Background & Summary

The Mediterranean basin is characterized by a squeezed coastal area with a high concentration of people and assets and by rapid demographic, social, economic as well as environmental change1,2. Between 1960 and 2010, the population of the Mediterranean has doubled from 240 million to 480 million3 and the urban population has increased by 20% (ref. 2). This human pressure is further amplified by international tourism. Around one third of the global tourist arrivals in 2011 have been registered in Mediterranean countries, predominantly along the coast. The number of arrivals is expected to increase further, and could reach 637 million per year by 2025 (ref. 3). The Mediterranean coastal zone is not only increasingly under pressure from local human activities, but also subject to future global environmental change. In particular, sea-level rise and associated hazards4 are expected to have significant impacts in Mediterranean nations during the 21st century57.

To address these pressures and to underpin future coastal management and adaptation policies, such as those included in the Integrated Coastal Zone Management (ICZM) Protocol of the Barcelona Convention8, policy makers and coastal administrations in Mediterranean nations require impact and vulnerability assessments. Such integrated assessments are cross-sectorial studies that require information from various fields and disciplines. A prerequisite for coastal impact assessment and for the planning of appropriate future interventions is the availability of consistent information on the physical, ecological and socio-economic characteristics of the Mediterranean coastal zone. Despite an increasing demand of decision-makers, planners and coastal researchers from various disciplines for such consistent scientific data9,10 there is currently no source of readily available information for the 22 countries that surround the Mediterranean basin. Due to the lack of such data only a limited number of studies exist that have analyzed the impacts of sea-level rise for the entire region10,11. Collecting and organising such data is a challenging task as consistent information on socio-economic and physical characteristics, both on current conditions as well as on future developments, of the coast is needed.

The developed MCD aims to meet these needs through the provision of an open access spatial database, that provides consistent information (in terms of format, resolution, quality, accuracy) for the entire region and that is based on a lean data model. The coastal database contains 160 parameters that characterize the natural and socio-economic systems of the Mediterranean coast. It relies on the structure that was originally designed for the Dynamic Interactive Vulnerability Assessment (DIVA) modeling framework1214, following the concepts described in reference15 and reference16 However, we have downscaled and extended these approaches by introducing spatial coastal assessment units that capture the spatial structure of population, assets and land exposed to coastal hazards. Further we have enhanced the transparency of the process of attribution of spatial data to the segments and coastal assessment units in order to make the database more user friendly. The developed coastal database is intended for use in regional scale analyses and provides a robust basis for all types of comparative coastal studies to future change as it allows results to be comparable across the entire region.

In this data descriptor we describe the generation of coastal segments and associated spatial coastal assessment units; the methods that we have used to attribute around 160 different parameters on current and future conditions in various formats to the coastal units; and the development of consistent data for the Mediterranean coasts.

Methods

Coastal data model

Finding a data model to represent coastal space for vulnerability, impact and adaptation assessment is not a straightforward task as coasts are highly dynamic and complex in terms of process interactions17. This introduces challenges when trying to depict this system into a format that allows spatial information to be stored into a database. A linear representation of the coastal zone has often been used in coastal studies due to the common perception of the coast as a linear boundary between the sea and the land18. The main advantage of a linear data model is its computational efficiency, which is essential for being able to conduct the large number of model runs needed in impact and adaptation assessments as a result of the large number of plausible sea-level rise, socio-economic or adaptation scenarios available. However, there is no direct way to attribute spatial data, such as the number of people living in the low-elevation coastal zone or landuse covering the coastal space, to a linear feature without losing spatial information. Therefore, a main disadvantage of a linear data model is that the explicit spatial structure of the system is lost19. The alternative data model frequently applied in order to preserve coastal spatial information is a grid. The disadvantage of this model is, however, that data volumes are large and computationally expensive for use in impact and adaptation assessments at broader geographic scales.

To address this challenge, we have created a data structure that combines the linear representation of the coast with spatial coastal assessment units that extend inland (Figure 1). Thus, we combined the advantages of a linear data model with a spatial representation of the coastal zone. To create such a data model, the first step was to divide the coast into homogenous segments in terms of impacts, vulnerability and adaptation to sea-level rise. For each segment, we then created a set of spatial coastal assessment units according to administrative boundaries and extent of land up to 20 m of elevation above mean sea level. Our aim was to generate assessment units that will respond uniformly to sea-level rise and can be treated as single units for the purpose of adaptation planning in the future. In the following section, we describe the data model and present the computational data processing that was undertaken for populating the database.

