High urban flood risk and no shelter access disproportionally impacts vulnerable communities in the USA (2024)

The frequency, intensity, and size of extreme flood events are drastically increasing, bringing about “profound consequences” for the U.S¹. Increased floods associated with climate change could reshape society, destroy homes and infrastructure, and endanger lives². A historic level of flooding and extreme weather was experienced in the U.S. in 2022 and water utilities encountered increasing stress on infrastructure not prepared for the effects of such high-rise floods³. During the summer of 2022, the country underwent four different flooding events qualifying as a 1-in-1000-year rainstorm. These events caused flash flood warnings, mudslides, road closures, and deaths across the U.S. This aligns with a concerning trend of high-impact events. On July 26, in St. Louis, 7.68 inches of rain fell in 6 h, and a total of 9.04 inches fell during a 15-h storm, shattering the previous rainfall record causing immense damage and requiring many emergency rescues⁴. In March 2021, floods in Nashville Tennessee caused a flash flood and 7 fatalities⁵. In August 2017, Hurricane Harvey ravished Houston, Texas resulting in 89 fatalities and damages totaling $126.3 billion⁶. In January and February 2017, California was impacted by a series of floods damaging roads and highways alone totaling over $1.05 billion and at least 5 fatalities⁷. Flooding is not a new hurdle for U.S. communities or global communities. Yet, the challenges associated with a warming climate coupled with urban growth have amplified the magnitude and impact of these events, as well as making them more frequent^8,9. The changing climate and “risky urbanization patterns” are therefore both to blame for the intensified impacts of flooding in urban areas¹⁰.

The changing rainfall coincides with poverty and vulnerability amplifying the impacts of flooding and putting lives in danger. Despite the high risk many socially and physically vulnerable communities face, they are the least likely to benefit from flood protection and risk management, when disasters strike^11,12. Urban populations are rapidly growing with the promise of improved quality of life, and much of this growth includes lower economic portions of society. In rapidly urbanizing watersheds, resource-limited groups are threatened by a greater risk of economic loss and health risks¹³. As a direct result of decades of redlining and environmental racism, poorer communities are often disproportionately located in low-lying, flood-prone neighborhoods with deficient infrastructure^14,15. Even where comprehensive protection is in place, these communities are likely to experience greater impacts during severe events, and would likely be overwhelmed by severe floods¹⁶. Vulnerability to such events is caused by both environmental and social inequalities, reflected in services, infrastructure, and access to resources¹⁷. Natural hazards collide with human vulnerabilities to create human disasters. The intensified risk from flooding stresses the need for strategic responses and the development of more holistic policies for mitigating the impacts of flooding on vulnerable populations in at-risk areas.

Efforts have been undertaken to understand the impact of flooding at the local, national, and continental scales. Previous research has examined the risk of flood for spatially and socially vulnerable populations¹⁸, as well asassessed the health¹⁹ and economic²⁰ impacts of floods on different communities. However, fully defining the impacts of flooding on the vulnerable has been elusive^8,21. A comprehensive understanding of flood impacts should not be confined to their social, spatial, and environmental consequences²¹, but also should incorporate safety and avoidance of risk, particularly for vulnerable communities. A proper response to mounting concerns requires having adequate “access to shelters.” Access, or ease of reaching shelters, is an obstacle for vulnerable populations during disastrous events. Lack of access to shelters amplifies flood risk vulnerability in under-resourced communities, particularly in the urban landscape where resource-limited communities rely on public transport and often reside in areas of higher flood risk². In the U.S., the combination of these factors impedes safety during flood events and exposes vulnerable populations to additional safety risks. Defining how access to shelters due to social vulnerability amplifies the impacts of flood risks creates an opportunity to ultimately address the impacts of urban flooding. Classifying flood risk through the incorporation of access to shelters allows for the tailored design of policies, plans, and budgets in responding to growing urban flood risks.

