municipalities. The average overestimation was 110.6 s in urban municipalities, 74.2 s in densely populated municipalities and 65.9 s in rural municipalities. Roughly 32% of all events occurred in rural municipalities (N = 19804) while nearly 41% occurred in densely populated municipalities (N = 25094) and around 27% occurred in urban municipalities (N = 16520). The standard deviation (SD) was furthermore highest in rural municipalities, and lowest in urban municipalities.

Table 2 Descriptive Statistics for Real and Estimated Response Times in Seconds, and the Difference Between them, Separated by Area Type
Full size table

Accumulating the events into the municipalities where they occurred, we get the maps presented in Figure 4. As expected, both estimated and real times are longer in the northern parts of Sweden. The municipalities with the largest gap between estimated and real response times are, however, spatially dispersed. As Table 2 indicated, the difference between estimated and real response times is relatively high in urban and densely populated municipalities. It is interesting to note that, especially in the north, there are multiple instances of municipalities having much longer differences in response times (dark), just next to municipalities with shorter differences (light).

Figure 4
figure 4

Average real (left) and average estimated (middle) response times as well as average difference between them (right) in seconds for each Swedish municipality

Full size image

4.3 Disparities in Estimated and Real Response Times Across Types of Events

Then, potential disparities in real and estimated response times between different types of events were investigated. The dataset was split into three different categories of events, namely those occurring in Nature (N = 7478), in Buildings (N = 10664) and in Traffic (N = 19777). A Nature event could be, for example, a fire or an emergency occurring in a park or a forest, while a Building event refers to fires or situations taking place at a building. A Traffic event involves vehicles, such as collisions and crashes. Table 3 indicates that median real response times were the longest in Nature events (667.0 s). The median difference between real and estimated was also greater in Nature events (160.7 s) compared to Building events (53.4 s), Traffic events (46.1 s) and for all other events combined (64.0 s). This likely reflects the fact that Nature events often occur off the road network.

Table 3 Descriptive Statistics for Real and Estimate Response Times, and the Difference Between them, Separated by Event Category
Full size table

4.4 Regression Results

We then turned to regression modelling to explain differences in real and estimated response times on the municipal level. The dependent variable was the average difference between the real and the estimated response time for the events in each municipality. Therefore, significant coefficients should be interpreted as explaining greater differences regardless of if they are positive or negative. In other words, the interpretation of the coefficients is thus that positive coefficients indicate that real response times were longer than estimated response times, while negative coefficients indicate that real response times were shorter than the estimated response times. Thus, variables that are statistically significant and have positive coefficients explain underestimated response times compared to real response times, while negative coefficients explain overestimated response times.

First a standard OLS model including all events (Full model) was run, based on all events aggregated to municipalities (N = 290); see Table 4. The residuals of this model exhibited problems with spatial autocorrelation, indicated by a statistically significant Moran’s I test (p = 0.18). Therefore, spatial autoregressive models were run to control for spatial autocorrelation, which improved the performance as indicated by lower AIC scores and higher R2 values compared to the OLS models, suggesting that omitted variables could help explain differences between real and estimated response times (error) and that there are clusters of similar values in neighboring municipalities (lag). The subsequent analysis is focused on the results from the spatial autoregressive models.

Table 4 Regression Output Where the Dependent Variable is the Average Difference Between Estimated and Real Response Times per Municipality
Full size table

Formal cooperation agreements consistently explained underestimations of response times. However, it came out as insignificant in the Nature model. Running separate models for different types of events, some differences in terms of explanatory variables were observed. In nature related events, the share of the municipality being covered by forests explained lower estimated response times compared to real response times, i.e. underestimation of response times. Forest coverage also explained underestimations of response times in the Buildings model.

Compared to urban municipalities, estimations of response times for events inside buildings tend to be longer than real response times in rural municipalities. In traffic related events, both densely populated and rural municipalities explained overestimations of response times, i.e. longer estimations than real response times. Potentially, this finding is related to the fact that traffic related events occur on the road network, meaning that they are easily accessible by FRS vehicles, while e.g. events occurring in nature may be located far away from the road network and thus taking more time to reach. Interestingly, variations in average differences between real and estimated response times were not explained by investments in FRS or population density in any of the models.