Figure 1
figure 1

Workflow of the data model generation for the Mediterranean coastal database.

Coastal segments

Following the concept of reference15 and reference20, we generated coastal segments of variable length, each segment representing parts of the coast with a uniform vulnerability to sea-level rise at a regional scale. As differences in vulnerability to sea-level rise are driven by variations in both socio-economic as well as geomorphic/physical characteristics of the coastal zone16, we used four parameters covering both these domains in the segmentation decision. Every time one of the four parameters changed in value, a new segment was started (Figure 2).

Figure 2
figure 2

Schematic segmentation procedure for the Mediterranean coastal database.

We employed the coastline of the Global Administrative Areas21 database level 01 as the base layer for the segmentation for the Mediterranean countries (Figure 2) and corrected artefacts related to the format (e.g., “pixelization” of coastline) using a smoothing algorithm (polynomial approximation) and a tolerance of 100 meters.

The first social system parameter included in the segmentation were (1) administrative boundaries. This parameter was included, because society’s capacity to respond to sea-level rise differs across jurisdictions. In this study, we used the GADM level 02 dataset (see Table 1).

Table 1 Summary of the parameter and data included in the Mediterranean coastal database on a segment level.

The second social system parameter we used in the segmentation was the (2) distribution of assets and people distinguishing the coast into two classes, (a) urban and (b) rural. This parameter is relevant for the segmentation, because population and asset density influence vulnerability by both determining the exposure to sea-level rise and storm surges, as well as by influencing society’s’ capacity to adapt13. We classified the coast using satellite imagery and photos from Google Earth and the Moderate Resolution Imaging Spectroradiometer (MODIS) global map of urban extent dataset with a spatial resolution of 500 m22. Classification decisions were based on visual interpretation of Google imagery and on the MODIS data, where urban areas are defined as places with predominantly built environment. That includes all non-vegetative, human-constructed elements, such as buildings, roads, runways and are greater than 1km2 (ref. 22). All pixels with a coverage greater than or equal to 50% built environment according to the MODIS dataset22 were classified as urban. As we were particularly interested in all human settlements that are lying directly on the coast, we refined the classification using Google Earth in order to also include smaller human settlements. Therefore, settlements with a maximum distance of 300 m to the shoreline and a minimum extent of 300 m x 300 m, predominantly covered by residential buildings, were defined as urban. Harbours were excluded from the urban classification, as they will require specific adaptation measures in the future.

The first geomorphological parameter we considered in the segmentation was the (3) coastal material. The material of the coast has a significant impact on the large-scale response to sea-level rise16. One of the major impacts of sea-level rise is long-term erosion and land loss due to permanent inundation12. For instance, sandy beaches will respond differently than rocky coasts to a rising sea level. We created a typology of four different geomorphic classes that respond differently to rising sea level and will require different adaptation measures. For the Mediterranean, no such dataset on coastal morphology and geological characteristics was available at that time. Four different classes, namely (a) sand, (b) unerodible, (c) mud and (d) rock with pocket beaches have been classified based on visual interpretation of Google Earth imagery and location-tagged photographs from the web-service Panoramio which offers geographically tagged photographs from users23. A similar method has been used in reference20 and reference23.

The second geomorphological parameter that we included in the segmentation process is (4) river mouths. This parameter was included because deltas and estuaries are among the most vulnerable coastal geomorphic features to sea-level rise24. We classified 47 river mouths for the Mediterranean region based on Google Earth imagery.

Finally, those layers with the four different parameters were combined to create the coastline segmentation for the Mediterranean. Overall, there are 11,975 segments with an average length of 4.5 km.

Spatial assessment units

The spatial assessment units expanding inland were created by generating inland buffer zones for every coastline segment and overlaying them with elevation data and the administrative boundaries (resolution: 3 arc second). We only account for the low-lying part of the coastal zone that is hydrologically connected to the sea and lies above 20 m from mean sea level. The low-lying part of the coast is particularly at risk due to higher extreme sea levels in the future25,26. We decided to extend the LECZ (Low Elevation Coastal Zone=Area below 10m2729) up to 20 m in order to account for all plausible scenarios of changes in mean sea level and associated hazards, including high-end scenarios, as well as to allow exploring different adaptation strategies such as coastal retreat. Every coastal assessment unit was linked to a coastal segment with a unique identifier code. The coastal segment unique identifier code consists of eight digits. The first three represent the administrative unit and the last five digits represent the coastline segment (Figure 1). The generated coastal data model forms the basis for the subsequent compilation of the database.

