The study area covers partially the city districts Hansaviertel, Reutershagen and Köpeliner-Tor-Vorstadt in Rostock (northeast Germany) at the estuary of the River Warnow at the Baltic Sea (Fig. 1). The city and its surroundings cover an area of 181.5 km². The elevation ranges from 0 m to a maximum height of 54.64 m a.s.l. (Landesamt Mecklenburg-Vorpommern n.Y.).

Figure 1.  

Location and Land use of the research area in the City of Rostock.

Due to its proximity to the Baltic Sea, the climate in Rostock is mild-maritime. The annual mean temperature is 9.2°C (1981-2010). The annual precipitation sum is 730 mm with a maximum monthly precipitation of around 70 mm in July (DWD Climate Data Center 2022). Several heavy precipitation events occurred in the past (Fig. 2). Especially, the summer of 2011 was very wet and floods occurred more frequently (Miegel et al. 2014).

Figure 2.  

Heavy Rainfall events from 1995 - 2019 with more than 15 mm/h and the chosen rainfall event on 06.08.2011 for the modelling in this study.

The study area with a size of 4.5 km² is located in the southwest of Rostock (Fig. 1). This area was chosen because of the variety of different land uses. Approx. 50% of the area comprises green areas (parks, forests and woodland), 23% consists of traffic areas and 25% contains sealed areas (settlements, urban dense areas and industry) (Steinbeis-Transferzentrum Geoinformatik 2017). The soil types in the study area are mainly luvisol-pseudogley and regosol. The texture of the substrate is sandy loam and loamy sand with wet characteristics (Hanse- und Universitätsstadt Rostock – Amt für Umwelt- und Klimaschutz 2019a, Hanse- und Universitätsstadt Rostock – Amt für Umwelt- und Klimaschutz 2019b).

The hydrological model requires an appropriate dataset of meteorological, land use, soil and elevation information. The main meteorological input dataset is precipitation. We used the data of the DWD climate station Rostock-Warnemünde at one minute resolution (DWD Climate Data Center 2021a). The selected event (6.8.2011 13:40 – 14:40 h) covers a rain duration of one hour, with a total amount of 21.74 mm and a maximum intensity of 2.93 mm/min (see small figure in Figure 1).

Other required meteorological data that are used if (canopy) evaporation is activated, are the minimum and maximum temperature (DWD Climate Data Center 2019a), wind speed (DWD Climate Data Center 2019b), solar radiation (DWD Climate Data Center 2021b) and relative humidity (DWD Climate Data Center 2019c). These data are only provided in 10 minute resolution. To resample these data to the one minute interval, we keep the 10 minute value constantly for all one minute values of the intervals.

Spatial data of the land use includes soil-sealing information (Steinbeis-Transferzentrum Geoinformatik 2017) and, in addition, a point shapefile of the tree locations with information on the tree diameter and tree type was available (Hanse- und Universitätsstadt Rostock - Amt für Stadtgrün Naturschutz und Friedhofswesen 2017). Spatial data on soil type (Hanse- und Universitätsstadt Rostock – Amt für Umwelt- und Klimaschutz 2019a) were used to set up the geometry. A digital elevation model was provided by the State of Mecklenburg-Vorpommern at one metre resolution (Landesamt Mecklenburg-Vorpommern n.Y.).

The FRES demand analysis also required a set of spatial data. In addition to the land-use data, which was used to identify the traffic infrastructure (Steinbeis-Transferzentrum Geoinformatik 2017), a map with land reference values (the average economic value for a majority of areas of mainly the same use and value characteristics (Lenzen 2014)) (Hanse- und Universitätsstadt Rostock – Kataster- Vermessungs- und Liegenschaftsamt 2021) and a map with the location of monuments (Hanse- und Universitätsstadt Rostock – Amt für Kultur Denkmalpflege und Museen 2017) were integrated. In addition, a collection of spatial data on critical infrastructure including hospitals (Hanse- und Universitätsstadt Rostock – Kataster- Vermessungs- und Liegenschaftsamt 2017), fire stations (Hanse- und Universitätsstadt Rostock – Brandschutz- und Rettungsamt 2017), schools (Hanse- und Universitätsstadt Rostock – Schulverwaltungsamt 2017), care facilities (Hanse- und Universitätsstadt Rostock – Amt für Jugend Soziales und Asyl 2017b) and institutions for the disabled (Hanse- und Universitätsstadt Rostock – Amt für Jugend Soziales und Asyl 2017a) were used in the analysis.

