Introduction

The MI-SAFE software and webviewer are developed under the EU funded FP7 project FAST (Foreshore Assessment using Space Technology). The MI-Safe viewer depicts the estimated contribution of coastal vegetation to wave height attenuation for user-selected coastal locations anywhere in the world. MI-SAFE is thus based on global datasets made available via open spatial services and presented in the web-viewer. The data from these global datasets are used to query the results of a modelling exercise conducted with XBeach for a range of different types of vegetated foreshores and a range of exposure to waves and tides. The results can be used to assess the requirements for the design of coastal flood protection measures with and without vegetation fronting those measures; they give a first estimate of the potential risk reduction that coastal vegetation has to offer with respect to specific coastal flood events on any particular shore across the world. 

For background information about the tool, the project and all the partners involved, please visit our website - http://www.fast-space-project.eu/

MI-SAFE illustrative summary

Figure 1: illustration of the principles of the FAST project (linked to RISC-KIT project)

Expert transect figure with datasources and outputs of the tool (Jasper) 

MI-SAFE data

Elevation

Global SRTM coupled with GEBCO

Coastal and nearshore topography is important factor determining the risk of flooding and thus features highly in the MI-SAFE application. The elevation of the surface over which the tides and waves travel determines how high the water level and the waves will be when they reach the most landward lying natural or artificial barrier. In addition, the type of surface and type of barrier determines how easily it is altered by tides and waves and how likely it is to suffer erosion. Hard rock coasts that rise up from the sea are less likely to suffer erosoin than soft cliffs or sandy coasts. Topography and sediment stability therefore are very important factors. Several datasets are identified as useful for the characterisation of topography in FAST, these are:

Topography is an important element of the risk of flooding and thus in the MI-SAFE application. Hard rock coasts that rise up from the sea are less vulnerable for flooding than soft sloping sandy coasts. Topography therefore is a very important factor. Several datasets are identified as useful for FAST, these are:

-       SRTM Topography (http://srtm.usgs.gov/)

-       GEBCO Bathymetry (http://www.gebco.net/data_and_products/gridded_bathymetry_data/)

-       ASTER (http://asterweb.jpl.nasa.gov/gdem.asp)

For the global version of the MI-SAFE toolbox a derivative product called SRTM15_plus is used. It offers a continuous global coverage of bathymetry and topography. The SRTM15_plus dataset is created by the Scripps Institution of Oceanography (http://topex.ucsd.edu/index.html). This dataset is mainly used for viewing purposes since it is a continuous dataset at global level. For use in the MI-SAFE application a more detailed dataset is created using the SRTM3 v4 and GEBCO data (see Figure 2, vertical accuracy is about 15 m). The finest resolution of approximately 90 meter is combined where GEBCO is rescaled to 90 meter and interpolated to the SRTM tiles. To reduce computation time this is done for tiles along the OSM shoreline of the coast.

Figure 2: Combined srtm + GEBCO elevation worldwide. 

Inter-tidal elevation (satellite-derived)

Precise elevation of the intertidal (foreshore) is crucial in the performance of the MI-SAFE tool, because this is the area where the actual wave attenuation takes place. 

Whilst in theory elevation and bathymetry data sources can be combined (see Elevation paragraph above) to create a continuous DTM that seamlessly covers intertidal regions, in practice a number of issues presently limit the usefulness of this approach:

• Intertidal regions are often missing from both land and seabed elevation data sets.
• Spatial and vertical resolutions may be below the minimum requirements of MI-SAFE (i.e. vertical accuracy of about 1 m).
• Most of the world is ‘data-poor’; meaning lower quality and reduced open-access to elevation data.

EU/ESA’s Copernicus, high resolution Sentinels (S1 and S2) and the NASA/USGS Landsat missions can help to fill this small, yet crucial gap. Extended temporal coverage means that a large number of images with different tidal elevations are available for most coastal regions. Identification of surface water (using indices and/or classification) from this large collection of images allows composite, time-ensemble averaged (TEA) images of the probability of inundation to be built up in tidal coastal zones. With calibration, these can be converted into inter-tidal elevation maps (Mason et al. 1995; Niedermeier, Hoja, and Lehner 2005; Murray et al. 2012). However, this technique is usually applied to relatively small regions and often includes in situ calibration via field measurements of intertidal elevation; hence the challenge for MI-SAFE was how to convert TEA inundation probability to elevation at a global scale without principally relying on field data?

To solve this challenge, as a first approximation, we assumed that:

• TEA water indices can be transformed to inundation probability [0,1] by normalising by the spatially-averaged, TEA water indices of regions identified as land and water, respectively.
• Once normalised, for a single pixel in an inter-tidal zone, the TEA inundation probability represents the long-term average of tidal height.
• If the inundation probability is derived from a collection of images that span a period of time similar to the tidal epoch, i.e., the time period over which tidal height statistics are derived (commonly 19 years), then pixels with a probability of 1 represent permanent water and have elevations less than or equal to the lowest astronomical tide (LAT), whereas land (p = 0) represents elevations more than or equal to the the highest astronomical tide (HAT). By deduction, p = 0.5 is then equivalent to mean sea level (MSL).