Extreme sea levels and waves

The MCD includes two extreme sea level datasets. The first dataset included is derived from the Global Tide and Surge Reanalysis (GTSR) dataset. GTSR is the first near-coast global reanalysis of storm surges and is based on global hydrodynamic modelling combined with meteorological forcing from ERA-Interim (1979–2014). A Gumbel extreme value distribution was fitted to the annual maximum to derive extreme sea levels for various return periods. GTSR generally provides extreme sea levels for the centroids of the coastal segments of the global database used in the DIVA model. However, for the MCD we increased the spatial resolution and saved the outputs for the 11 975 centroids of the Mediterranean coastline segments (GTSR-MED). The general methodology is described in detail in reference30.

The second dataset that we included in the MCD is the DINAS-COAST Extreme Sea Levels (DCESL). This dataset has been the first global extreme water level dataset and was developed with the use of a simple empirical model described in detail in refs 31,32.

Besides the data on extreme sea levels, we have also included information on mean wave heights for the period 1971–2000 for the Mediterranean basin. The wave data have been computed using the wave model WAM33 at a resolution of 0.25 degrees, resolving the spectrum using 12 directions and 25 frequencies. The wind meteorological forcing was generated using the hourly meteorological fields produced by the regional climate model COSMO-CLM at a resolution of 0.12 degrees. The model framework has been validated by reference34 and reference35. These mean wave heights should be considered representative of offshore conditions, before depth induced wave breaking and interaction with the bottom in the near shore zone occur.

Computational data processing

The database was developed with the use of ArcPy, which is a site-package that builds on the ArcGIS scripting module. It enables users to perform geographic spatial data analysis, data conversion and data management with the programming language Python. One main advantage is that every Python script constitutes a precise documentation of the computational data processing that was conducted. The input spatial data were available in various formats depending on the information that they represented. Therefore, we used different methods to attribute the data to the coastline segments and spatial assessment units.

Data processing differed according to two characteristics of the input data, namely: (1) whether the data were originally in vector or raster format; and (2) whether we attributed them to the coastline segment or to the assessment units (e.g. data representing information that extends several kilometres inland). For instance, the extreme sea level data were available in a vector (point) format representing extreme sea levels for different return periods directly at the coast. We spatially joined the nearest extreme sea level data point to the centroid of every segment in order to have a common attribution approach. Every coastal assessment unit obtained the extreme sea level information of the corresponding coastline segment (Figure 1). Another example for the attribution of raster data to the coastal assessment units is the information about the distribution of people in the coastal zone. For this purpose, a gridded population dataset was combined with gridded elevation data in order to calculate the number of people below a certain elevation (in 1 m increments, up to 20 m). Then we calculated the zonal statistics for every coastal assessment unit in order to get the number of people per elevation increment. A detailed documentation for every step employed for attributing the different parameters to the coastal units can be found in the Python scripts (Table 1 and Table 2).

Table 2 Summary of the parameter and data included in the Mediterranean coastal database on at coastal assessment unit level.

The database consists of various parameters about current conditions of the coastal zone. In addition, the database provides information on plausible future changes that will drive future impacts, such as sea-level rise or socio-economic development scenarios. Table 1 and Table 2 summarizes all parameters that are included in the database; their source; a short description and the name of the python code where the detailed computational data processing steps are documented. With the exception for the extreme sea level datasets and the wave data, we only used datasets that are publically available.

Code availability

The python code to populate the Mediterranean coastal database is available to download in the figshare repository (Data Citation 1). The code consists of ArcPy commands, which can be used if an ArcGIS for Desktop license is installed. Each script is internally documented with explanation of the different data processing steps. The internal documentation of the scripts should be used in combination with this manuscript.

Data Records

The developed Mediterranean coastal database described in this article is publicly and freely available through the figshare repository. We have included csv files with all the information on a segment (MCD – coastal segment level) and assessment unit level (MCD – coastal assessment unit level) into the repository. Furthermore, we provide the coastal segments and administrative boundaries in a shapefile format as well as the coastal assessment unit in a tiff format in the repository. The database will be updated and expanded as new and improved data become available.