Fully understanding the urban flood vulnerability has been limited by the complexity of accounting for vulnerability across sectors²². Studying the risk of floods in isolation of access to shelters can turn a natural hazard into a human disaster. The adverse effects of flooding on communities can be mitigated only if we understand spatial and social vulnerabilities and adopt appropriate preparedness strategies for severe events. Previous studies have emphasized this by exploring the suitability and effectiveness of shelter locations^23,24,25,26 and their connection to community resilience^27,28, identifying the optimal siting of shelters for flood evacuation^29,30,31, evaluating response capabilities of shelters³², bringing attention to the low access to shelter for residents of vulnerable housing units³³, exploring community-based solutions for disaster mitigation strategies³⁴, and providing a foundation for understanding the nuances of shelter location, their role in fostering community resilience, and the intricacies of optimal placement for flood evacuations.

The association between disaster management, resilience, and urban planning allows for the formation of a dynamic cycle to target the challenges disasters pose for cities³⁵. Incorporating the access to shelters in identifying priority areas and communities at risk must happen at the beginning of the cycle with an initial assessment of risk and vulnerability. This assessment informs urban planning about areas in need of shelter allocation to strategically design urban environments and build resilience. This community-centric approach contributes to disaster management by reducing casualties and minimizing the impact of flood-related challenges. Communities become empowered and better equipped to handle emergencies. The dynamic nature of this cycle lies in the feedback loop. Continuous assessment of risk and vulnerability from each phase feeds back into the initial assessment, refining future urban planning, resilience-building, and disaster management efforts. With each iteration, the cycle aims for continuous improvement in resilience, preparedness, and response. It encourages adaptive strategies, allowing communities to evolve and become increasingly resilient to flooding and other climate-related threats.

The overarching goal of this study is to integrate access to shelters into flood risk measures to ultimately circumvent natural hazards from turning into human disasters. Our contribution is twofold. First, we examine urban flood risk areas by spatially incorporating access to shelters into flood risk index³⁶ within eight high-risk cities in the U.S. We identify four risk-exposed areas: (i) areas with high flooding risk and high access to shelters (HH), (ii) areas with high flooding risk and low access to shelters (HL), (iii) areas with low flooding risk and high access to shelters (LH), and (iv) areas with low flooding risk and low access to shelters (LL). The terms “high” and “low” refer to values that are respectively above or below the average value of the variable (i.e., flooding risk, access to shelters) in the specific geographic area. This approach is practical for planners and emergency managers as (i) the categorization provides a straightforward and intuitive way to organize and prioritize the risk landscape without the need for complex technical assessments, (ii) by classifying areas based on their risk exposure and access to shelters, decision-makers can allocate resources more efficiently, (iii) the categorization aids in emergency response planning by allowing responders to tailor their strategies to the specific needs of each category, and (iv) the clear and easily understandable information facilitates effective communication among stakeholders. Second, we synthesize the population of each four risk-exposed areas. This delineates the social disparity in equity to be integrated with flood risk, allowing for the identification of priority zones where vulnerable communities with low access to shelters are likely to be impacted by severe floods. The outcomes of this work support planning and policy needs by identifying and prioritizing areas to improve emergency responses and resource allocations. Although it is tested on eight physiographically diverse U.S. cities experiencing the high risk of riverine flood, the framework is expandable to any regions experiencing urban flood³⁷.

Flood risk and access amalgamated

We revisit the risk of flooding by studying FEMA’s Riverine Flood Risk Index and auto access to shelters in tandem. We use the Bivariate Local Indicator of Spatial Autocorrelation (BiLISA) clustering technique to spatially capture and visualize the characteristics of block groups under the concurrent influence of flooding risk and their neighboring block groups’ access to shelters (e.g., educational, community, health, civic, religious centers). Figure1 depicts the results of cluster analysis. Three observations are discerned. First, shelter allocation is not aligned with flood risk. Areas with a low level of access to shelters is frequently bifurcated by risk of flooding. This is indicated by bundles of HL and LL clusters. Dwellers of low access regions experience flood risk disproportionally. A similar conclusion is drawn for areas with high level of access to shelters, which is perhaps less critical than areas with low level of access to shelters. Second, shelters are more accessible in the central business districts and inner-city than edges. Residents of edges are probably affected more than inner-city residents when urban flood strikes. This is a manageable discrimination that necessitates the attention of city officials, planners, and emergency mangers. Third, a noteworthy proportion of block groups fall into the “Not Significant” (NS) cluster category across the study areas; Detroit has the highest share of NS clusters at 91%, followed by Chicago at 82%, while Nashville exhibits the lowest figure at 43%. It is important to highlight that our analysis employed a relatively lenient threshold for statistical significance, set at a 90% significance level. Despite this leniency, the prevalence of NS clusters underscores the importance of examining the distribution of block groups across other cluster categories.