5 Discussion

This study has analyzed and compared estimated and real response times of the FRS in Sweden, and investigated some potential explanatory variables for disparities between real and estimated response times. In a majority of events (81.3%) real response times were longer than estimated. Similar to previous findings [6], response times were in other words underestimated in our models. Real and estimated response times were furthermore on average the longest in rural municipalities.

Despite having the longest real and estimated response times, the difference between them was lower in rural areas compared to densely populated and urban municipalities. Estimations thus reflect real response times to a higher degree in rural municipalities, while underestimating to a greater degree in urban areas. This is contrary to previous evidence that underestimations are more common in rural areas [see e.g. 6]. A likely explanation is that unpredictable factors that influence real response times are more common in urban areas. For example, sudden traffic jams and having to slow down at intersections may be more common in urban areas [7]. This is reflected in real response times, but not in our estimations as such factors were not included in the modelling. Travel speeds also tend to be higher in rural contexts because the road network is less complex there compared to urban areas, as Claridge and Spearpoint [18] points out. This, too, would potentially make estimated response times based on speed limits closer to real travel speeds of FRS vehicles in rural areas.

Difference between real and estimated response times also varied between different types of events. The largest average difference, almost 3 min, was for events occurring in nature. In comparison, the differences was less than 1 min in Building and Traffic events. This could depend on a key difference between nature-related events and the other types of events. Nature-related events likely occur in places that are more difficult to access by vehicle, i.e. located away from the road network. Such inaccessibility was not accounted for in the network analysis. Instead, the closest point to the emergency on the road network was assumed to be the end point, i.e. the emergency scene. In reality, this is not necessarily the case. Vehicles may have to travel on smaller, unofficial roads, or completely off road, to reach the emergency scene. This could explain why the forest coverage variable significantly explained underestimation of response times in nature events.

The R2 values were low in the OLS models. Likely, this was because dependent variable was an indicator of over- and underestimation, and not actual response times. The performance of the models was furthermore improved by controlling for spatial autocorrelation suggesting that the data is both spatially clustered and that there are omitted variables that could help explain spatial patterns in the data. One such variable could be type of forest ownership. Industrial, large-scale forestry of pine forests is common in northern Sweden, while small-scale private owners with mixed vegetation are more common in southern Sweden. In the north, fires could potentially be more difficult to reach because of a lack of roads and infrastructure required to reach the fires, in turn forcing emergency vehicles to travel in suboptimal ways to reach the fires and thereby making estimations of response times less accurate. This would also help explain the finding that real response times were longer than estimated in municipalities with higher shares of forest coverage in events occurring in nature, which is common in the northern parts of Sweden. Considering wildfires are more likely to spread in pine forests [15], vegetation coverage could also be investigated as a potential covariate. Mountainous terrain and harsher winter conditions in the north likely also influence real response times but were not accounted for in our models, and should be investigated in future studies.

It has previously been argued that increased numbers of formalized cooperation agreements between Swedish municipalities has not increased cost efficiency in FRS resource utilization [22]. In our regression models, cooperation significantly explained underestimated response times for all event types except nature events. Cooperation, in other words, does not necessarily entail more efficient use of FRS resources in terms of decreasing response times. That said, cooperation agreements could have other benefits that we did not study here. For example, rural FRS tend to rely on part-time fire fighters to a higher degree than urban due to lower numbers of incidents [22]. We can speculate that this lack of full time personnel could make it more difficult for rural FRS to work proactively in different ways to contribute to safety. In contrast, urban FRS organizations with full time fire fighters may have greater possibilities to focus not only on quick response, but also on other ways to contribute to society. Cooperation may help in assuring such work can be carried out also in municipalities where demand is sparser, and temporally fluctuating.

In Sweden, municipalities are free to organize resources as they wish. As Jensen [21] point out, such decentralized responsibility to organize the FRS locally can help solve context specific challenges which cannot be solved by national level directives. However, although municipalities know the context well and face common challenges, cooperation does not appear to improve FRS performance in terms of response times. While we cannot determine the precise reasons for why real response times tend to be longer than estimated response times in municipalities that cooperate, different ways of organizing FRS does seem to influence response times in some way that is not captured by our network analysis. We can, like previous research [5], speculate that this depends on problems with information sharing and unclear roles, which has also been argued to be a barrier to successfully implementing fire safety programs in Sweden [24]. The decentralized Swedish FRS system, where municipalities are free to innovate in order to solve local, context specific challenges, may in this sense simultaneously be a blessing and a curse. The former by allowing innovation and the latter by preventing a general understanding of how best to solve problems faced by the FRS.