Technical Validation

The database presented here has been created using a number of publically available datasets, which are thoroughly documented and described in reports or scientific articles (Table 1 and Table 2). Thus, these datasets have undergone rigorous quality controls and/or validation. In addition, for those parameters where consistent information for the entire basin did not exist, new datasets were generated. Technical validation therefore has focused on the evaluation of these new datasets (i.e. the coastal material, GTSR-MED extreme sea levels); and on the correct attribution of data to the assessment units.

Geomorphological classification

The geomorphological classification dataset was compared to the recently compiled geomorphological dataset of the Mediterranean Sea and Coast Foundation (MEDSEA, 2017) which was developed independently, using a similar methodological approach. Further validation for some of the Mediterranean countries, where this was possible, was undertaken based on expert judgement and on national digital datasets of coastal morphology.

Comparing different coastal classification datasets can be a challenging task as the number and type of classes used can differ substantially, depending on factors such as scale or user requirements36,p.267. The MEDSEA data included similar classes to our dataset, namely: (1) sandy beach and beach with uncertain grain size; (2) river, deltas; estuaries and soft sedimentary strands; (3) artificial structures and artificial frontage; (4) soft rock shores; and (5) hard rock shores (see Table 3). 23% of the coast were classified as sandy in both datasets. The MCD class ‘unerodible’ includes rocky coasts and artificial structures. When comparing the MCD class (2) and the MEDSEA classes (3) and (5), the datasets show good agreement, accounting for 46 (MCD) and 40 (MEDSEA) percent of the total coastline. The remaining classes (“rocky with pocket beaches” and “soft rock shores” from MCD and MEDSEA respectively) are not directly comparable. Overall, the two datasets are in agreement for around 70 percent of the Mediterranean coast. The spatial patterns of coastal types are similar in the southern and eastern Mediterranean basin. In the north western part of the basin differences in the classification are visible as the MEDSEA dataset indicates a higher extent of hard and soft rock shores. However, this difference is primarily due to the difference in the definition of the classes for the two classification schemes.

Table 3 Comparison of coastal morphology classification.

Local experts from Spain, Greece and Croatia undertook additional checks, based on visual inspections and on national or local datasets. For Spain, there was agreement for 75% of the coast. For the remaining 25%, we implemented changes based on the expert suggestions and additional checks with Google Earth. For Croatia, we compared our dataset to a national spatial dataset on the distribution of sandy beaches provided by the Ministry of Environmental and Nature Protection and found that all sandy beaches were included. Further qualitative tests based on expert judgement and visual assessment were carried out for Greece for approximately 100 randomly selected segments, indicating an agreement for approximately 85% of the inspected segments. Accordingly, discrepancies were checked and corrected based on Google Earth and expert suggestions. Finally, further comparisons were carried out for the districts of Lazio and Emilia-Romagna with available classifications used in previous analyses20.

It must be noted that despite the extensive evaluation of the geomorphological classification and the agreement with all employed data sources, some errors may still exist. These errors can result from numerous factors, such as differences in the quality of Google Earth imagery for the entire region; differences in scale; errors in the location of Panoramio photographs; coverage of location-tagged photographs varying considerably between the Northern and Southern part of the basin; errors in classification; or differences due to subjectivity in class definition. Nevertheless, the dataset will continue to be updated as new data become available. We are currently exploring options for extending the validation using crowd sourced data from the project Coastwards (http://coastwards.org/) in order improve our current classification.

Extreme sea level datasets

Here we compare the two datasets of extreme sea levels against observations. First, we evaluate GTSR-MED and DCESL which include tides and surges.