Exploring the distribution of block groups among the other clusters, the spatial heterogeneity in shelter access and flood risk across different urban areas is evident. For HL clusters, San Antonio stands out with the highest proportion of HL clusters at 13%, while Detroit exhibits the lowest percentage at only 2%. For LH clusters, Nashville has the highest share at 26%, whereas Detroit again has the lowest percentage at 2%. For HH clusters, Fresno leads with the highest proportion at 20%, while Detroit has the lowest percentage at 2%. For LL clusters, Fresno also has the highest share at 20%, while Detroit has the lowest figure at 2%. These observations emphasize the spatial heterogeneity in shelter access and flood risk across different urban areas, underscoring the need for tailored disaster preparedness and mitigation strategies. We have documented the detailed proportion of each cluster in Supplementary Table1.

While identifying the proportion of block groups associated with each cluster is undoubtedly a valuable step in understanding the spatial distribution of shelter access and flood risk, it is important to emphasize that the population exposed to each cluster carries greater significance than merely measuring the number of block groups. Two key reasons support this assertion. First, the primary goal of assessing shelter access and flood risk is to safeguard and support human lives during flooding events. Therefore, focusing on the population exposed to each cluster directly addresses the core objective of disaster preparedness and mitigation. Second, disaster response and recovery efforts are resource-intensive efforts. The allocation of resources (e.g.,emergency personnel, supplies, funding) should be based on the potential scale of the impact. It is thenimportant to identify the communities at risk and its characteristics.

Delineating the populations at risk

Clustering regions to identify at risk areas, albeit necessary, is not enough. Natural hazards expose social inequalities with disproportionate negative externalities³⁸. To combat natural hazard-induced social injustice, risk assessments should be accompanied by demographic and socioeconomic synthesis of the population at risk. Here, this is achieved by a detailed assessment of seven socially vulnerable cohorts: (i) disabled individuals, (ii) the elderly, (iii) carless, (iv) low income, (v) Hispanics, (vi) Asians, and (vii) African Americans. We specifically look at the portion of these communities living in flood risk areas per cluster. The statistical breakdown of our visual analysis illustrates the socioeconomic and demographic of individuals residing in each clustered region and areas exposed to flood risk. Figure2 visualizes: (i) the portion of the population living in risky areas (i.e., HL, LH, LL, HH) and non-flood risk areas per city and cohort, (ii) spatial distribution of each cluster per city and the share of population living in each cluster per city, and (iii) break down of the demographic and socioeconomic characteristics of the population living in each cluster per city.

An unsurprising observation is an inequal distribution of population in flood risk areas across cities. It becomes interesting however, when the socially vulnerable cohorts are visualized separately. Cities do not discriminate socially vulnerable cohorts. If a socially vulnerable cohort (e.g., Hispanic, Asian, African-American) has a higher share of population at risk, other socially vulnerable cohorts have a higher share of population at risk as well. It can be ascertained that socially vulnerable cohorts show a clustered pattern. More specifically to our cities, Indianapolis, Nashville, Fresno, and San Antonio are at the top of the inequality chart regardless of the socially vulnerable cohort. Pittsburg, Cincinnati, Chicago, and Detroit are at the bottom.

Another observation is an inequal distribution of socially vulnerable cohorts in flood risk areas within and between cities. Contrary to previous research suggesting African-Americans asthe most population at risk, our analysis, which combines flood risk and access to shelters, identifies Asians and the elderly as the most at-risk groups. Our analysis also expands the knowledge of population at risk beyond ethnicity and synthesizes the carless and disabled population. Both cohorts are exceptionally vulnerable when flood strikes, but have frequently been overlooked in previous research. They are less mobile. Their access to shelters is hampered, even if they reside in an area with a high level of access to shelters. Their population at risk is not marginal in comparison with other cohorts emphasizing the need for careful integration of them in equity analysis. The percentage of the population residing inside and outside of risky areas is documented in Supplementary Table2. A report on the percentage of the population living in different clusters per city is also documented in Supplementary Table3.