These findings have important implications for the management and planning of FRS. While it is obvious that estimations do not mirror real response times perfectly, over- and underestimations of response times vary in quite predictable ways, depending on both the place and the type of event. While both real and estimated response times are comparatively long in rural municipalities, the estimations tend to be more accurate there than in urban municipalities. As such, planning of efficient and equitably accessible FRS faces different challenges in urban and rural municipalities. In urban municipalities, underestimations of response times may generate naïve perceptions about the performance of the FRS. Understanding which factors affect response times, and implementing them in estimations, is therefore a key challenge for urban municipalities. While this is important also in rural municipalities, estimations were found to be comparatively similar to real response times there. Instead, reducing long response times caused to a large degree by greater geographical distances is a key challenge for rural municipalities. Technical innovations such as satellite detection of emerging forest fires could help in doing so by reducing the time it takes before the FRS are alerted, thereby reducing the total response times.

As challenges differ between places and between types of events, planning of FRS may require involvement of different actors depending on the context. For example, in rural areas, technical innovation from private companies and entrepreneurs could have a greater impact on FRS performance not only by assisting in detecting fires, but also by solving problems such as reaching physically inaccessible areas, e.g. in forest areas without road networks. In cities, it could instead be more important to involve traffic planning offices to assure traffic flows can be directed in a way that makes FRS response easier. Moreover, different challenges also entail that fire fighters may require slightly different skills—driving in areas with high traffic flow, for example, is likely a more important skill in urban areas than rural.

5.1 Limitations

This study has several limitations. The Modifiable Areal Unit Problem (MAUP), a problem present in all spatial analysis, should be mentioned. Our system of spatial division inevitably influenced our analysis. Urban and rural differences in under- and overestimations of response times could be different with another system of spatial division. However, in this study, municipality borders were employed because planning of FRS is made on the municipal level. Thus, results may be more useful in practice than other systems of spatial divisions.

Human error may also have affected the response times. In some events, recorded response times could be longer or shorter because of delays in reporting. Some events may also have been categorized incorrectly, because of the difficulty to define some events. For example, a fire may be located both in a building and in the surrounding nature. Defining it as either “Nature” or “Building” may be difficult in such cases. Such difficulties could have resulted in inconsistent categorization, which in extension may have impacted our analysis.

5.2 Future Research Directions

The forest fires of 2014 in Sweden may be viewed as a reminder of the importance of organizing the FRS resources effectively. Future research should investigate how different organizational schemes and cooperation (and information sharing in particular) influence the performance of the FRS. While it was outside of the scope of this study, urban and densely populated municipalities appear to have formalized cooperation agreements to a larger extent than rural (see Figure 1). Why this is, and how it affects practice, should be studied further. Asking questions about how much it costs to establish and uphold cooperative organizations could shed more light on what implications cooperation have on municipalities’ FRS. Urban–rural differences in FRS delivery should also be investigated further as possibilities and problems differ between contexts. Different emergencies, in different contexts, may require different ways of organizing and providing FRS, and thus solutions may be suitable in one context but not in another. Furthermore, case studies similar to ours that include more factors, e.g. forest ownership types and vegetation types, could also further improve the accuracy of future modelling of response times.

6 Conclusions

This study set out to investigate the spatial patterns of estimated and real response times to fires in Sweden, and to assess potential explanatory factors of under- and overestimations of response times. Employing an equity perspective, it also assessed potential disparities in FRS service between urban and rural areas. In a majority of the events, real response times were longer than estimations. These findings provide important insights into how response times of the FRS vary spatially and how explanatory factors for disparities between estimated and real response times may differ depending on the type of emergency, and highlight the importance of being event and area specific when planning for FRS across the country. If planning is based on underestimated response times, there is a risk of ‘false security’ and overconfidence in the ability of the FRS. Apart from rendering municipalities unaware of where resources could be needed, this also entails potential risks for individuals in emergencies who have to wait longer for the FRS to arrive.

Moreover, cooperation is viewed as a key to meet future challenges, including rising demand and costs for the FRS. However, our results show that cooperation between administrative bodies (municipalities) does not necessarily improve the FRS ability to respond to emergencies. There are potentially other benefits of cooperation, and research should investigate what cooperation entails for practice and planning, and what works where. Without proper evidence to back it up, cooperation otherwise risks incurring unnecessary costs and little increase to performance.