Figure 3a shows a comparison of the modelled and observed extreme sea level with a 10-year return period for the GTSR dataset. Extreme sea levels are generally in the range of 0.15 m and 1.24 m. Off the coast of Tunis, the Strait of Gibraltar, and near Venice extreme sea levels are relatively high. This is corresponding with the relatively high tidal range in these areas. For example, the Gulf of Gabes off the coast of Tunisia has a tidal range of nearly two meters, while tides are generally small in other parts of the basin. Reference30 evaluated the GTSR extreme sea levels against observed sea levels from the archive of the University of Hawaii (http://uhslc.soest.hawaii.edu). However, the set of tide gauges used for the global validation has a very limited number of tide gauges available in the Mediterranean basin. Therefore, we performed additional validation using the GESLA-2 dataset (http://gesla.org), which includes data from many more tide gauges37. Here we use the estimates of the return periods of extreme sea levels from reference38. They processed the raw data and fitted a Gumbel distribution to the annual maxima using all stations that contain at least 20 complete years that is less than 25% missing data. For the Mediterranean region, this resulted in 17 observation stations. To evaluate the performance of the GTSR-MED extreme sea levels, we calculate the mean bias and the mean absolute error between the modelled and observed extremes. The modelled extreme sea levels are generally characterized by a negative bias. For a 10-year return period the mean bias is −0.21 m (s.d. 0.20 m), whereas for a 100-year return period the mean bias is −0.34 m (s.d 0.41 m). As extreme sea level are generally below 1.5 m, the relative differences exceeds 25% for a number of locations. This is depicted in Figure 3a-c. The relatively strong negative bias may be due the fact that in semi-enclosed basins, such as the Mediterranean Sea, extreme sea levels are largely controlled by local conditions and mesoscale dynamics. Hence, the representation of the global bathymetry and the resolution of the global tide and surge model may not be sufficient in this region. Moreover, in areas with a complex orography, such as the Adriatic Sea, global climate reanalysis data can have difficulties in reproducing local winds39. Unfortunately, almost all tide gauges are located in Portugal, Spain and Italy. Hence, the eastern and the southern part of the Mediterranean Sea are under–represented. Therefore we were not able to assess the performance of the model for the entire basin.

Figure 3: Performance of the GTSR-MED extreme sea levels for the Mediterranean Sea for the 10-year return period.
figure 3

Observed extreme sea levels are taken from the GESLA dataset. (a) GTSR-MED and GESLA extreme sea levels. (b) The difference of GTSR-MED with GESLA in m. (c) Relative differences of GTSR-MED and GESLA.

The DCESL data have been validated by reference32. The study concluded that the return periods of DCESL are generally too high, compared to observed return periods from GESLA-2. Reference31 compared the GTSR and DCESL against UHSCL observations. They found that both GTSR and DCESL capture the spatial variability of extremes. However, DCESL generally overestimates the extreme sea levels, whereas GTSR generally underestimates the extreme sea levels, but with smaller errors compared to observations.

However, this comparison included only few stations in the Mediterranean. The comparison of GTSR and DCESL by reference38 is based on the GESLA-2 dataset and includes more stations. It shows that DCESL overestimates the 100-year return period values by more than 0.5 m. If we compare the DCESL return periods against the GESLA-2 stations used for the GTSR-MED validation (see above), the mean bias is 0.16 m (s.d. 0.43 m), 0.17 m (s.d. 0.59 m), and 0.20 m (s.d. 0.76 m), respectively for the 10-, 100-, and 1000-year return periods. For specific stations the bias is up to 1.5 m, whereas for other stations the bias of DCESL is less than a few centimetres. Hence, although the performance is variable, DCESL generally also overestimates extreme sea levels in the Mediterranean basin. Again, there is no information to assess DCESL for the eastern and the southern part of the Mediterranean Sea.

We would like to highlight that due to the limited numbers of observations it is difficult to assess the overall performance of the extreme sea level datasets. However, we recommend users that would like to perform an analysis on a regional scale to use the GTSR-MED extreme sea levels as the standard deviation (observed vs modelled) is smaller. Users that are interested in a specific location may be more interested in the DCSEL dataset as the mean bias is smaller than for the GTSR-MED dataset, which indicates that the modelled extreme sea levels may be closer to the observed values in some locations. Users interested in coastal flood risk assessment to extreme sea levels should be aware that both datasets do not include waves. Omitting waves could lead to an underestimation of potential impacts (see reference40).

Data attribution

The attribution of data to the coastal units involved several processing steps, which varied depending on the type of dataset (as documented in the python scripts, see methods section). Every parameter of the database was then checked manually in a Geographic Information System, by at least two different users, to ensure correct attribution. This validation process was introduced to ensure internal consistency and, to identify and correct errors or mismatches in the database compilation process.

Usage Notes

We envisage that academics, managers and planners will use the developed database for coastal applications. We emphasize that the database is designed for regional-scale applications. Caution is required when using the database for local applications.

Additional information

How to cite this article: Wolff C. et al. A Mediterranean coastal database for assessing the impacts of sea-level rise and associated hazards. Sci. Data 5:180044 doi: 10.1038/sdata.2018.44 (2018).

Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.