A final observation is the positive correlation between the percentage of the population and the percentage of the land areas exposed to the risk of flooding in each city. In Indianapolis, Nashville, Fresno, and San Antonio, 63.2%, 56.7%, 55.2%, and 54.7% of dwellers are in regions with the risk of flooding. This percentage is as low as 9.7% in Detroit, which also had the lowest relative area of flood risk. More detailed information is encapsulated by visualizing the characteristics of each cluster in Fig.2. As depicted, in San Antonio, Nashville, Indianapolis, and Fresno, the HL cluster possesses a higher share of risk-prone lands with 53.0%, 43.4%, 26.1%, and 22.5%, respectively. Cities with a high share of LH lands are Cincinnati, Pittsburgh, Indianapolis, and Fresno with 11.1%, 10.5%, 9.9%, and 9.4%, respectively. Overall, on average, HL with 22.9%, has the greatest spatial distribution. HH, LL, and LH possess 10.3%, 9,5%, and 8.2% of the cities’ areas. This is an indication of low access to shelters in selected urban areas. The percentage of the total population living in each cluster is also demonstrated for each city. In San Antonio, Indianapolis, Nashville, and Cincinnati, the share of the population living in HL areas is higher with 39.2%, 26.7%, 24.7%, and 19.6%. The spatial distribution of each cluster per city and the share of the population living in each cluster is documented in Supplementary Table4.

As of August 28, 2023, FEMA announced the final selection of 149 subapplications, totaling nearly $642.5 million in grant funding. Applications have been chosen to receive a total of $44.36 million in management³⁹. The allocation of these substantial funds is important as it directly impacts the efficacy of flood mitigation strategies across regions. Proper allocation ensures resources are directed toward projects and initiatives most effective in reducing flood risks and protecting communities from recurrent flood damage. Strategic allocation is, therefore, the key to achieving equitable and efficient flood mitigation outcomes, ultimately contributing to the overall resilience of the nation in the face of increasing flood hazards. We elaborate on this with an example from our findings. Chicago and Pittsburg are both grappling with intensifying flooding challenges and arein need of financial aid. Let us consider two contrasting scenarios for budget allocation: a land-centric and a community-centric scenario. A land-centric strategy considers priority areas and the extent of land exposed to high flood risks. This means that Pittsburgh, with a larger flood-prone area, receives a greater share of the funding. A community-centric strategy considers communities at risk. This qualifies Chicago, with larger communities at risk, to receive a greater share of the funding. This example underscores the critical need to analyze urban flood risk by incorporating inequality in access to shelters,aiming for a more equitable budget allocation and enhancedcommunity resilience.

The incorporation of access to shelters into flood risk assessments yields three profound outcomes for planning and management: (i) planners transcend the boundaries of geographic risk to gain a holistic understanding of citizens’ needs and concerns, delving into socioeconomic vulnerabilities that shape their experiences during flood events; (ii) resources are consciously directed toward areas with the most pressing needs, bridging the equity gap that often separates advantaged from disadvantaged communities; and (iii) the consideration of communities at risk and access to shelters becomes a powerful catalyst for initiating dialogues among planners, emergency managers, and policymakers. These dialogues foster collaborative endeavors aimed at developing strategies that comprehensively address risks and vulnerabilities within communities.

It is essential to approach flood risk management strategies with careful consideration to avoid exacerbating existing socioeconomic inequalities. The current national risk assessment method employed by FEMA relies on three components: expected annual loss, social vulnerability, and community resilience⁴⁰. This assessment forms the basis for identifying priority areas susceptible to natural hazards. However, the incorporation of expected annual loss can contribute to existing inequities by prioritizing properties with higher values, often associated with households with higher incomes. Our solution was to continue the assessments after identifying priority areas by identifying at-risk communities and their characteristics. Considering factors such as the household income of communities within each cluster strikes a balance between the imperative of flood mitigation and the overarching goal of reducing socioeconomic disparities, effectively guiding investments and resource allocation. Doing so integrates flood risk management with broader equity considerations, fostering a more inclusive and effective strategy for the future.