For the hydrological modelling, we used LEAFlood (Landscape vEgetAtion and Flood model) (Wübbelmann and Förster 2022), which is based on the Catchment Modelling Framework (CMF) (Kraft et al. 2011). CMF is an open source programming library for hydrological modelling. The modular structure of this Python package provides high flexibility and is adaptable to different research questions. Hydrological processes can be selected from a range of different approaches depending on the question, as demonstrated by Förster et al. (2021) for physically based hydrological modelling of green roofs.

LEAFlood adopts CMF features to create the geometry, based on polygon cells out of GIS shapefiles on the spatial resolution that is required for adequate hydrological modelling on the city scale. Most models designed for urban areas focus on urban drainage with a simplified representation of vegetation (Iffland et al. 2021). LEAFlood considers hydrological processes of canopy interception, through-fall, canopy evaporation, soil infiltration and surface run-off (see Fig. 3). It accounts for a detailed representation of interception and a lateral surface run-off simulation through a 2D kinematic wave approximation. The model was described and tested in detail by Camarena et al. (2022). They compared measured run-off data with LEAFlood model results and verified the good representation of both interception by tree canopies and run-off at the quarter scale. In the present study, neither run-off nor interception measurements were available for Rostock, but model comparison and on-site inspections confirm the plausibility of computed spatial inundation patterns. In general, models should be calibrated for new sites, whenever possible. Even though some parameters are site-specific, the results demonstrated by Camarena et al. (2022) highlight that the model is capable of representing pluvial flooding for landscape elements, which is why we chose the model.

Figure 3.  

   The hydrological processes, defining parameters and storages in LEAFlood (Wübbelmann and Förster 2022).

The geometry for our study was created on the basis of an irregular polygon shapefile consisting of 4750 cells with an average size of approximately 1000 m², in order to best possibly resolve relevant urban landscape elements on the one hand side and numerical stability on the other. The canopy closure, which defines the amount of through-fall and canopy interception, was given by the quotient of the projected canopy area and cell area. Each tree was assigned a Leaf Area Index (LAI) value and an interception capacity, based on its species-specific attributes using the datasets of Breuer et al. (2003) and Iio and Ito (2014). After that, for each cell, a mean LAI and interception capacity was calculated from all trees included in that cell. For missing values, a mean of all existing values was used. Since literature values of interception capacity, like those compiled by Breuer et al. (2003), cover a broad range of precipitation events in terms of precipitation total and event duration, they do not necessarily represent the maximum retention governed by interception during heavy rainfall events, as demonstrated by Asadian and Weiler (2009) and Alves et al. (2018). Therefore, supported by the validated modelling experiments conducted by Camarena et al. (2022), a scaling factor of 5 was applied to the literature values to compensate the mismatch in temporal scale and to acknowledge the higher possible interception load of trees during heavy rainfall events.

LEAFlood uses a one soil layer approach, assuming that only the upper layer is relevant for infiltration and that percolation does not play a role due to the time delay. The used infiltration approach is Green-Ampt, which is an approximate theory adaptation of the Darcy equation (Rawls et al. 1993). Except for saturated conductivity (Ksat), all other parameters were the same throughout the study area (Table 1). The base value for Ksat was 0.3 m/d resulting from the soil property sandy loam (Sponagel et al. 2005). Depending on the degree of soil sealing, Ksat was reduced by the following function: \(K_{sat, reduced} = (1-\frac{sealing}{100})*K_{sat}\)

A higher degree of sealing, therefore, resulted in a lower Ksat value (Table 2). The porosity of sandy loams, given in literature (0.45) (Sponagel et al. 2005), was lowered to 0.3 due to urban compaction and resulting reduced infiltration rates (Gregory et al. 2006). The Manning roughness coefficient was assigned for each land-use class (Table 2).

