Assessing Macrophyte and Ecosystem Service Changes in Shallow Eutrophic Coastal Waters Using Remote Sensing Methods

Abstract

Knowledge of the structure and spatial distribution of coastal water habitats is crucial for understanding coastal water systems. However, spatial habitat data are largely lacking, hampering ecological and ecosystem service assessments as required by EU policies. Mapping the structure, spatial distribution, and temporal dynamics of macrophytes is a particular challenge. In this study, we combined long-term macrophyte data with remote sensing methods (i.e., aerial and underwater drones, as well as SENTINEL-2 data) to assess their potential for spatial macrophyte monitoring and habitat-based ecosystem service assessments, in which ecosystem services were linked to habitats using the expert-based Baltic Ecosystem Service Potential Matrix. Greifswald Bay in the German Baltic Sea served as the case study for this research. Our aerial drone detected macrophytes up to a depth of 3 m that could be integrated into the existing macrophyte monitoring scheme of the Water Framework Directive. Reliable data from SENTINEL-2 were only obtained in optically shallow waters and could therefore only be used as proxy indicators to assess changes at a water body level. Despite the uncertainties and inaccuracies of the SENTINEL-2-based macrophyte maps, they were crucial for filling data gaps and enabled a spatially differentiated ecosystem service assessment for Greifswald Bay. However, we have shown that the commonly used matrix approach does not allow for the assessment of spatiotemporal changes at the water body level and is thus not suitable for supporting coastal and marine policy implementation.

1. Introduction

Coastal waters are of great ecological importance as transitional zones between land and sea. They provide habitats that serve as shelter, spawning, nursing, or feeding ground for a variety of marine species [1]. In addition, they act as natural filtering and buffering systems, particularly in the shallow and semi-enclosed Baltic Sea, where they play an important role in the retention of nutrients and other pollutants [1,2,3]. Benthic habitats, like seagrass meadows, absorb nutrients, produce oxygen, and stabilize sediments, while mussel beds filter out particulates and pollutants, thereby improving water transparency and reducing algal blooms [4]. Hence, knowledge of the structure and spatial distribution of these habitats is of central importance for understanding coastal water systems and the Baltic Sea as a whole.

Due to their ecological importance and functions, as well as the indication of ecological integrity, benthic habitats play a central role in the status assessment of coastal waters. This is reflected in the requirements of several European policies: The Marine Strategy Framework Directive (2008/56/EC) (MSFD) includes benthic habitats in its descriptors for the Good Environmental Status, the Habitats Directive (92/43/EEC) (HD) aims to protect and restore specific benthic habitat types (e.g., reefs, sandbanks. and seagrass meadows), and the Water Framework Directive (2000/60/EC) (WFD), which uses benthic habitats and species (i.e., macrophytes and macrozoobenthos) as biological quality elements for assessment of the good ecological status of coastal waters.

While benthic habitats at greater water depths are mainly determined by substrate and salinity, the distribution of emersed and submerged macrophyte habitats in shallow coastal waters is also determined by their exposure to currents and waves [5]. As a result, the presence and stability of these macrophyte habitats can rapidly change, for instance, due to extreme events like storm surges. Capturing the structure, spatial distribution, and temporal dynamics of macrophyte habitats is therefore a particular challenge.

Current macrophyte monitoring schemes in the Baltic Sea represent a compromise in which costs and effort play a major limiting role. They primarily rely on dive surveys along transects, in which the species composition and maximum depth distributions of submerged macrophytes (i.e., macroalgae and angiosperms) are assessed [6,7,8,9]. Due to the high costs of divers, sampling is conducted at low frequencies (usually every 3 to 6 years per waterbody), particularly in the shallow inner coastal waters. Thus, available data are spatially limited, and information on the exact extent of submerged macrophytes and population trends is lacking.

New remote sensing technologies such as drones and satellites offer a remedy and can complement existing monitoring to obtain a better overall picture of the macrophyte distribution [6]. While these methods have been successfully used to map benthic habitats and especially submerged macrophytes in the Mediterranean and Northeast Atlantic [10,11,12], as well as the outer coastal waters of the Baltic Sea [7,13], there is little information on their effectiveness in shallow, turbid, eutrophic waters such as the inner coastal waters of the Baltic Sea, including bays, lagoons, and estuaries.

While status assessments of the WFD and MSFD primarily focus on temporal changes in the species inventory and the depth distribution of macrophytes within their assessment units (i.e., coastal water bodies and Baltic Sea sub-basins) [14], the management of Natura 2000 sites and implementation of the new EU Nature Restoration Law (EU 2024/119) require an integration of spatial approaches. The new EU regulation requires member states to restore at least 30% of degraded marine habitats according to the European Nature Information System (EUNIS) classification by 2030. For this, better knowledge on the spatial distributions of marine habitats is urgently needed.

At the same time, a consideration of ecosystem services is increasingly required by EU policies [15]. Ecosystem services are defined as “the benefits nature provides to humans” [16] and are typically divided into provisioning (e.g., animals or plants for nutrition), regulating (e.g., erosion and flood control), and cultural services (e.g., recreation or landscape aesthetics) [17]. In addition to the required ecological or environmental assessments, ecosystem service assessments introduce a socio-economic perspective and thus offer a holistic view on nature protection [18]. Also, the new EU Nature Restoration Law has a strong focus on ecosystems and habitats that help to mitigate climate change, reduce the impacts of natural disasters, or ensure food security. Hence, macrophyte-like seagrass and macroalgae are of particular interest.

For ecosystem service assessments to support the implementation of the WFD or MSFD, the assessment of temporal changes in ecosystem services on a water body scale was still sufficient, as demonstrated by Inácio et al. [19,20] and Schernewski et al. [21,22]. A higher spatial resolution and a higher level of habitat differentiation are needed to enable ecosystem service assessments in the context of nature policies (e.g., EU Nature Restoration Law or Natura 2000 sites). However, spatially explicit mapping and the assessment of coastal and marine ecosystem services are still limited, and assessments of spatial and temporal changes largely lacking [23,24,25,26]. Hence, the mapping and monitoring of coastal and marine habitats have been proposed as key priorities to enable ecosystem service assessments in coastal and marine areas [23,27].

This study aims to answer the following overarching questions: What can the spatial monitoring of macrophyte habitats in shallow waters look like? Can this form the basis for spatially differentiated ecosystem service assessments? And do such assessments support policy implementation? For this, we examined the Greifswald Bay, a large shallow bay in the German Baltic Sea, as a case study and applied a mixed methods approach, as shown in Figure 1. We combined long-term WFD macrophyte monitoring data (provided by LUNG) with drone- and satellite-based macrophyte mapping to produce habitat maps that were used to assess changes in ecosystem services between different points in time using a matrix-based approach.

2. Materials and Methods

2.1. Study Area

Greifswald Bay is the largest shallow bay at the southern Baltic Sea coast, with a surface area of 494 km² [28] and average depths of 5.8 m (max. 13.5 m). It is located in the northeast of Germany and bordered by the German mainland in the south and west, as well as the island of Rügen in the north. As shown in Figure 2, its coastline is strongly indented and characterized by smaller bays and lagoons. Greifswald Bay has two connections to the Baltic Sea, including a large opening in the eastern and a narrower channel on the western side. With salinity values between 5.3 and 12.2 psu (average 7.5 psu), it is a mesohaline inner coastal water according to the classification of coastal waters of the German Baltic Sea under the EU WFD [1,28]. Like the rest of the Baltic Sea, Greifswald Bay is a micro-tidal system, and its hydrodynamics are mostly wind-driven [29]. Periods of northeasterly winds can generate water level fluctuations of up to 1.5 m at the coast and lead to a water inflow from the Baltic Sea, renewing its waters around 8 to 12 times per year [29,30]. The seabed substrate consists mostly of sandy sediments and clay–gravel mixtures, particularly in shallow areas, while muddy sediments occur only in deeper areas. Reefs occur in the central and western part of Greifswald Bay, but in general, hard substrate is sparse [30,31]. Historically, Greifswald Bay was macrophyte-dominated and characterized by various species of charophytes (e.g., Chara aspera, Chara baltica, and Chara cenescens), macroalgae such as Fucus and Furcellaria, and spermatophytes such as Zostera marina, Najas marina, and Ruppia. Dense spermatophyte meadows down to 6 m depth and and macroalgae down to 8 m depth have been reported [30,32,33]. Due to an increase in nutrient concentrations and consequently turbidity, between 1950 and 1980, macrophyte cover declined from 90% to 15%, and the bay shifted to a phytoplankton-dominated ecosystem [29]. Despite nutrient load reductions in the past decades, macrophytes have not recovered, and Greifswald Bay is considered to be in a poor ecological state according to the WFD [14,28].

With an area of 665 km², the catchment area is rather small and dominated by tourism. There are numerous marinas and water sports schools around Greifswald Bay, as well as popular activities that include sailing, wind surfing, canoeing, and swimming. Shipping and transport take place in particular around Greifswald and the industrial park Lubminer Heide, which is being developed into an energy and industry hub for regional development. In addition, Greifswald Bay is a key spawning area for the western Baltic spring spawning herring and provides good feeding conditions for many fish species [30,34]. Yet, in the past decades, fish catches strongly declined and only amounted to 212,600 t in 2021 [35].

2.2. Analysis and Documentation of the WFD Macrophyte Monitoring Approach and Data Set

We analyzed the long-term WFD macrophyte monitoring data set (provided by the Mecklenburg-Vorpommern State Agency for Environment, Nature Conservation, and Geology (LUNG)), to document the available information on the distribution and state of macrophytes in Greifswald Bay and to identify strengths and weaknesses of the WFD macrophyte monitoring approach. The data set includes macrophyte data from 27 sampling stations in Greifswald Bay (see Figure 2) collected between 2004 and 2021 by the company MariLim on behalf of LUNG within the scope of the WFD implementation.

Since the introduction of the WFD, the macrophyte monitoring and assessment scheme applied to German coastal waters changed several times. First, an assessment index for the ecological status of macrophytes (called ELBO) was developed, which considers the parameters maximum depths distribution and characteristic plant communities and includes water body specific reference values for each parameter [33,36]. The applicability of ELBO was tested in practice between 2004 and 2007 and then slightly adjusted and used for an annual monitoring between 2008 and 2014 [36]. Following an evaluation of the ELBO system in 2013/2014, it was replaced by the currently used assessment scheme PHYBIBCO (PHYtoBenthic Index for Baltic inner COastal waters) in 2015. Since then, the inventory of macrophyte species and their ecological value have been assessed instead of characteristic plant communities. As described in detail in the PHYBIBCO guidelines [9], the assessment of the maximum depth distribution and species inventory is conducted in defined depth intervals along underwater transects. For each depth interval the total vegetation cover, macrophyte species, and sediment classes are recorded. However, for the assessment of the ecological status, only the composition of macrophyte species is relevant, but not their depth-dependent coverage. The maximum depth distribution is determined by a continuous dive between the last two depth intervals and is defined as the maximum depth at which the coverage of submerse rooted plant species is ≥10 % [9].

2.3. Drone-Based Habitat Mapping

To test the suitability of consumer aerial (UAV) and underwater (ROV) drones for macrophyte monitoring in Greifswald Bay, we conducted field sampling during November and December 2020, as well as during May and June 2021. The selected sampling site (shown in Figure 2 and Figure 3), season, and days were chosen to ensure optimal conditions, taking into account factors such as water transparency and turbidity, macrophyte growth conditions, wind speed, wave height, light conditions, and cloud coverage.

For the aerial habitat mapping, we used a DJI Phantom 4 Pro V2 (DJI, Frankfurt am Main, Germany) consumer aerial dronecoupled with a gimbal RGB camera of 20 MP (see Figure 3. To filter out light reflections at the water surface and adapt to the prevailing light conditions, the integrated camera was additionally equipped with a circular polarization filter of Walimex, which was manually adjusted before each flight. Four aerial transects were selected (see Figure 3, with the objectives to cover a large area, reach maximum depth at which the bottom would be visible with the UAV, and to cover a heterogeneous area with different benthic structures, including unvegetated seabed, seabed covered by submerged macrophytes, and boulders or rocks. The flight routes for the aerial transects were predetermined using the third-party software DroneDeploy and autonomously flown by the UAV at maximum altitude (i.e., 100 m), resulting in a resolution of 3 cm/pixel.

The aerial habitat mapping was supplemented by an underwater habitat mapping in order to validate the UAV-based habitat maps and for assessing the potential of underwater drones to identify macrophyte species. For the underwater mapping, we used a BlueROV2 (Blue Robotics, Torrance, CA, USA) (see Figure 3. The ROV was operated via a 100 m long cable, a laptop with the open-source software QGroundControl, and a gaming controller. In addition to a front-mounted full-HD (1080p, 30 fps) RGB camera with a field view of 80° horizontally, we attached a GoPro7 (GoPro, San Mateo, CA, USA) of 12 MP to the bottom of the ROV at nadir position. The underwater transects were monitored on the same day as the corresponding aerial transect. Since the northern shore of the field sampling site turned out to be too shallow to operate the ROV, we conducted the underwater monitoring in transects T3 and T4 with an inflatable boat and wading using a handheld GoPro7 camera.

Following the field sampling, we generated orthomosaics from the UAV images using the DroneDeploy software. Subsequent processing and analysis were done in ArcGIS (version 10.8.1). We applied an object-based supervised classification to identify macrophyte habitats in the UAV images. For this, we created benthic structure classes and training samples based on underwater images, making a key distinction between macrophyte-covered areas and areas without submerged macrophytes. In Transect 4, rocks were included as an additional benthic structure class. In the object-based classification, we utilized medium spectral and spatial detail segmentation, as well as the Support Vector Machine classifier. The shoreline and 3 m isobaths were used as classification boundaries. The latter was selected in line with the median depth distribution limit (i.e., 2.8 m) according to the WFD monitoring data for Greifswald Bay (provided by LUNG).

The accuracy of the supervised classification of the UAV images was assessed through a confusion matrix with total accuracy and kappa coefficients (0 indicating no better than random, >0 better than random, and <0 worse than random) [37] using 100 validation points created in an “equalized stratified random” manner.

2.4. Satellite-Based Habitat Mapping

In addition to the drone-based approach in the field sampling site Lanken, we conducted a satellite-based mapping with the aim to assess the macrophyte distribution in the entire Greifswald Bay. For this, we used freely available SENTINEL-2 images from the CODE-DE browser of the Copernicus mission [38], which provides atmospherically corrected SENTINEL-2 images. SENTINEL-2 samples 13 spectral bands with spatial resolutions of 10, 20, and 60 m. Selection criteria for the images included high underwater visibility, low cloud cover and high vegetation vitality. SENTINEL-2 images of Greifswald Bay recorded on 23 September 2020 (Processing baseline: N0209; tiles: T33UUA and T33 UVA; relative orbit: R065) were most suited to fulfil the criteria, and they were used to represent the initial time around the field sampling period. In line with the frequency of the current WFD macrophyte monitoring procedure [9], a second image recorded three years later was selected to assess changes over time. For this, an image recorded on 8September 2023 (Processing baseline: N0209; tiles: T33UUA and T33UVA; relative orbit: N0509) was used.

The pre-processing (including resampling and mosaicking) and the object-based classification of the SENTINEL-2 recordings were done using the Sentinel Applications Platform (SNAP) version 9.0.0. Different algorithms for the object-based classification were tested. After visual inspection of the classified images, the Random Forest Classification was found to yield the best results and was further applied. We used the same benthic structure classes (i.e., macrophyte-covered areas and areas without macrophytes) and classification boundaries (i.e., shoreline to 3 m depth) as in the classification of UAV images, and we also assessed the accuracy of the classification results based on kappa coefficients (c.f. Section 2.3) using 100 validation points.

2.5. Assessment of Spatial Ecosystem Service Changes

For the ecosystem service assessment, we used the matrix-based approach [39], which links ecosystem service potentials to land use/land cover types or habitats. Due to the possibility of using various data sources, including expert judgement, the matrix-based approach is applicable in data-scarce areas and has become a common approach for mapping and assessing marine ecosystem services [40,41,42,43,44,45]. Marine ecosystem service potentials are commonly linked to benthic habitats using classifications like EUNIS [45,46].

In this study, we applied the German Baltic Ecosystem Service Potential Matrix (Baltic ESP Matrix) developed by Schumacher et al. [37,38], which covers terrestrial, coastal, and marine habitat types and is generally applicable to assess ecosystem service potentials in northern Germany and the southern Baltic Sea Region. Its marine habitat typology, which is relevant for this study, is closely related to the EUNIS typology and is shown in Figure 4 (upper row). It combines the substrate, benthos (in particular, submerged macrophytes), and the water body (depth and salinity).

Habitat maps form the basis for spatial assessments of ecosystem services and their changes over time. We prepared habitat maps for Greifswald Bay for the years 2020 and 2023 according to the habitat typology used in the Baltic ESP Matrix (see Figure 4, upper row). For this, the macrophyte distribution data obtained from the satellite-based mapping (c.f. Section 2.4) were further processed in ArcMap (version 10.8.1) and combined with publicly available seabed sediment data obtained from the Federal Maritime and Hydrographic Agency (BSH). In addition, we prepared a habitat map reflecting the historical distribution of submerged vegetation in Greifswald Bay, as described by Kanstinger et al. [30].

The ecosystem services and scoring range that are shown in Figure 4 and used in this study were directly taken from the initial Baltic ESP Matrix. Only services that can be logically excluded for coastal waters (e.g., timber) were excluded from the initial matrix. The Baltic ESP Matrix uses a scoring range up to 100 (highest possible ecosystem service potential). The score 5 indicates a provision that can be logically excluded. As described by Schumacher et al. [47,48] these scores have been derived from an expert-based assessment, including more than 60 experts.

3. Results

3.1. Macrophyte Monitoring, Distribution, and Composition in Greifswald Bay

The long-term monitoring data on the maximum depth distributions of macrophytes in Greifwald Bay (provided by LUNG) are summarized in Figure 5 and cover a total of 27 stations sampled between 2004 and 2021. Changes in the monitoring approach between the different assessment schemes are clearly shown. During the practical tests and under the ELBO system, macrophyte monitoring was conducted largely in the northern part of Greifswald Bay, with the exception of the sampling station in Lanken (2-LNK). With the introduction of PHYBIBCO in 2015, there has been a shift from sampling a few stations annually to employing an increased number of stations with a reduced sampling frequency (i.e., every 3 to 6 years per station). This led to a better coverage of the entire water body, even though there has still been a strong focus on the more sheltered areas in the northern part.

An effect of the different assessment methods (i.e., Practical tests, ELBO, and PHYBIBCO) on the maximum depth distribution data is not shown (see Figure 5). For most stations, the variability between different years is high, and the maximum depth distribution varies between a few decimeters (e.g., in Lanken (2-LNK) and Wreechen (13-WRE)) and up to 2 m in Zicker Nord (24-ZIN)). A clear trend in the maximum depth distribution of macrophytes between 2004 and 2021 is not shown. The results of the PHYBIBCO monitoring indicate a positive trend between 2015 and 2021, but due to the low number of sampled years per station, this trend needs to be regarded with caution and requires continuous data. Overall, the monitoring values for the maximum depth distribution of macrophytes range from 0.9 m to 3.6 m and indicate a poor ecological status for Greifswald Bay. Only three stations located in the northern part of Greifswald Bay (i.e., 18-GSW, 22-RHS and 27-KZI) achieved moderate status results in individual years.

Macrophyte species identified during the WFD monitoring are summarized in Figure 6. It shows that Baltic stonewort (Chara baltica) is the most widespread of the higher-rated charophyte species in Greifswald Bay and was identified in almost all sampling stations. In contrast, other high-valued charophytes (i.e., Chara canescens and Tolypella nidifica) were only recorded occasionally and at single locations. In particular, the distribution of the critically endangered bird’s nest stonewort (Tolypella nidifica) was mostly restricted to the northern shore around the Island of Vilm (i.e., sampling stations 15 to 18).

Among the spermatophytes, fennel pondweed (Stuckenia pectinata) was most abundant and found in all sampled stations. With the exception of the station Zicker Nord (24-ZIN), horned pondweed (Zannichellia palustris) was also found in all stations but not in all sampled years. Both species are considered to be indicators for eutrophication and therefore only have a low ecological value (score 1). A dominance of these high-growing submerged macrophytes indicates unstable water turbidity conditions and eutrophication [2,7]. Eelgrass (Zostera marina) and widgeon grass (Ruppia maritima) were also found to be widespread in Greifswald Bay. Both species are considered to be more sensitive toward eutrophication and therefore have a medium ecological value (score 3) [9]. Rarer spermatophyte species (e.g., Najas marina and Zostera noltei), which are considered to be of high ecological value, were not found.

The total ecological value (Figure 6, lower part) was found to vary strongly between sampling stations and years, and no trend is shown. According to the monitoring results, the status of most stations can be rated as poor. Only for single stations in 2015 and 2018 was the parameter rated as moderate. Among them is also the station Lanken (2-LNK), which is located within our field sampling site.

The long-term monitoring data show that the results for the maximum depth distribution, species composition, and ecological value are highly variable within a sampling station. Particularly, the assessment of the ecological value was strongly affect by the absence or presence of a species. Thus, a high spatial resolution and high level of accuracy are needed to assess the composition and ecological value of macrophytes within the scope of the WFD. In addition, the concentration of sampling stations in the northern part of Greifswald Bay and variation in sampling stations hamper an extrapolation of the monitoring data and assessment of macrophyte population trends.

3.2. Detection of Underwater Habitats Using Aerial and Underwater Drones

During our field sampling with the UAV, we covered a total area of 67 ha (0.14% of GWB) distributed over four, partially overlapping, transects (see Figure 7). Despite a careful planning of the field sampling days in November/December 2020 and May/June 2021, taking weather conditions into consideration, it was not possible to operate the UAV on all sampling days due to sudden weather changes. In addition, the UAV-derived image quality was strongly affected by water turbidity and unfavorable lighting conditions causing reflections on the water surface. Hence, not all images were suitable for further analysis. As a consequence, the recorded images of the four transects resulted from different sampling days and were taken under different environmental conditions. Transects 1 and 2 were recorded on November 30, 2020, during southwesterly winds with a speed of about 4 m/s, while Transects 3 and 4 were recorded on June 11, 2021, during northwesterly winds with a speed of about 3 m/s.

Despite the differences in sampling conditions, the resulting UAV orthomosaic images show that in all transects, benthic structures, and underwater habitats are best visible up to a water depth of 1.5 m, which is in line with the median Secchi depth of Greifswald Bay between 1990 and 2018 according to the WFD water monitoring data (provided by LUNG). We were able to detect underwater habitats up to a depth of 3 m in all transects, but the visibility in the optically deep waters was strongly dependent on water clarity and lighting conditions and differed between sampling days (see Figure 7). Also, during a sampling day, the conditions varied, as shown by the different water clarity conditions of Transects 3 and 4, which were both recorded on 11 June 2020.

According to the results of the supervised object-based classification of the UAV orthomosaics, 42% (28 ha) of the classified area is covered with macrophytes. Clear differences between the four transects are visible (see Figure 7). Transect 1, located at the steeper, western shore of the field sampling site Lanken, shows the smallest macrophyte cover (18 % of the transect). Transect 2, which covers a shallow stretch along the northern shore, depicts the largest macrophyte cover (46%). Transects 3 and 4 have a similar macrophyte coverage of 27% and 31%, respectively. In Transect 4, rocks were used as an additional habitat class and identified in 4% of the transect area. With kappa coefficients between 0.85 and 0.90, the accuracy of the classification results for the UAV orthomosaics was found to be very high. Yet, overlaps between Transects 1 and 2 (1 ha), as well as Transects 2 and 3 (7 ha), allowed us to also assess the agreement between the classification results. With a kappa coefficient of 0.37, there is a fair agreement in the classification results for the overlapping area between Transect 1 and 2. However, only a slight agreement was derived (kappa coef. 0.12) for the overlapping area of Transect 2 and 3. Since the detectability of submerged marcophytes depends on their vitality and size [49], the relatively low agreement between Transects 2 and 3 can be explained by the seasonal differences in which the UAV images were recorded.

Images recorded in the underwater transects with the ROV (Transect 1), as well as by wading (Transect 3) and boat (Transect 4), were used to verify UAV results and for the identification of dominant benthic structures and macrophyte species (see Figure 7). The underwater images enabled the differentiation of unvegetated seabed with and without signs of macrozoobenthos (i.e., invertebrates living on or in the seabed). The latter was only found in deeper waters in Transect 1 and detected with the ROV. In addition, four macrophyte species could be differentiated and include Stuckenia pectinata, Myriophyllum spicatum, gut weed (Ulva intestinalis), and brown algae—Pylaiella littoralis/Ectocarpus. The four species were abundant and found multiple times in all transects, but only the first two are relevant for the WFD assessment (see Figure 6). Shallower waters close to the shore (in Transect 1) and the large shallow areas covered by boat (in Transect 3) and wading (Transect 4) showed a higher species diversity. As in the WFD macrophyte monitoring, Stuckenia pectinata was also found to be the most widespread species in the underwater drone-based assessment. However, we could not identify more eutrophication-sensitive species with a higher ecological value, such as Zostera marina and Chara baltica, despite their widespread distribution in Greifswald Bay, as well as their prior detection in nearby sampling stations (2-LNK and 3-LUD) using the WFD monitoring procedure.

The comparison of the WFD macrophyte monitoring data (Section 3.1) and the results of the drone-based habitat mapping show that the latter does not provide the accuracy required for an ecological assessment of macrophytes within the scope of the WFD and cannot replace existing WFD monitoring. The ROV in particular could not be used in shallow waters (<1.5 m) and did not allow for an assessment of the ecological value of macrophytes, as too few species were identified.

3.3. Detection of Underwater Habitats and Their Changes Using Satellite Images

The satellite-based habitat mapping for shallow waters up to a depth of 3 m for the entire water body of Greifswald Bay covered a total area of 106 km² (i.e., 21% of GWB). In comparison to the UAV orthomosaics, the SENTINEL-2 images have a much lower resolution (i.e., 10 m/pixel vs. 3 cm/pixel), and a differentiation between rock and vegetation was not possible. Thus, the classification of the SENTINEL-2 images only differentiates between vegetated and unvegetated areas.

As summarized in Figure 8, a total area of 57 km² was classified as vegetated and 49 km² as non-vegetated for the year 2020. The classification results for 2023 show a vegetated area of 52 km² and an area of 54 km² as unvegetated. With kappa coefficients of 0.68 (for 2020) and 0.76 (for 2023), respectively, the accuracy of the classification results is considered to be substantial. An extrapolation of the vegetated area to the entire water body revealed in a macrophyte cover of 11% in 2020 and 10% in 2023. This is slightly higher than the macrophyte coverage of ~7% identified in an aerial survey with a plane carried out in 2009 [30]. Studies conducted in the 1990s estimated the total macrophyte cover to be between 4% and 10% [22]. It should also be noted that comparisons of the total macrophyte distribution between studies should be treated with caution due to differences in the sampling and extrapolation methods [30].

Based on the analysis of the WFD macrophyte monitoring data, we assume that an increase in macrophyte cover from 7% in 2009 (detected by a Kanstinger et al. [30]) to 11% in 2020 (detected in this study) is unlikely. Even though the spatial distribution of macrophytes is not directly assessed in the regular monitoring, the monitoring results of the maximum depth distribution do not indicate significant changes (see Figure 5), making an increase from 7% to 10% unlikely.

The spatial overlap of the classification results for 2020 and 2023 (Figure 8, map) shows that 72% (76 km²) of the classified area remained the same for both years. This includes 39 km² of vegetated area (37% of the classified area) and 37 km² of unvegetated area (35% of the classified area). In contrast, a loss of macrophytes was found for 16% (17 km²) and a gain for 11% (12 km²) of the classified area.

The satellite images for years 2020 and 2023 in the three focus areas (Figure 8A–C) were used for a visual inspection and qualitative comparison between both years:

According to the classification comparisons, the macrophyte coverage around the field sampling site Lanken (Focus area A) remained largely unchanged between 2020 and 2023. This is also supported by a comparison of the satellite images for the area for both years (Figure 8A). An increase in macrophyte cover is shown for the shallow waters in front of the nature reserve “Peenemünder Haken, Struck und Ruden” (Focus area B). A comparison of the satellite images for 2020 and 2023 confirms this increase partly. However, according to the classification results, there was also an increase in macrophyte habitats in the very nearshore area around the Struck Peninsula. Here, the dark stripe, which is shown in the satellite images (Figure 8B), was classified as a macrophyte habitat. However, it is more likely that this is due to nearshore waves and currents causing an accumulation of suspended sediments, including drift algae, which poses a challenge for the satellite-based mapping of macrophyte habitats in shallow nearshore areas [13,49].

For the lagoon “Schoritzer Wiek” (Focus area C), a loss of macrophytes is indicated in the nearshore area, whereas stable macrophyte cover is shown for the inner part of the lagoon. Yet, a visual analysis of the corresponding satellite images (Figure 8C) shows that the water transparency in the lagoon was very low, especially in 2020. Hence, a low accuracy of the classification results for the lagoon “Schoritzer Wiek” can be assumed. Thus, the macrophyte loss indicated in the classification comparison was likely caused by changes in water transparency and resulting differences in satellite image quality rather than showing real macrophyte loss. The loss of macrophytes indicated for several lagoons in the northern part of Greifswald Bay (Figure 8) was likely due to differences in water transparency and image quality.

Our results also show that the satellite-based habitat mapping enabled a detection of underwater habitats. The SENTINEL-2 images in Figure 8 show that in optically shallow areas with light sandy sediments (e.g., Figure 8A), dense and clearly defined macrophyte habitats could be detected up to a depth of 2–3 m. Our drone-based field sampling confirms the presence of dense macrophyte habitats in this location. Hence, the results of the satellite-based mapping for the optically shallow waters seem reliable. However, the macrophyte cover in optically deeper waters and areas with lower water transparencies, such as the lagoons around Greifswald Bay, has likely been overestimated.

3.4. Changes in Ecosystem Service Potentials

Despite the limitation of our habitat-mapping approach and uncertainties in the resulting macrophyte cover for the years 2020 and 2023, the results enabled the preparation of habitat maps according to the habitat typology of the Baltic ESP Matrix by Schumacher et al. [47,48] and were therefore combined with publicly available sediment data. Together with the historic map (with a macrophyte coverage of ~90%), they formed the basis for assessing the changes in ecosystem service potentials in this study. Since separate sediment maps for the three years (i.e., 2023, 2020 and 1940s) were not available, the sediment maps from Tauber et al. [18] had to be used as a basis for all three habitat maps. The resulting habitat maps for the three points in time are shown in Figure 9 (upper row).

In the historic habitat map, areas with submerged vegetation are clearly dominating (88.5%), while non-vegetated sand, gravel, and sandbanks (3.2%), as well as non-vegetated clay and mud (8.2%), cover only small parts of Greifswald Bay. In contrast, the habitat maps for 2020 and 2023 are dominated by non-vegetated sand, gravel, and sandbanks (~60%), as well as non-vegetated clay and mud (~31%), while submerged vegetation covered only around 9% of Greifswald Bay in 2020 and 2023 (note: the area considered in the ecosystem service assessment differs slightly compared to the habitat mapping, leading to differences in the indicated macrophyte cover). The changes in submerged vegetation detected between 2020 and 2023 with the satellite-based habitat mapping (cf. Section 3.3) barely affected the resultant habitat maps.

Consequently, the ecosystem service potential values, which are shown on an aggregated level for provisioning, regulating, and cultural ecosystem services, were the same for the years 2020 and 2023 (see Figure 9).

In contrast, the comparison of ecosystem service potential values between the historical and the current state shows an increase in provisioning services (from 38 to 44) but a decrease in regulating services (from 64 to 51). The expert-based values for provisioning services shown in the Baltic ESP Matrix (see Figure 4) suggest that the increase in provisioning services is mainly caused by two services, namely, ‘Crops for human nutrition’ and ‘Minerals’. Both services received higher ESP values for non-vegetated habitats, which significantly increased due to the loss of macrophytes between the historic and current state. In contrast, submerged macrophytes in shallow aquatic ecosystems were associated with high potentials for regulating services, such as ‘Global climate regulation’ (R02), ‘Flood protection’ (R03), and ‘Nutrient regulation’ (R06), and they led to a decrease in the ESP value when comparing the current state to the macrophyte dominated historical state.

The maps and ESP values for cultural services are largely the same for the historical and the current state (CES: 63 vs. 66), which is also shown in the Baltic ESP Matrix, in which only the services ‘Recreation and tourism’ (C01) and ‘Natural heritage’ (C06) have different ESP values among different habitat types (see Figure 4). For the remaining cultural ecosystem services (i.e., ‘Landscape aesthetics and inspiration’ (C02), ‘Knowledge systems’ (C03), and ‘Regional identity’ (C05)), the ecosystem service potential is the same for all habitat types in mesohaline inner coastal waters according to the expert-based matrix.

Our results show that major habitat changes within a water body are required to achieve different outcomes from matrix-based ecosystem service assessments. The potential of marine ecosystem services depends not only on the benthic habitats but also for some services (e.g., seafood, abiotic energy, and landscape aesthetics) mainly on the water body, which are determined by factors such as depth, salinity, and exposure. Therefore, the potential values for some services are very similar or equal between habitats within a water body.

4. Discussion

4.1. Methodological Assessment

Our approach and resulting data have several limitations that must be acknowledged. First, we used a consumer RGB drone, freely available SENTINEL-2 data, and simple object-based classification algorithms to map the distribution of submerged macrophytes in Greifswald Bay. The accuracy of our results was assessed using kappa coefficients. A key strength of our approach lies in the combination of different methods and data sources (e.g., aerial drone and satellite imagery, long-term monitoring data, and underwater drone imagery), which allowed us to evaluate the reliability of our results. However, due to differences in sampling locations, years, and the spatial resolution of the methods used, the reliability assessment was largely qualitative, and a comprehensive quantitative evaluation is lacking. Nonetheless, the methods used provide sufficient data to answer our overarching questions. Yet, it is important to note that more advanced techniques (e.g., drones with multi-spectral sensors, higher-resolution commercial satellites, and advanced classification methods) are available [49], which could improve macrophyte detection and increase the accuracy and reliability of our results. Therefore, the overall macrophyte distribution should be interpreted with caution, and our conclusions about the potential of remote sensing methods for macrophyte monitoring can be considered conservative.

Second, we used the expert-based Baltic ESP Matrix [47,48] to assess the spatiotemporal changes in ecosystem services. The limitations of matrix-based assessments, including simplification, expert bias, and subjectivity in scoring, are well-documented [45,50,51]. Despite these limitations, the matrix-approach is widely used for mapping and assessing marine ecosystem services [45]. Rather than repeating the general limitations of the approach, we focus on its potential for assessing changes in marine ecosystem services, thus addressing the need for spatiotemporal assessments crucial for policy implementation.

4.2. Towards Spatial Monitoring of Macrophytes

WFD monitoring provides a long-term and comparable data set on the depths distribution and the species inventory in coastal waters. However, our analysis of the WFD data set for Greifswald Bay showed that an extrapolation of the monitoring data and an assessment of population trends is hampered by variations in sampling stations, as well as their concentration in the northern part of the lagoon, where macrophytes are more abundant, diverse, and stable.

Our results show that none of the remote sensing methods tested in this study provides sufficient detail to replace the expensive and time-consuming use of divers in shallow eutrophic coastal waters like Greifswald Bay. Even though underwater videos and photos recorded by a camera sled or underwater drone are considered promising and already used for monitoring macrophyte species and depth limits in the outer coastal waters of the Baltic Sea [6,7,8], the field sampling in Greifswald Bay showed that this is not feasible in shallow eutrophic inner coastal waters due to the stirring up of sediments and low visibility.

Aerial drones are best suited to complement current WFD monitoring efforts, enabling improved spatial coverage. Results from the Lanken field sampling site demonstrated that a UAV can detect the maximum macrophyte depth distribution of 3 m under favorable environmental conditions. While our study cannot confirm detectability at greater depths, the ability to record a 3 m depth limit with a basic consumer RGB drone suggests that UAVs are generally capable of identifying the depth limit of macrophytes in coastal waters. This assumption is supported by Lønborg et al. [6], who reviewed monitoring methods for submerged macrophytes in European coastal waters.

In practice, WFD macrophyte monitoring dives could be complemented by aerial drone flights to enhance spatial coverage. Flying 1 km to the left and right of each dive transect would provide a broader spatial perspective of the sampling stations. The flight time and area covered between the shoreline and the 3 m depth contour depend on the coastal morphology. For the 27 sampling stations in Greifswald Bay, coverage areas range from 43 to 149 hectares. Using the DJI Phantom 4 Pro V2 consumer drone, flight times would vary between 25 and 60 min, requiring two to four batteries per station (according to calculations with the DroneDeploy software). Battery capacity and travel time between stations are the primary constraints on the number of sites that can be surveyed in a single day.

During the WFD macrophyte monitoring in 2018 and 2021, 14 and 13 stations, respectively, were sampled within four days (according to the WFD monitoring data set provided by LUNG). Covering three to four stations in one day would also be feasible with a consumer aerial drone and could be increased with a professional drone. Therefore, integrating the drone-based approach into the ongoing WFD monitoring would be feasible, enabling a more comprehensive and systematic assessment of population trends. This would not only support the WFD, but it would also benefit other directives, such as the MSFD, Habitats Directive, and Nature Restoration Law, which currently rely heavily on the WFD monitoring data. From a research perspective, regular simultaneous monitoring using both divers and UAVs would help assess the detectability of macrophyte depth limits in coastal waters and provide a systematic, consistent, and repeatable approach to macrophyte mapping, allowing for the recording of macrophyte distribution and dynamics over time.

Yet, the described spatial macrophyte monitoring approach (i.e., sampling 1 km to the left and right of each dive transect) covers only a small area. Even if it is applied to all 27 stations in Greifswald Bay, it would encompass approximately 22 km² (4% of Greifswald Bay), requiring more than 16 h of flight time (excl. travel, flight planning, and drone setup). Given that only three to four stations could be sampled per day with a consumer aerial drone, at least seven days of fieldwork under favorable environmental conditions would be necessary. Consequently, consumer aerial drones are unsuitable for mapping the entire macrophyte covered area of a water body. Such comprehensive coverage is achievable only through satellites or aerial surveys using planes.

This study demonstrates that freely available SENTINEL-2 imagery produces high-quality results in optically shallow waters with dense and well-defined macrophyte habitats, but it is limited by its 10 m resolution. Previous studies mapping macrophytes in the Baltic Sea with SENTINEL-2 reported varying depth restrictions: for mapping macrophytes in the Baltic Sea with SENTINEL-2, the results varied. High accuracy was achieved up to 5 m along Germany’s outer coast [13], 3 m along the Swedish coast [7], and 1.5–2 m in the Gulf of Riga [52,53]. In Greifswald Bay, the results were reliable down to 1.5 m, being consistent with the median Secchi depth from long-term monitoring (provided by LUNG). However, macrophyte detectability varied with the density, substrate, and water turbidity. On the exposed southern shore, macrophytes were detectable to a depth of 3 m, but turbid lagoons (Secchi depths < 1 m) were unsuitable for mapping. Overall, the Secchi depth proved to be a good proxy for estimating the maximum detection depth consistent with Lønborg et al. [6]. Beyond this depth, the mapping quality declined significantly, often overestimating the macrophyte extent [53].

Aerial surveys by planes, which have been used for macrophyte mapping in Greifswald Bay in single studies [30,54,55] provide higher spatial resolution than SENTINEL-2, but they have similar limitations and cannot reliably detect macrophyte species composition or depth limits [6,30]. In addition, due to their high costs, regular aerial surveys are impractical, making satellites a better alternative. Huber et al. [7] suggest that periodically updated macrophyte maps of optically shallow areas could serve as baselines for detecting changes. This approach could complement field monitoring, allowing more frequent assessments of annual or seasonal changes, but it needs to be further investigated. To enable validation, satellite data covering the same time period as the field surveys with divers and aerial drones should be used.

4.3. Towards a Spatiotemporal Assessment of Ecosystems Services

The lack of spatially explicit benthic habitat data significantly hampers assessments of coastal and marine ecosystem services [23,56,57] and poses a challenge for practical applications. Our study demonstrates that remote sensing methods, particularly UAVs, can help address data gaps in submerged macrophytes, which play a vital role in providing ecosystem services in coastal waters [4,58]. However, for policy implementation, ecosystem service assessments must be scaled up to entire water bodies, where satellites provide the only cost-efficient method for large-scale mapping [7,49].

Despite the limitations of SENTINEL-2 images and uncertainties in our satellite-based results, SENTINEL-2 data can still be valuable for large-scale assessments covering entire water bodies. Combined with seabed sediment and bathymetry data obtained from public authorities, we could derive habitat maps for Greifswald Bay according to the EUNIS classification of marine habitats [46] and the related typology by Schumacher et al. [47,48]. Combined with the Baltic ESP Matrix, these maps enabled a spatially differentiated ecosystem service assessment in Greifswald Bay.

Previous studies have relied on geospatial macrophyte data compiled from different years and which are often only available for specific areas of interest, such as marine protected areas [47]. Satellite-based macrophyte mapping can help to overcome the lack of temporally homogeneous data and particularly in coastal waters, which are usually not or only insufficiently covered by broad-scale seabed habitat maps like the EUSeaMap [46,59].

While the commonly used matrix-based approach has the potential to assess ecosystem service changes through repeated applications, our applications of the Baltic ESP Matrix to the habitat maps for Greifswald Bay for different points in time (i.e., 1940s, 2020, and 2023) showed that the approach used for land is not easily transferable to marine systems. Changes in terrestrial ecosystem services are commonly assessed based on the 6-year updated CORINE land cover maps [60,61,62], while our results indicate that larger timescales and significant ecosystem changes are needed to assess changes in marine systems.

Although remote sensing methods allow for frequent and large-scale monitoring of macrophytes, the lack of regularly updated seabed sediment maps such as the EUSeaMap [46] or the seabed sediment data for Greifswald Bay (provided by BSH) [31] affected the quality of the habitats maps and consequently the ecosystem service potentials. However, seabed sediments in the micro-tidal Baltic Sea are considered relatively stable over several years [63]. Thus, significant changes on larger spatial scales are not expected even if the data were updated more frequently. This means that the resulting changes in ecosystem service potential are driven only by changes in macrophyte distribution (i.e., the habitat/ecosystem extent). Yet, according to the long-term WFD monitoring data and the satellite-based macrophyte mapping for 2020 and 2023, significant changes in habitat extent in a water body within a few years are not expected. Consequently, detecting significant changes in marine ecosystem services requires long temporal scales, limiting the applicability of habitat-based approaches for supporting coastal and marine policies, which typically follow 6-year cycles.

Our findings also suggest that a high level of habitat differentiation within a water body is relevant for only a few services, such as erosion regulation, which are closely linked to benthic habitats like seagrass meadows or reefs, while the potential of many services varies only slightly within a single water body. Due to the three-dimensional nature of marine systems, many ecosystem services depend more on water column characteristics than on benthic habitats. For example, while macrophytes provide spawning and nursery areas, the potential for fish and seafood also depends on pelagic conditions like salinity, oxygen, and nutrients. Similarly, abiotic energy potential is influenced by water depth and proximity to shore and cultural services, such as tourism and recreation or regional identity, which are generally associated with coasts and seas rather than specific benthic habitats. Consequently, assessments at the water body level, such as for Greifswald Bay, are sufficient for many services, reducing the need for fine-scale habitat differentiation. Thus, the matrix approach, which determines ecosystem service potentials based on habitat type and extent, is not suited for holistic assessments, which assess a large variety of services and cover large areas. A hierarchical habitat typology, adaptable to a specific service being assessed, seems more practical but further limits matrix-based methodologies.

Consequently, this study shows that habitat-based matrix approaches are not suited for supporting coastal and marine policy implementation. While they have been applied to assess habitat and ecosystem services changes on a very small spatial scales (i.e., few hundred hectares), for instance, within marine reserves [41,64], they fail to meet the requirements of marine policy and management, which typically focus on the water body level.

4.4. Transeferability of Approaches and Results and Future Perpectives

The methods and findings of this study extend beyond Greifswald Bay and the Baltic Sea. The lack of spatial habitat data, macrophyte monitoring approaches, and spatiotemporal assessment methods is a widespread issue across European seas. However, the Baltic Sea stands out as a pioneer in marine policy implementation and is leading with regard to monitoring activities and data availability [65]; it thus serves as an ideal case study.

Most EU Member States rely on divers for monitoring macrophytes [5]. The shortcomings identified in this study, particularly the lack of spatial data, thus apply broadly to European coastal waters. These monitoring schemes could benefit significantly from complementary remote sensing methods. However, water turbidity was shown to be a major factor hampering the quality of our results in this study. Transferring satellite- and drone-based approaches to highly turbid coastal waters with limited optically shallow areas is likely not feasible. In contrast, applications in less turbid systems appear viable, as demonstrated by several studies [10,66,67].

The inability of the widely used matrix approach to assess spatiotemporal changes in marine ecosystem services emphasizes the need for alternative methodologies. A recent study by Schernewski et al. [22] highlights the use of 3D modeling as a promising alternative for large-scale marine ecosystem service assessments. Such models provide quantitative, spatially resolved results for different points in time [26]. However, further research is needed to determine whether 3D modeling can provide meaningful assessments within policy-relevant timeframes and achieve spatially explicit results for single services.

5. Conclusions

This study combined long-term macrophyte monitoring data with remote sensing methods, including consumer aerial and underwater drones and freely available SENTINEL-2 data, to assess their suitability for complementing current monitoring approaches. Aerial drones were found to be particularly effective in improving the spatial coverage of macrophyte monitoring under the Water Framework Directive, allowing for better trend assessment at sampling stations. However, their limited spatial coverage restricts their ability to capture the full distribution of macrophytes within larger water bodies.

In contrast, the SENTINEL-2 satellite is the only viable option for large-scale mapping, providing reliable results in optically shallow waters. In deeper waters, however, macrophyte distribution is likely to be overestimated, highlighting the need for further validation and refinement. Despite these challenges, changes in macrophyte distribution in shallow waters are considered a promising proxy for broader macrophyte distribution trends.

Despite the uncertainties and inaccuracies of our SENTINEL-2-based macrophyte distribution maps, the use of satellite data is the best option to fill data gaps for macrophytes in shallow coastal waters. Combined with sediment and bathymetry data, it is possible to produce habitat maps according to the EUNIS classification, which are needed as a basis for spatially differentiated assessments of ecosystem services, as well as for the implementation of EU policies, particularly the Nature Restoration Law.

When combined with an ecosystem service matrix, such as the expert-based Baltic ESP matrix used in this study, these habitat maps allow for spatially differentiated ecosystem service assessment. However, to the best of our knowledge, this study is the first to show that repeated applications, as carried out for terrestrial systems, do not allow for the assessment of large-scale spatiotemporal changes in marine systems and are therefore not suitable for supporting the implementation of coastal and marine policies.