Incorporating access to shelters into the natural hazard risk assessment offers two advantages: (i) it leads to the identification of priority areas based not only on their inherent risk but also on their access to emergency shelters and (ii) it extends to the identification of at-risk communities that have limited access to shelters, elevating their significance within the risk assessment. This integration elevates the risk index into a complex interdependent body of information that accounts for the spatial trajectory of (i) flooding risk, (ii) socioeconomic and demographic characteristics of communities, (iii) location of shelters, (iv) travel time and mode of transport, and (v) cumulative access to shelters. This comprehensive knowledge also yields benefits for disaster managers in enhancing both long-term and short-term emergency strategies. Long-term strategies for mitigating flood impact might include establishing more permanent sheltersby prioritizinghigh-risk access areas. Short-term strategies for addressing immediate flooding concerns might involve establishing temporary shelters in high-risk areas during flooding seasons through collaborative efforts. This target can be achieved by close collaboration with federal and local authorities, non-profit disaster relief organizations (e.g., Salvation Army), and private entities willing to volunteer space and resources. Enhancing mobility options to cater to individuals lacking proper means of transport to reach shelters, if evacuation is needed, is another important planning outcome derived from this framework. This might include offering temporary transport services such as on-demand and out-of-station services. Implementing these strategiescan enhancecommunity preparedness and response to floods,thereby protecting lives and reducing the impact during vulnerable times.

The integration of access and risk also has the potential to inform policy-making and urban planning,providing abasis fordeveloping an interactive strategic planning platform. This platform can generate outcomes based on various “what-if” input scenarios. In essence, it offers a dynamic tool that allows stakeholders to explore the potential consequences of different strategies and interventions in a virtual environment. This capability empowers decision-makers to make informed choices, optimize resource allocation, and design policies that are responsive to the unique needs and challenges of their communities. This platform serves as an instrument for exploring how preparedness strategies can contribute to narrowing the disparities in access to shelters between disadvantaged and advantaged communities. It not only fosters transparency but also encourages a collaborative approach among stakeholders, enabling them to collectively shape the future of disaster preparedness and response efforts in their regions. This interactive tool, presented in the form of an interactive map, can also be a practical and informative resource for community members. It increases their awareness of potential risks in their residential locations and facilitates the sharing of findings on precautionary, preventive, and motivational policies, as well as possible solutions for existing problems.

We acknowledge that this framework is not free from limitations and potential biases. One limitation is that we utilized population-weighted centroids of block groups as starting points for travel time calculations and, subsequently, access value determinations. While this method was deemed suitable for our case studies, given their relatively small block group areas, future research could benefit from more precise starting locations to enhance access accuracy. Another limitation is that our analysis solely considered the presence of shelters. In reality, some of these shelters might not be operational during flooding events, or they may lack accessibility for specific populations, such as individuals with disabilities or the elderly. Subsequent studies could incorporate more accurate data concerning shelter operational status and their suitability in accommodating diverse needs during emergencies. This study does not offer a definitive solution. It, however, serves as a foundational stepping stone toward adopting a more comprehensive, multidisciplinary perspective on disaster preparedness. It encourages us to rethink and reimagine risk assessment and mitigation strategies that genuinely account for the distinct challenges faced by different communities. By doing so, we can work toward a future that is not only more equitable but also more resilient in the face of the evolving threats posed by climate change and other related hazards.

Data description

Our multi-faceted framework proposed in the current research utilizes three datasets: (1) FEMA National Risk Index³⁶, (2) National Shelter System Facilities⁴¹, and (3) 2015-2019 5-year American Community Survey (ACS)⁴². We obtained the flooding risk index at the census track geographical level from FEMA National Risk Index. The national risk index defines risk as the potential for negative impacts induced by flooding integrating three components of an expected annual loss, social vulnerability, and community resilience. We extracted the spatial location of the emergency shelters from the National Shelter System Facilities of the Homeland Infrastructure Foundation-Level. This includes educational, community, health, civic, and religious centers. We obtained the demographic and socioeconomic information including ethnicity, level of income, and car ownership as well as the elderly and people with disability population at the block group geographical level from the 2015-2019 5-year ACS.