The FRES analysis was undertaken with Python and ArcGIS Pro using an intersection of spatial information of population, economy, land use and hydrological model results. Fig. 4 gives an overview of the workflow and Table 3 shows the definitions of the used FRES terms and the used indicators. The potential demand was determined by a GIS analysis of various economic and social indicators. The output of LEAFlood was the water storage on the surface, the soil water and the intercepted canopy water. While the last two variables were used as indicators for the supply of FRES, the surface water, together with an intersection of the potential demand, indicated the actual demand. Finally, the FRES supply and demand budget was calculated by the difference of the supply and actual demand.

Figure 4.  

Workflow of the FRES analysis.

t₀ is the starting time at the beginning of the rainfall event. c_i is the water depth of an individual polygon. c_max is the maximum water depth of all cells for the respective parameter or the 90% quantile for the surface water depth.

The indicators for the FRES supply were the water depth of interception and soil. Each indicator and the total supply, which resulted from the sum of interception and soil water depth, were converted into relative scales from 0 to 1, respectively. The maximum interception depth was 7 mm, soil water depth was 3 mm and the total supply in one cell reached a maximum of 10 mm. Fig. 5 displays the indicators and the total supply in maps and a chart diagram of the area weighted mean supply of the indicators and the total supply for each land use.

Figure 5.  

FRES supply. a) Interception by canopies, b) Soil storage, c) Total supply.

In general, green areas, such as forests, parks and woodlands, had the highest supply. A very high supply was provided by forests with ~ 0.9. Both the supply through interception and through the soil were very high. The supply on green areas and woodland were mainly provided by soil (very high), while the supply by interception on this areas was low (0.2 to 0.3).

Traffic areas had a low supply, which resulted from a low supply by interception (0.3), while the supply by soil was very low. The other sealed areas had a very low supply, which was also due to the very low supply by interception (0.15).

Over the entire area, the interception supply was low (~ 0.3), the soil supply medium (~ 0.5) and the total supply low (~ 0.3) on a relative scale. However, the results also showed, if a canopy were present, the absolute amount of interception storage was higher than the soil storage.

The demand components of potential demand, flood hazard and actual demand are displayed in Fig. 6. While the maps show the spatial distribution of the relative scaled demand, the bar chart shows the area weighted mean demand over the entire study area grouped by land uses. The individual indicators for the FRES potential demand are mapped in the Suppl. material 3.

Figure 6.  

ES components of FRES demand. a) potential demand, b) flood hazard, c) actual demand.

The potential demand (Fig. 6a), based on the indicators of population, reference land value, monuments and infrastructure, was relatively high (0.6) (see also Suppl. material 3), with maximum values on traffic areas and urban areas (~ 0.75). On green areas and settlements, the potential demand was high and, on industry and woodland, it was medium. Additionally, forest areas in the south were indicated with a high potential demand due to the cultural heritage status of the area.

The flood hazard (Fig. 6b) resulted from the surface water from the LEAFlood model and was converted to a relative scale, based on the 90% Quantile of 31.4 mm, while the maximum reached up to 2047 mm in depressions. Therefore, a very high flood hazard was indicated on water bodies (0.95). Traffic areas had a high flood hazard, while the remaining land uses of forests, green areas, industry, settlements and urban density were exposed to a medium flood hazard.

The actual demand (Fig. 6c) resulted from the product of potential demand and flood hazard on each individual cell and can be a maximum of 1. On average, there was no high or very high actual demand for any land-use. The highest actual demand were on traffic areas (0.5) and urban density areas (0.4; medium), whereas, green areas, water, woodland, industry and settlements had a low actual demand on average.

The budget map (Fig. 7) shows the difference between the supply and actual demand with a relative scale from -1 (unmet demand) to 1 (supply surplus) and a balance of 0.

Figure 7.  

Budget of the FRES supply and demand resulting in unmet demand and supply surplus.

Greened spaces, such as forests, green areas and woodlands, had an average supply surplus. Thereby, forests had a very high supply and low actual demand, which resulted in a medium supply surplus (~ 0.55). Green areas and woodlands were exposed to a medium supply (> 0.4) and a low actual demand. On average, there was a very low supply surplus on green areas (0.1) and a low supply surplus (~ 0.2) for woodlands.

On the contrary, sealed spaces were indicated with an unmet demand. While the supply was low on nearly all land-uses, the actual demand was low or medium (traffic areas, urban dense areas). This resulted in a very low unmet demand for settlements, industry, traffic areas and urban density areas.

In total, the study area had a low supply and low to medium actual demand. Therefore, the budget was calculated with a very low unmet demand of -0.1.

By using hydrological models for FRES-assessment, it is possible to take different rainfall events and initial conditions into account. Thus, the results of the actual FRES demand and budget analysis are only valid for a specific event. However, this also gives the opportunity to test different scenarios and replicate real events with different initial conditions to get a bandwith of possible impacts. Designed events and ideal (drier) soil conditions, for example, could lead to an improbably high supply and are far from reality. The total capacity would be determined rather than the actual used flood regulating capacity available to the population .  

CMF fills the gap of flexible and modular hydrological modelling structure that the community is asking for (Elga et al. 2015). With this framework, it was possible to build up LEAFlood, which provides the opportunity to value single landscape structures and elements of ungauged urban areas including their lateral connectivity instead of entire catchments, while having a flexible choice of hydrological processes. This way, LEAFlood enables hydrological predictions below the scale of typical land use classifications as it allows for resolving single elements of urban districts in a distributed way such as parks, buildings, streets or even elements of green infrastructure (for instance swales or trees) (Camarena et al. 2022). It is capable of incorporating vegetation-related hydrological processes, which are an important FRES-supply element (Nedkov and Burkhard 2012, Burkhard and Maes 2017) and which is unaccounted in typical stormwater models used in urban hydrology. The use of point shapefiles for the trees, a land use dataset that resolves individual elements of a city and a 1 m DEM can, therefore, be considered as sufficient for the application in this model. However, modelling and programming experiences are required and, even if it is possible to take vegetation and individual landscape elements into account with LEAFlood, this hydrological model is - like all models - only a simplified representation of reality.

Besides the vegetation-related hydrological processes, we considered the infiltration with the Green-Ampt approach and the kinematic surface run-off. For infiltration, we only looked at the upper soil layer and did not consider percolation (water flow from the unsaturated to the saturated soil zone),  because of a time delay, most of the water infiltrates into the upper layer during an short rainfall event (Markovič et al. 2014). The same applies to interflow (horizontal water flow in the unsaturated soil zone), which is not considered in LEAFlood. These limitations are acceptable for event-based modelling in urban areas, as surface run-off is the most dominant process. Furthermore, evapotranspiration is neglected. This is based on the assumption that the rate of evapotranspiration losses is at least one order of magnitude lower than corresponding rainfall intensities during a heavy rainfall event (Elga et al. 2015).

The advantage of hydrological modelling, especially of LEAFlood, for valuing FRES, is its flexibility. Depending on the available data and the research question, the complexity of the model can be adapted and extended by the processes, input data or resolution. For instance, we used a simple soil approach with regard to the spatial distribution because detailed information about soil texture distribution was unavailable. Urban soils are highly heterogeneous, which is why a dense measurement network is necessary for detailed soil mapping. In addition to the enormous measurement effort, it is difficult to obtain according permissions. Therefore, the existing level of detail is sufficient for the research question and, even with the simpler approach, good conclusions can be drawn about FRES.

A calibration of the ungauged urban study area in Rostock is not possible because of missing field measurements, which is common for urban areas since these are not demarcated catchments on this scale (Krebs et al. 2014) and high measurement efforts are required. However, the functionality of LEAFlood has been proven in the study area of Vauban (Freiburg, Germany) by showing that the model is capable to model run-off and canopy interception (Camarena et al. 2022). Both study sites in Vauban and Rostock are smaller districts within a city with a soil texture of sandy loam. Information about trees is more detailed in Rostock; however, the mean characteristics of the trees and so the settings of LAI and interception capacities are similar. Therefore, the functionality of infiltration and interception processes can be assumed as transferable from one site to the other. Nevertheless, the set and calibrated model parameters in Vauban cannot be transferred directly to other areas for hydrological modelling, as they are site and event specific (e.g. saturated depth) and, thus, reflect uncertainties. However, the results can be considered as sufficient, since an on-site inspection showed comparable flooding of past pluvial flood events in Rostock (for instance, at the Holbeinplatz) (OZ 2014) and also other modelling studies in Rostock emphasise these areas, which were indicated as hotspots in our study (Biota 2013). Furthermore, Rostock provides a comprehensive dataset, which is sufficient for the mode setup of the qualitative analysis of FRES.

The results showed that the interception by vegetation has a large share of the total FRES supply, which is particularly true when the soil is highly saturated. This confirms the statement by Nedkov and Burkhard (2012) and Burkhard and Maes (2017) that interception plays an important role in FRES supply assessment. The importance of urban trees is further supported in other studies, which have found that urban trees have a larger canopy circumference than forest trees because they have more space to grow. Consequently, they have a greater interception capacity (Asadian and Weiler 2009, Carlyle-Moses et al. 2020). With LEAFlood, we can model the interception and the importance for FRES supply on the spatial resolution of individual landscape patterns and, therefore, perform a spatially-detailed FRES supply assessment for the indicator by modelling, instead of working with general assumptions of interception classification based on literature (e.g. Nedkov and Burkhard 2012, Liyun et al. 2018). We avoided a weighting of the two supply indicators according to their importance and influence on flood regulation by summing up the absolute water depth values of both indicators and calculating the relative scale of the total supply afterwards. Furthermore, we did not incorporate the supply by upstream areas as other studies did (Goldenberg et al. 2017, Shen et al. 2021). Since we investigated a short heavy rainfall event and the topograhy of the study area is realtively flat, it can be initally assumed that the inflow or retention from the surrounding areas is low during this short time period. Rather, the present study is concerned with the direct regulating contribution of the areas in the study area.

In addition to the common processes of interception and infiltration, smaller landscape and green infrastructure elements, such as green roofs, have great potential to contribute to flood regulation (Zölch et al. 2017, Basu et al. 2022, Twohig et al. 2022). Camarena et al. (2022) considered green roofs in LEAFlood with a simple approach by increasing the saturated conductivity. However, for a comprehensive consideration of these elements in a FRES assessment, LEAFlood should and can be adapted accordingly (Förster et al. 2021). Since no spatial information on green roofs was available for the study area in Rostock and, in addition, satellite images did not show any green roofs, we have not taken this element into account as a regulation function.

Flood regulating demand should not only be roughly estimated by land-use or population density, as it is often used in other studies (Nedkov and Burkhard 2012, Goldenberg et al. 2017, Vallecillo et al. 2019), but is multidisciplinary and multiple data types and indicators from different disciplines, such as societal, economic, ecological and cultural demand should be analysed (European Parliament 2007, Milcu et al. 2013, Dworczyk and Burkhard 2021). By taking into account economic and cultural values, a different demand results than the estimation based on land use. Monuments are classified with a high demand, which is missing in an estimation based on land use, for example, if it is a park. While most studies focus on a detailed analysis of the supply of ES, the demand is often neglected or considered by using comparably simple analysis approaches, for instance, based on land-use classifications (Campagne et al. 2020). Using different sectors and data types, including different levels of demand, increases the knowledge and details of the demand assessment (Dworczyk and Burkhard 2021). By considering various protected assets, such as population, economy, infrastructure and monuments, we are tackling this point. Nevertheless, not all aspects of demand could be taken into account and not all potential demand indicators could be quantified monetarily or biophysically. The existing data can be refined spatially and temporally by adding further information and including stakeholders. For instance, instead of the population density in 100 m grid resolution, the number of persons per building could be used. Another possibility is the intersection of the land-use with the land reference value for a more detailed damage potential built environment.

We defined the protected assets of population, economy, cultural and infrastructure as potential demand, by arguing that all vulnerable areas and activities have a demand for flood protection regardless of whether they are actually exposed to the hazard. The actual demand results from the areas of potential demand that would be flooded. Therefore, it must be noted that the actual demand calculated here is only valid for the selected precipitation event and its initial conditions. For a more comprehensive assessment, other possible extreme precipitation events and initial conditions need to be considered.

A mismatch analysis of FRES demand and supply is important to identify priority areas for adaptation with an unmet demand, which is necessary, for instance, for adaptation planning (Syrbe and Grunewald 2017). By comparing supply and demand through subtraction, areas with a balance, as well as unmet demand and areas with a supply surplus, can be identified. Such an analysis is an indication of various parameters, which visualisation with maps supports decision-makers for urban and adaptation planning (Lüke and Hack 2018). Since demand indicators are expressed in social or economic units (for instance people/100 m² or euro) and supply indicators on biophysical units (for instance mm or mm²), we transferred them to a relative scale from 0 to 1, in order to compare these different units of indicators. This means that   a direct comparison of the biophysical values cannot be conducted (Czúcz et al. 2018), but it does make it possible to compare indicators from different units for the supply and actual demand mismatch analysis. Additionally, interpreting the relative scale of indicators with the same unit relies on: 1) different upper boundaries related to the maximum or 90% quantile of each parameter and 2) the fact that this value is site specific and can be computed for arbitrary sites. Therefore, the approach is an indication instead of absolute values, because they are not comparable due to different units and maxima, but still reflect the bandwith of supply or demand.      Identifying mismatches or balances should, therefore, always be done with respect to scaling classification. Nevertheless, we additionally showed the imbalance between the biophysical values of supply and hazard by their quotients to draw attention to the unequal distribution of water in a cell and to illustrate the high run-off fractions over the surface.

Furthermore, the budget analysis can be strongly influenced by site-specific and short-term aspects. This is particularly true for this study, where event-based modelling was used. The results are valid for a one-hour event with a total amount of 22 mm with a saturated soil, as has already been observed in Rostock. A less saturated soil at the beginning could increase the FRES supply and consequently the supply surplus in green-related areas. However, no improvement is expected for areas with a high unmet demand since these areas are mainly highly sealed. Whereas, with a prolonged or more intense rainfall, the proportion of land with unmet demand is expected to increase.

We would like to emphasise here that the results of the mismatch analysis do not constitute a flood hazard map. Rather, it serves as an input dataset in the FRES analysis and as an indicator of hazard, which in turn, is the output of the modelling. Unlike the Flood Framework Directive (European Parliament 2007) or the IPCC's vulnerability and risk approach (Oppenheimer et al. 2014), a comprehensive FRES study, that considers and compares both supply and demand, captures the contribution of natural ecosystems to flood regulation. With this approach, missing biophysical FRES supply on demand areas can be identified. This information can help to identify adaptation areas in order to create a sustainable city by ecosystem and biophysical adaptation measures by increasing the FRES supply where the analysis highlights the need. In an area with high potential demand for flood hazards, such as a city, biophysical structures and ecosystems   thus have a social and economic contribution to protect the population and reduce damage costs. Long-term consequences of FRES loss might be high economic costs and increasing vulnerability and decreasing resilience (Gómez-Baggethun and Barton 2013). Therefore, the mapping of scaled FRES demand and supply indicators and their mismatch delivers an easy-to-comunicate and important tool to identify the benefit and missing FRES supply for a sustainable urban planning. 

Since CMF is a modular python package, it is possible to connect it with other models, including models from other disciplines (Kraft et al. 2010). Against this background, it would be interesting to examine whether the hydrological model can be linked to regional climate model information. With this, the FRES assessment could be conducted for different possible future climate scenarios. An extensive spatial analysis of the soil would further improve the model. Run-off measurements are necessary for a local calibration of the model. Adding the effect of urban drainage system is an other outlook for future research. This would allow a FRES analysis with more emphasis on artificial elements (Vallecillo et al. 2019).

In terms of ES research, it is interesting to compare the results obtained in this study with the well-known and frequently used ES matrix method, based on land-use classifications (Burkhard et al. 2009). So far, only a few studies have compared the matrix method to quantitative estimations (Campagne et al. 2020). To counteract the issue of scaling and different units, all indicators must be aligned to one unit. For this, in a next step, the ecosystem value could be converted into economic values (Constanza et al. 1997, Constanza et al. 2014).

It has already been mentioned that demand is multidisciplinary (Dworczyk and Burkhard 2021) and often neglected in ES research (Campagne et al. 2020). Besides developing, adapting and extending the demand indicators, modelling of temporal changes (e.g. land-use changes and population development) could be conducted to improve ES research on the demand side. Furthermore, the damage costs could be calculated in monetary values to add another aspect of economic value. The temporal and time aspect should also be considered on the supply side, by analysing the development of the used supply capacity and the mismatch with demand during a heavy precipitation event.

So far, we did not consider future climate and land-use scenarios. For urban planning, the method would be an interesting approach to test adaptation measures in terms of their FRES supply functionality under changing climate conditions.