Development and production of inter-tidal elevation maps covering most of the global coast were carried out using the Google Earth Engine (GEE) computing platform; which facilitates rapid development and testing. The specific Open Access (OA) data sets used for product development are shown in Table 1.

Table 1. Earth observation and ancillary data sets used in this study. Abbreviations: United States Geological Service (USGS), European Space Agency (ESA), European Union (EU), Centre national d’études spatiales (CNES). Note pixel resolution is that used for the final product; other resolutions maybe available.

Tidal statistics were derived from the GLOBAL TIDE - FES2012 model. Specifically, GeoTIFF’s with global coverage of mean LAT and HAT between 2005 and 2025 were derived from model simulations. These were added as “assets” to the Google Earth Engine platform for use in automated transformation of TEA water index images to inter-tidal elevation (Table 1).

For the purpose of validation and accuracy assessment we used insitu elevation datasets collected using dGPS at each of the FAST case study sites and high resolution digital elevation models (Table 2).

For the UK and NL sites, OA, quality Digital Terrain Models (DTM) generated by local authorities were available. For the Spanish FAST case study sites, available DTMs were deemed unsatisfactory (they had poor coverage in inter-tidal areas) hence a Unmanned Aerial Vehicle (UAV) flight from the Univ. Cadiz central services was commissioned and a high-resolution digital surface model was derived using structure-from-motion techniques for site ES_2.

Table 2. Insitu elevation data sets used in this study. Abbreviations: Foreshore Assessment using Space Technology (FAST), Univ. Cádiz (UCA), Univ. Cambridge (UCAM, Netherlands Institute for Sea Research (NIOZ), Actueel Hoogte Bestand Nederland (AHN), unmanned aerial vehicle (UAV), structure-from-motion (SfM). Note pixel resolution is that of the original data set; digital terrain models were bilinear re-sampled to 20 m.

Work flow protocol and algorithms

The initial processing of optical images includes the transformation of at-sensor radiance to geometrically corrected (orthorectified) top-of-atmosphere reflectance, resulting in a level 1 product suitable for public use (Figure 2). During this process quality assessment (QA) and including per-pixel classification, such as clouds, may also be carried out; usually by the data providers i.e., USGS and Copernicus and provided as a QA band. Top-of rather than bottom-of (or surface) -atmosphere reflectance was used because more TOA scenes were available, and issues were observed with the standard atmospheric correction process in coastal regions.

Further pre-processing included filtering for scenes that intersected an area of interest (AOI, see below), were collected within a specific time range (1997-01-01 to 2017-01-01), and excluding scenes with a metadata cloud cover value > 30 %. A relative cloud score was calculated using band combinations for each pixel of every image; pixels were set as cloud if they had a value > 90% of the time-averaged cloud score (per pixel) of the stack of images. When available, such as Landsat 8 OLI and Sentinel 2 MSI, pixels set as cloud, cirrus and shadows in the QA bands were also added to the mask, which was finally expanded by 3 pixels, before being applied to the image.

A number of approaches are available for mapping surface water using optical imagery; ranging from expert systems and evidential reasoning (Pekel et al. 2016) to thresholding based on normalised difference spectral indices (NDSI) (Boschetti et al. 2014; Donchyts et al. 2016). As a first approximation, in part because of the relatively low computational costs, we used a single NDSI, the modified normalised difference water index (MNDWI) (Xu 2006), which has been shown to be slightly more robust in vegetated systems (Boschetti et al. 2014); (SWIR1 - Green) /  (SWIR1 + Green),    where, SWIR1 and Green are the reflectance of the short-wave infra-red (~ 1600 nm) and green (~ 560 nm) bands, respectively. MNDWI values of land and water tend to be > and < 0, respectively, and were calculated for every non-masked pixel of every scene within the filtered image collection.

Considering each scene as a spatially-explicit measurement of tidal water inundation sampled from the tidal dynamics of that particular region, the temporal-ensemble of pixels at each point in space, is potentially a statistical representation of relative tidal heights. Thus, stacks of scenes were reduced to a single image by calculating the time-ensemble average of each pixel. Land and water ‘end points’ were automatically extracted from the histogram of mean MNDWI values and used to interpolate MNDWI values to inundation probability [0, 1] (with values outside the range of the end points masked). Of the remaining pixels values between 1 and 99% were retained, logit transformed and interpolated to the LAT (p = 0.99) and HAT (p = 0.01) values (m) relative to mean sea level extracted for the particular AOI from the global raster of tidal statistics.

Application to the global coastline

The global coastline was divided into ~ 25000 AOI of about 40x40 km² using the OSM shoreline of the coast. AOI in polar regions were removed, as the technique is not suitable in areas with semi-permanent ice cover. Per AOI, all available scenes within the specified time range from the combined Landsat and Sentinel 2 collections were selected, filtered and processed to derive a single inter-tidal elevation (m, MSL) raster image. The median number of scenes per AOI was 315 (range: 2 to 2223). AOI with a low number of scenes were removed from consideration. As ancillary data, we also exported LAT/HAT, the histogram used for rescaling, the timestamps of the images used and a polygon-vector representing land (defined as elevation > MSL).

Product specifications

• Inter-tidal elevation: Each pixel represents a floating point value of time-averaged (1997 - 2017) surface elevation (m, relative to MSL) in the inter-tidal zone (between LAT and HAT).
• Land mask: This multi-polygon represents land which is inundated less than 50% of the time-period (1997 - 2017). In open coastal zones this is usually land with an elevation higher than MSL (although this is not always the case).
• Scenes were selected between 1997-01-01 and 2017-01-01; however the number of images, and their periodicity varied per AOI (median collection timestamp per AOI was 2010-02-25).
• The inter-tidal elevation raster product has a pixel (geometric) resolution of 20 m, however the input collections, Sentinel 2 and Landsat, have pixel resolutions of 20 and 30 m, respectively. Considering the variability in the number and periodicity of scenes, as well as per-pixel masking, every AOI potentially has a slightly different nominal resolution, however this must be between 20 and 30 m.
• The data products are produced as GeoTIFF raster files of intertidal elevation, GeoJSON multi-polygon geometry files of land-mask’s and CSV files containing ancillary data (per AOI).
• All geospatial products have a Web Mercator (EPSG:3857) datum/projection.

Quality, validity and accuracy assessment

The initial product generation resulted in a proportion (~ 15%) of AOIs that failed during the processing. For many of these there was an obvious explanation; they tended to be AOIs with no shoreline, either offshore or inland, or areas with no LAT/HAT predictions (such as the Caspian Sea). In general the product appeared to represent plausible inter-tidal topography in a number of tidal regions. Although, visual differences between adjacent AOIs were regularly observed, highlighting potential issues with using the self-contained AOI approach (Figure 3).

Figure 3. Map showing inter-tidal elevation of Ria Formosa, Portugal (AOI 164_030, 164_035 and 164_037, lat:36.98, long:-7.80).

Indeed, erroneous results were systematically observed in a range shorelines, mainly related to ‘false-positives’ in terms of identifying open water in the image collections. For example, volcanic substrates, which have MNDWI values similar to water, resulted in the false assignment of inter-tidal elevation. The same type of effect was also caused by snow cover and shadows from terrain. Errors of omission, e.g., known tidal flats not represented by the product, were also observed, for example the Wadden Sea, NL. Here it is not directly clear why the technique failed, although the dynamic nature of the tidal flats might an issue.

Quantitative validation and accuracy assessment was carried out by comparing predicted inter-tidal elevation values to in-situ data sources (Table 2) using the coefficient of determination (R²), root-mean squared error (RMSE) and mean absolute error (MAE). In-situ elevation data was supplied relative to an appropriate local vertical datum, roughly equivalent to ‘m above MSL’ at each site. For the ES, NL and UK sites, this was based on a local tidal station (Cadiz III), Normaal Amsterdams Peil (NAP) and Ordnance Datum Newlyn (ODN), respectively. Insitu dGPS measurements from both case study sites in each country (apart from micro-tidal RO) were combined with digital terrain model (DTM) data, bi-linear re-sampled to 20m pixels, to allow an overall comparison of the accuracy of predictions (Figure 4).

Figure 4. Scatterplot showing observed versus predicted inter-tidal elevation (m, MSL) at the case study sites. A) UK_1 and UK_2 on the east coast of the UK, B)  NL_1 and NL_2 in the Westerschelde, SW Netherlands, C) ES_1 and ES_2 in Cádiz Bay, SW Spain. Green and blue points represent dGPS measurements (vertical accuracy of ± 0.02 m) from case study sites 1 and 2, respectively. Grey points are data derived from a high resolution DTMs (see Table 2), bilinear re-sampled to 20m pixels. Dashed lines represent LAT and HAT values for the region. Solid line is the 1:1 relationship. Statistical measures of ‘goodness-of-fit’ are also shown.

Coefficient’s of determination for a linear relationship between predicted and observed elevation ranged from 0.45 to 0.75, suggesting a reasonable fit. Examination of the scatter plots hinted at some non-linearity in the relationship towards the extremes of the inter-tidal ranges. RMSE and MAE values, estimates of the accuracy of predictions, ranged 0.32 to 0.92 m, suggesting predictions were generally within 1 m of the observed value. Particularly at the UK sites, which are separated by large distances, this maybe improved by further adjustments of the insitu data for ODN to MSL bias. To put in context the preliminary accuracy of the product, DTMs derived from Lidar often have a stated vertical accuracy of +- 0.2 m, whereas global DTMs tend to be in the 5-20 m range. Hence, this global 20m pixel inter-tidal elevation product potential has accuracy between these 2 sources.

Waves

Field measurements 

To acquire the results of the modelling exercise conducted with XBeach for a range of different types of vegetated foreshores and a range of exposure to waves and tides, the XBeach model was validated and calibrated against detailed water level data recorded using high frequency (4 Hz) dynamic water pressure measurements to resolve even small (2 Hz frequency) waves through the use of bed-mounted pressure sensors at a series of eight vegetated foreshores in Europe, ranging from reed beds in Romania (outer Danube Delta) to Sarcocornia salt marsh (Bay of Cadiz), and NW European estuarine and open coast salt marsh (The Netherlands and United Kingdom respectively). Data recording took place with a telemetered data logging system that captured waves and water levels during almost every inundation of the vegetated foreshores from early autumn to late spring over one year. The Romanian field sites Jurilovca (Razelm) and Histria (Sinoe) experienced continuous inundation, albeit with varying water levels, such that data acquisition at these sites varied from that at the United Kingdom, Netherlands, and Spanish sites with wave records triggered every 8 hours (three times per day). Water pressure records acquired in this way were processed into water depths, wave spectra, and summary wave statistics by the University of Cambridge, using tried and tested programming routines (Möller et al. 1996).

Data and metadata for all the calibration sites of the FAST project is available at: …..

Era interim, translated to nearshore depth limited waves  (why, how and where with links to deliverables/products) (Kees)

• era interim wave return periods

Wave data for return periods (T_r) of 1, 10 and 50 years are extracted from the ERA-interim dataset. Waves are generally lower in the tropics (e.g. less than 2 m between -30° to + 30°) than in temperate zones (where H_s>5m between 40°-70°). A table with worldwide mean and median values of H_S and T_p, extracted from the ERA-interim dataset for return periods of 1, 10 and 50 years is given below:

The picture below gives a global view of wave heights for a return period of 1 year.

As the ERA-Interim wave data is based on offshore characteristics, waves are translated to onshore conditions by comparing the waves from ERA-Interim with a depth limited wave. If the depth limited wave height is smaller, this wave height is used for calculations whereas the period is maintained. The wave direction is not taken into account explicitly; the wave direction is assumed to be coast-normal, as it would be during a worst-case-scenario storm, for which an assessment of the protection function of the foreshore is likely to be most important. 

Water levels 

Wave attenuation over foreshores is typically most relevant during storm conditions that create a surge (water level set-up) in combination with high tidal levels. For the MI-Safe tool, a water level that has a probability of occurrence of 10%, i.e. once in 10 years, is considered to be the most relevant: This represents a storm that is both frequent enough to be relevant to users’ needs for planning coastal protection (a 1/100 or 1/1000 year condition may seem too extreme) and severe enough to be a serious threat to coastal regions. The representative water levels or hydraulic boundary conditions are derived from a global D-Flow Flexible Mesh model (Muis et al., 2016) that includes tides, storms and hurricanes. The output of this model is mapped to DIVA segments, so local anomalies can occur for coasts with irregular shapes (bays, estuaries). More extreme or locally tailored conditions can be studied using the more advanced versions of the MI-SAFE tool, which can take into account hydraulic boundary conditions that are specified by users -e.g. in local coastal management guidelines- or derived from dedicated modelling or field observations at the relevant location.

Vegetation

Field data  

To acquire the results of the modelling exercise conducted with XBeach for a range of different types of vegetated foreshores (and a range of exposure to waves and tides), detailed information was obtained on the vegetation of all the vegetated foreshores at which the XBeach model was tested (see above). For each of the eight sites (two each in Romania, Spain, The Netherlands, and the UK), vegetation species richness, percentage cover, and height was recorded seasonally in 1 x 1 m² survey plots over a period of one year. In addition, a ‘photoframe’ was used to  capture the density of the vegetation layer as seen from the side (and as experienced by water flowing over the surface). The vegetation photographed in this way also harvested (an area of 20 cm x 60 cm) to determine its biomass and, where appropriate (grass species), the number of stems per unit surface area. At the Romanian sites, tall reed was present and a sub-sample of stems was harvested from each survey location instead of using the photo frame area for harvesting.

Data and metadata for all the calibration sites in the FAST project is available at:……

Earth Observation products of vegetation  

We provide the following vegetation products derived from Earth Observation: vegetation presence/absence (a global product), vegetation type (a product on European scale), and biophysical characteristics of the vegetation (examples for the case study sites).

The latter include NDVI, the normalised difference vegetation index, of both the saltmarsh and the adjacent mudflat and waters. Values range from -1 to 1. On the saltmarsh, higher values indicate larger biomass, density and health of the saltmarsh, and different vegetation types. On the mudflat, higher values indicate higher biomass of microphytobenthos biomass, higher biomass, density and health or different types of macroalgae and seagrass. Negative values typically indicate water. Below are examples for NDVI derived from atmospherically corrected Sentinel-2 MSI images at case study sites in the Netherlands (Paulina) and Romania (Jurilovca). In these products, land is masked. Spatial resolution of the product is 10 m.

A core biophysical variable for MI-SAFE is Leaf Area Index of the marsh, derived from Sentinel-2 MSI level 2 products. The product as implemented here refers to Leaf Area Index (i.e., green leaf area per unit ground area) of the marsh only. Areas outside the saltmarsh, either subtidal area or emerged tidal flat are set to 0, and land is masked. Below are examples for Leaf Area Index of the marsh at case study sites in the Netherlands (Paulina) and Romania (Jurilovca). Spatial resolution of the product is 10m. In MI-SAFE, Leaf Area Index of the marsh is used to calculate wave attenuation on the marsh.

Yes/no vegetation map  (why, how and where with links to deliverables/products) (Josh/Ebi)

• global landcover
• global distribution of mangroves

MI-SAFE services

MI-SAFE provides detailed information (10-20 m resolution) for case study sites. For other foreshore areas in the world MI-Safe reverts to the global datasets mentioned above under 'MI-SAFE data'.

Casting a transect in MI-SAFE

When a user queries a location in the MI-SAFE tool combines data from four parameters to assess the effect of the foreshore on wave attenuation (See Figure Below):

1)      The local (sea)bed level;

2)      The local water level, including tides and storm surge;

3)      The local wave conditions;

4)      The local vegetation type and cover.

To acquire these data for a queried location (point), the tool first determines a transect perpendicular to the nearest coastline. This transect runs 1000 m inland of MSL and 1000 m seaward; the area of interest where most wave attenuation occurs. If relevant vegetation is encountered along this transect, the upper indicator on the left of the screen turns green; if not, red. The properties encountered along this transect are used to find the nearest match in a table that contains wave attenuation results for many thousands of combinations of conditions. The wave attenuation thus obtained is compared to the wave attenuation over a similar but bare transect. If the difference is considerable, i.e. more than the average in this dataset, the lower green indicator turns green too. If the difference is small, the lower indicator displays a red cross.

Surge levels used for the MI-SAFE Expert tool 

For the Expert version of the MI-Safe tool, locally derived hydraulic boundary conditions have been used wherever available (NL and UK sites) to have the closest resemblance with actual flood defense design conditions. Elsewhere, the same global D-Flow Flexible Mesh model (Muis et al., 2016) as used for the Educational version is used to derive representative water levels, but for a return period of 1/100 years instead of 1/10 years because the former is closer to realistic design values. For more advanced studies using MI-Safe, it is possible to derive localized hydraulic boundary conditions via a combination of hydraulic modelling and historical analysis in case these conditions are not known.

Wave conditions used for the MI-SAFE Expert tool (Iris/Kees, perhaps you can still use some of text below?)

For waves, a similar reasoning as for surge was followed, leading to the selection of a 1 in 10 year wave height and period. These representative waves are derived from the ERA-interim 40-year reanalysis (ECWMF, 2014).

Schematisation of vegetation types for the MI-SAFE Expert tool (Iris, perhaps you can still use some of text below?).

The MI-SAFE tool needs to function without (yet) having the detailed vegetation properties that are to be determined from EO data. Therefore, where possible, vegetation types are derived from existing maps such as the Corine Land Cover (CLC) 2012 and Globcover maps and the characteristics of these vegetation types are based on published data, including that derived from the FAST project calibration sites. End-users wishing to improve on the model calibration / drag that we have developed in the FAST project, should contact their respective country partner (who could then contact UCam if needed) to discuss implementation of wave measurements at their field site to allow special calibration of the MI-SAFE tool for their specific site.

General approach (move this to wave model/ XBeach part?)

XBeach requires four parameters to represent the presence of vegetation:

1)      Length or height h (m);

2)      width or diameter d (m);

3)      number of stems per horizontal area n (m^-2);

4)      drag coeffient C_D (-).

Additionally, these can be varied over any number of layers over the vertical to represent plants with a complex morphology (van Rooijen et al. 2016). Worth noticing is that from these parameters, XBeach calculates a ‘vegetation factor’ that is a multiplication of diameter, number and C_d. As a consequence, 200 stems of 1 mm diameter have the same effect as 100 stems of 2 mm diameter.

For deriving representative properties, three principles were followed:

1)      The vegetation factor should be relatively conservative, so as not to give an overly optimistic estimate of wave attenuation. Thus, plant dimensions are chosen with winter conditions and relatively small individuals in mind. The choice of the drag coefficient beforehand is troublesome, because this not only depends on the plant properties but also on the hydrodynamic conditions. Therefore, a relatively conservative estimate is made with large waves (that give large Reynolds/Keulegan-Carpenter numbers that are associated with low drag coefficient values) and the flexibility of the vegetation in mind. This will be refined once more reliable drag coefficient estimators are available, e.g. based on observations from the FAST field sites.

2)      The vegetation factor should be representative for all occurrences of a particular type across Europe, not for a specific site.

3)      The vegetation factor should be large enough and differ enough between vegetation types to meaningfully differentiate the effects of different vegetation covers from each other.

Note that these basic assumptions are used if only global information is available in the Educational version of the tool at present. In the Expert version, the local EO data will be used to derive vegetation properties.

Intertidal vegetation (salt marshes)

Intertidal vegetation is derived from the CLC class Salt marshes (421): areas submerged by high tides where vegetation dominates. The cover of such marshes can vary considerably between locations and throughout the year. For example, Spartina spp  stands can well be 70 cm high during summer, whereas Salicornia spp.  can be nearly absent in winter. The properties as observed by Möller et al (2014) were selected as representative (Table 1), because they are for a mixed marsh typical for North-Western Europe and because they were measured with wave attenuation studies in mind rather than just observing biomass. Moreover, the drag coefficient of that particular vegetation cover has been derived from large-scale flume experiments under (near) real world wave conditions.

Inland marsh (reed beds)

Reed beds are of interest for inland locations such as lakes that are often fringed by reed beds, and for the Romanian study sites of FAST where large reeds grow in coastal lagoons. The CLC map does not account for reed beds as such, but are based on class 411 Inland marsh. As a result, the MI-SAFE tool will apply reed bed properties to every inland marsh, also the ones predominantly made up of lower shrubs, and might overestimate the wave attenuation capacity of such marshes.

For the moment, the properties of the reed beds at the Romanian sites (Table 1) have been used because these are readily available and because they represent the situation at the study site, which enables a comparison with local observations of wave attenuation. It should be noted that these are possibly taller reeds than usually found along lake banks.

Riparian willow forests (broadleaved forest)

All areas classified by CLC as ‘broad leaved deciduous forests’ (class 311) are considered to be willow forests if they are located in Europe, where no mangroves occur. The CLC class also contains many forests far from any large water body, but these are not relevant as they will not be queried by users of the MI-Safe tool. Forests hardly occur directly adjacent to the coastline, and if they do they are usually coniferous so they will not lead to a false positive identification as willow forest. As a consequence, riparian forests are the most likely forests to be identified in this class.

The composition, and consequently tree size, of riparian forests differs considerably among floodplains but willows are very common in European floodplains. The age and size of willows depends strongly on the management of floodplains: in natural rivers they are older and taller than along strictly managed rivers, where they can be cut regularly to prevent flooding as a result of the additional hydraulic resistance they cause. Such managed areas are more likely to be of interest, and the MI-Safe tool should not overestimate the wave attenuating effect. Therefore, the willow dimensions are chosen to be representative of relatively young, regularly trimmed trees. Data are available from sites in the Netherlands: commonly found regularly cut pollard willows (‘knotwilgen’) of several years old and young willows (less than 1 year old) of a field especially planted for wave attenuation in front of a levee near Fort Steurgat, Werkendam. 

The wave model: XBeach

In order to quantify wave attenuation by vegetation for a given salt marsh or mangrove coastline, MI-Safe uses the numerical modeling software XBeach (van Rooijen et al., 2016) . Xbeach is a depth-averaged, two-dimensional process-based model that solves the time dependent short wave action balance for the entire wave group, suitable for simulating wave attenuation over foreshores. XBeach has three wave energy dissipation processes relevant for MI-Safe simulations: dissipation due to (depth-induced) wave breaking, dissipation due to bottom friction and dissipation due to vegetation.XBeach also has three simulation modes, from simple to advanced: stationary, surfbeat and non-hydrostatic. The stationary mode is fast but lacks wave groups (surfbeat) that are important for wave height variations near shore. The non-hydrostatic mode is physically the most complete but at substantial computational cost. The surfbeat mode does represent the effects of wave groups at reasonable computational cost and represents the effects of vegetation via the well known relations of Mendez & Losada (2004), and is therefore selected as the most useful mode for this application.

The example in Figure XX below shows the importance of these processes on the FAST Tillingham (UK) field site, in combination with observations at this site as studied for FAST Deliverable 5.4 ('Prototype linkage of general rules for engineering requirements of dike design'). The thick black line indicates the bed level along a coast-normal transect, the thick green line the different vegetation covers observed. The thin black line represents the wave height change over the transect without bed friction or vegetation presence, i.e. purely depth-induced breaking. The thin blue, red and green lines represent wave height change in the presence of vegetation, for a range of vegetation drag coefficients. The dashed blue line represents wave height change in case of high bed friction due to very irregular small-scale bed topography (ridges and swales < 1m). Under these mild conditions (small waves, relatively low water depth), the attenuation by bed forms at the marsh edge dominates wave attenuation whereas under more extreme conditions (higher waves, deeper water) the attenuation by vegetation over the marsh plain will become more important.

For the Educational version, XBeach was used to generate a lookup table of attenuated wave heights for a range of possibly occurring combinations of nearshore waves and water levels, foreshore slopes and -widhts and vegetation types. The MI-Safe tool searches this table using the conditions at the selected site as input, resulting in a reduced wave height at the end of the vegetation where the levee is supposed to be. Subsequently, this reduced wave height is used to calculate a reduction in required crest height of the supposed standard levee, in comparison with a bare foreshore under the same forcing. 

For the Expert version of MI-Safe, a number (typically 6) of ~ 2km long transects has been defined at each study site, running from the nearshore to the position of the levee estimated from EO images. At some sites (NL, UK) the foot of the levee can be clearly distinguished, on other sites (RO, ES) the relevant end of a transect is more difficult to define. For all transects, dedicated site-specific XBeach simulations have been performed using the local bed level, hydraulic boundary conditions and vegetation cover. Just like in the Educational version, this results in the attenuated wave height at the foot of the levee that is used to calculate the required crest height, but with much greater precision because the actual situation is simulated rather than a substantial simplification.

The viewer (Gerrit, Amrit, Arjen, Ed, perhaps you can still use some of this text below?)

The viewer or better the dissimination platform is build up of 3 main parts, these are:

1. the canvas where the maps are shown
2. the data part where a selection of layers can be toggled on or of 
3. the results. 

Not yet mentioned is the back end of the dissemination platform which plays an important role in the data management of FAST.

The results part enables users to draw a profile of the coast. Via OGC services (a WPS in this case) data is extracted over the profile. This data is then classified for certain characteristics, such as elevation, water levels, wave characteristics and vegetation presence and type. This data is then used to query the table of model results. The result is shown  indicating whether or not vegetation is existent and or contributes to wave attenuation. 

The case study sites

Insert figure with all study sites.

United Kingdom 

Tillingham: This field site is a macro-tidal (Mean spring tidal range of ca 4.8 m) open coast or outher estuarine salt marsh of the UK east coast (Southern North Sea) on the Dengie Peninsula in Essex. The estuaries of the Rivers Blackwater and Crouch lie to the north and south of the peninsula. The exposed marshes of this peninsula extend up to a maximum width of ca 700 m and are fronted by extensive intertidal mudflats, with an earthen embankment landward of them and low lying agricultural land behind this defence. Over the past 100 – 150 years the marshes have experienced several phases of advance and retreat. Marsh surfaces are composed of clayey silts and are approximately horizontal, with elevations of between 2.4 – 2.7 m ODN (Ordnance Datum Newlyn, which approximates to mean sea level). The Dengie Peninsula currently experiences rates of sea level rise of approximately 2 - 3 mm a^-1 (Burningham and French 2011). The vegetation here is typical of UK east coast salt marsh with Aster tripolium, S. anglica, Suaeda maritima and Salicornia europaea present in the seaward areas. Higher up and further landwards, a canopy of P. maritima and A. portulacoides with E. athericus occurring on levees along creek margins is present. These species form mixed canopies but also exist in distinct mono-specific patches of several square metres in size, such that approximately uniform vegetation types can be found in close proximity to each other. Offshore wave heights have been estimated as averaging 1.09 m (on Long Sand Head, 42 km NE of Tillingham), with wave heights larger in winter (January), when mean monthly maxima reach 1.45 – 1.70 m (Herman 1999).

Donna Nook: This site lies to the immediate south of the Humber Estuary on the East coast of the UK. It comprises a foreshore that consists of mixed clays, silts, and mud deposits and is occupied by a salt marsh in the upper regions. The marshes here extend out to a width of ca 750m, with high exposure to winds from the north-east. The tidal range is 8.1m at Immingham, providing water depths above the marsh surface at highest astronomical tide of between 0.9 and 1.4m. Vegetation here consists of mixed north-western European salt marsh species (Atriplex, Puccinnellia, Elymus, Aster, Salicornia, and others). The marshes are flanked by extensive (> 1000m wide) tidal flats to seaward and dunes and/or earth embankments on their landward margin. Evidence suggests a marsh progradation rate of 1-1.5m per year over at least the past 5-10 years, through sediment derived from updrift (the Humber estuary and further north). Landward of the salt marsh for which data for FAST was obtained lies a recent managed realignment site (regulated tidal exchange through a breach in the most seaward embankment to create natural habitat to landward).

The Netherlands 

The Westerschelde estuary is located in the southwest of the Netherlands. The Westerschelde has a macrotidal regime; discharge is relatively low compared to the tidal prism of the estuary. Wind-driven waves are particularly important at the mouth, where fetch is longer. The Westerschelde has a salinity gradient from saline at the mouth near Vlissingen, to brackish at the Belgian border. The bed material typically consists of sandy sediments, but particularly the foreshores fringing the estuary tend to be muddier. The estuary is affected by human impact, such as dredging to facilitate navigation to the ports of Vlissingen, Terneuzen and Antwerp. The Westerschelde also has a high ecological value, particularly due to its multi-channel system surrounding tidal flats and shoals, that accommodate a rich benthic community, providing food to birds and (flat)fish. At the higher parts of these tidal flats, saltmarshes can develop, consisting of Spartina spp in the more saline parts to Phragmites spp. in the brackish areas. We focused on two case study sites, one sheltered saltmarsh (Paulinapolder), dominated by Spartina spp. and one wave-exposed eroding saltmarsh site (Zuidgors), consisting particularly of Elymus spp. and Atriplex spp.

Spain 

The Spanish FAST study sites are located within the inner system of the Bay of Cadiz. The Bay of Cádiz is located in the west of the Gulf of Cádiz, SW Spain, between 36° 23’ to 36° 37’N and 6° 8’ to 6° 15’W. This area was declared a Natural Park in 1989 (and since then, it has been also declared as SPA, SAC and Ramsar site), having the particularity that the fringes of the Natural Park are highly urbanised.

The bay is divided into 2 basins, a shallower inner bay with a mean depth of 3 m mean low water (MLW), and a deeper outer bay with a mean depth of 12 m MLW (Rueda & Salas 2003). The area of the inner system is profusely occupied by salt marshes and the tidal flats in front of the salt marshes are inhabited by seagrasses. In this area, sediments are composed mainly of fine sand and mud, with a high organic carbon content (Carrasco et al. 2003, Rueda & Salas 2003). However, specifically in FAST study sites, the grain size is mainly composed of mud and silt. The entire system is strongly influenced by semidiurnal co-oscillating tides with mean amplitude of 1.5 m (Alvarez et al. 1997). For the inner system, the tidal exchange is relatively high, with average water turnover rates ranging from 50 to 75% per tidal cycle (Alvarez et al. 1997). The area of the bay is protected from the action of Atlantic swell by a barrier island (city of Cadiz), but it is clearly influenced by wind waves that generates waves up to 0.5 m (Hs, periods <5s) on the fringes of the bay (tidal flats, with seagrasses).

Dominant saltmarsh vegetation on the intertidal area is mainly represented by Spartina maritima on the low saltmarsh and by Sarcocornia species (Sarcocornia perennis subsp. perennis; Sarcocornia fruticosa) and Halimione portulacoides (or Atriplex portulacoides) on the medium saltmarsh (García de Lomas et al. 2008). However, also other species like Suaeda vera and Limonium spp. can be found. Judging by historical data, the saltmarshes are rather stable and characterised by a complex patterned distribution (although detailed studies are not available).

Romania 

Razelm-Sinoe Lagoon System is included in the Danube Delta, that is the final part of the Danube River. The average annual precipitation in the Razelm - Sinoe Lagoon is 350-400 mm/year, with a large evaporation which should lead to the raising of the ground water, saturated by chlorides and sulphates, and, consequently, to humid soils salinization.

The climate of the Lagoon System is continental, with hot dry summers and very cold winters. The water bodies from the lagoon system happen to be partly or entirely frozen during winter, with frizzing periods varying from days to week. Complete ice cover for long periods, however, is rare. The Lagoon System is located in one of the windiest areas in Romania.

During the past century, the system has been subject to major changes due to human interventions. These changes resulted into a complete change of the Lagoon specific ecosystems compared to its pristine state. Throughout a series of hydro-technical interventions, the Lagoon System has been transformed into a fresh water reservoir, to be used for agriculture and fresh water aquaculture, considered at that time much more viable economically.

The planned functioning regime of this hydrotechnic system was:

-                      between April 1 - June 1, water accumulates up to the level of +80cm;

-                      from June 1 to August 15, level maintained at +80 cm;

-                      between August 15 and September 15, the level - descended between +30 +50 cm to ease the fishing (fishing continues until October 30);

-                      between November 1 and March 30, the level of retention - maintained at +50cm (to “protect” the northern compartment from the brackish waters of Sinoe Lagoon).

All these hydrotechnical interventions have automatically led to drastic changes of the hydrological, hydrochemical and biological conditions of these lacustrine and lagoon ecosystems.

Variations of the maximum and minimum water level regime

The hydrologic regime of the lagoon system is characterized by two specific elements, namely:

-       a seasonal variation related to the seasonal variation of the Danube River regime (via the water regime of the Sf. Gheorghe branch)

-       a variation related to the changes in the wind direction and intensity. The wind plays a major role in the water and sediments circulation between the parts of the lagoon system but also in the exchange at the remaining inlets (Edighiol and Periboina).

The highest frequency of strong winds is from N and NE – and under these conditions large water masses are circulated from the north of Razelm to the south, which determines an increase of the levels at Jurilovca or Canal 5 and a level decrease at Sarichioi or Sarinasuf.

Quantitative descriptors and their respective indicators for presenting hydrological status of the Razelm - Sinoe  lagoon.