We augmented our datasets by measuring auto access to shelters at the block group geographical level for 30- and 60-minute travel time thresholds. We calculated it by employing the Cumulative Opportunities Measure^43,44 counting the number of shelters that an individual can reach from a centroid of a block group as formulated in Eq.(1):

Here, \({A}_{i}\) is the access of individuals residing in block group \({i}\) to shelters \(O\) located in block group \(j\). The cost of travel between \(i\) and \(j\) follows a binary function as illustrated in Eq.(2), where \(t\) is a particular travel-time threshold:

Bivariate Local Indicator of Spatial Autocorrelation (BiLISA)

We adopt a bivariate Local Indicator of Spatial Autocorrelation (LISA), also known as bivariate Local Moran’s I, to examine the statistically significant association between the flooding risk index at location \(i\) and the average of the neighboring values for access to shelters for 30-min travel-time threshold (i.e., the spatial lag)⁴⁵. The bivariate LISA clustering is measured by Eq.(3), where \({w}_{{ij}}\) are the elements of the first-order queen contiguity weight matrix, and \({x}_{{i}}\) and \({y}_{j}\) are standardized such that their means are zero and variances equal one.

The significance level of locations is determined through a rigorous conditional permutation method. This method involves conducting permutations for each observation 2000 times. A pseudo p-value is thencalculated for each location, serving as a metric to assess significance at the90% level. What makes this approach particularly robust is the combination of indications of significance with the location of each observation. This combined information allows us to make credible classifications of locations into four distinct categories. The bivariate LISA statistic is expressed as High-High (HH), Low-Low (LL), Low-High (LH), andHigh-Low (HL). The first letter represents the risk of riverine flood in a core area and the second letter is associated with the level of auto access to shelters in the neighboring areas. A High-High (HL) region, for example, displays regions with a “High” flooding risk surrounded by “Low” auto access to shelters. The H element in the HL cluster, however, does not necessarily represent FEMA’s high-risk index. This is attributed to the Local Moran’s statistical assessment where a high value indicates anabove-average risk value in the studied area.

When there is no compelling statistical evidence to support meaningful or non-random patternsin risk and access at a location, the areas are categorized as Not Significant (NS). Locations deemed statistically insignificant are not categorized into any significant spatial cluster or outlier category in the analysis. It is important to note that when we refer to “high” and “low” in this context, we are doing so relative to the mean of the under-consideration variable in the study area.These terms should be viewed not in an absolute sense but as a relative assessment of significance within the dataset.

Studied areas

In recent years, efforts, such as First Street, have created platforms for assessing risk across the U.S. For this analysis, eight cities were selected based on their flood risk according to the First Street Analysis. The cities, listed in Table1, are considered among the top U.S. cities for flooding risk³⁷. All the example cities are dominated by fluvial and pluvial flooding (as opposed to coastal flooding). These cities have been classified based on population and properties at risk, physiographic region, and types of flooding risks.

The flooding risk index data can be accessed through FEMA Data Resources at https://hazards.fema.gov/nri/data-resources. Shelter locations can be accessed through the National Shelter System Facilities provided by Homeland Infrastructure Foundation-Level Data (HIFLD), available at https://hifld-geoplatform.opendata.arcgis.com/datasets/geoplatform::national-shelter-system-facilities/about. Socioeconomic and demographic can be accessed through the American Community Survey (ACS) 2019 5-year estimates at https://data.census.gov/table. Underlying data for graphs and charts are included as an excel file in source data at https://doi.org/10.6084/m9.figshare.24596718⁴⁶.

Authors and Affiliations

1. Department of Geography and Geoinformation Science, George Mason University, Fairfax, VA, 22030, USA

   Alireza Ermagun&Fatemeh Janatabadi

2. Department of Civil and Environmental Engineering, Villanova University, Villanova, PA, 19085, USA

   Virginia Smith

Contributions

A.E.: Conceptualization, Methodology, Formal analysis, Writing—original draft. V.S.: Methodology, Writing—original draft. F.J.: Data curation, Visualization, Writing—review and editing.

Corresponding authors

Correspondence to Alireza Ermagun, Virginia Smith or Fatemeh Janatabadi.

Competing interests

The authors declare no competing interests.

Peer review information

Communications Earth & Environment thanks the anonymous reviewers for their contribution to the peer review of this work. Primary Handling Editors: Niheer Dasandi, Joe Aslin and Aliénor Lavergne. A peer review file is available

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions