In wetlands, multi-annual water level drawdowns and herbivory can induce cyclic vegetation succession. While water level drawdowns can be used in wetland management to increase the area of reed vegetation, an important habitat for wetland birds, herbivory may interfere with this process. Here, we studied the combined effects of a human-induced water level drawdown, i.e. the intentional temporarily and large scale lowering of the water level, and herbivores on wetland vegetation development.

In the Oostvaardersplassen wetland, we used satellite imagery to assess vegetation development with and without water level drawdown and with and without red deer presence (introduced in 1992). An herbivore exclosure experiment (2022-2024) across an elevational gradient tested the effect of grazing on vegetation development during a drawdown.

Satellite imagery showed an expansion of reed cover by 560 ha in the period without red deer (1987-1991) and by 420 ha with red deer (2020-2024), only in the area with drawdowns. The exclosure experiment highlighted an interaction between herbivory and water depth: The presence of red deer at drier locations had minor effects on reed expansion, whereas reed expansion was strongly inhibited at wet locations with presence of geese.

Our findings provide large-scale quantitative evidence of the interaction between a water level drawdown and herbivory on the restoration of reed-dominated wetlands. We show the effectiveness of a water level drawdown, when dry conditions can be maintained for several consecutive years, as a restoration tool to promote reed development and the potential to steer the impact of herbivores during restoration.

Study Site

The Oostvaardersplassen is a eutrophic wetland in the Netherlands (52°27'N 5°19'E) (Figure 1A). It is in a former part of the Zuiderzee estuary, which was closed off from the North Sea in 1932 and gradually turned into a freshwater lake due to input from the river IJssel (Van Leeuwen et al., 2021). In 1968, land was reclaimed, mainly for agriculture purposes. The lowest and wettest part of the polder (around 5600 ha), destined for industrial activities, was left unmanaged during the first years after reclamation. This part developed into a marsh and attracted many rare and characteristic wetland bird species. This natural development prompted a shift in management strategy, leading to its designation as a protected nature reserve (Cornelissen et al., 2014; Jans and Drost, 1995). The reserve consists of a reed-dominated marsh (c. 3600 ha) and a border zone with wet and dry grasslands (c. 2000 ha). In this paper, we focussed on the reed-dominated marsh of the reserve (Figure 1B).

The Oostvaardersplassen is located in a polder and isolated from the Dutch river systems. The water level is controlled by the water board using a pumping station. The water level is kept at a relatively constant level throughout the year for the benefit of agriculture, industry, and housing. As the surface level of the marsh is higher than its surroundings, there is no inflow from surface water. Adjacent to the west side of the polder and the marsh is lake Markermeer, which is separated from the marsh by the Oostvaardersdijk. The water level of the lake is about 4 m above that of the marsh. Percolation from the lake through this dike to the marsh is thought to be minimal due to the fine clay particles that limit water infiltration. Because of its higher surface level, an embankment was made around the marsh to keep the water in. The water level in the marsh is controlled by a weir. Through this embankment, water percolates to the lower area surrounding the marsh. The amount of seepage through the dike and embankment is unknown. Between 1992 and 2024, the annual precipitation was about 800 mm/year and the annual reference evapotranspiration about 200 mm/year (source KNMI). All together, during periods without a water level drawdown, water level fluctuations between winter and summer were around 20 cm. To create a diverse landscape in the border zone consisting of grasslands, heck cattle (Bos taurus) and Konik horses (Equus ferus caballus) were introduced in the 1980s (Ejrnæs et al., 2024). In 1992, also red deer (Cervus elaphus) were introduced, with 42 individuals being released in the grassland area. Up to 2018, red deer numbers were controlled by food supply, severity of the winter and inter- and intra-specific competition, as they live in a fenced area without large predators (Cornelissen, 2014). As a result, the number of red deer increased to more than 2500 in 2017. In 2018, the management of the large herbivores changed in response to a decrease in bird biodiversity in the border zone and animal welfare concerns. The population of red deer was gradually reduced to 750 in 2024 through population control. During the drawdown, it was estimated that more than 90% of the red deer population utilized the marsh area (pers. comm. P. Cornelissen).

Due to the relatively small water level fluctuations (Appendix in manuscript for water level data from 1980-2024) and high grazing intensity because of tens of thousands of moulting greylag geese (Van Eerden et al., 1997; Vulink et al., 2010; Vulink and Van Eerden, 1998; Zijlstra et al., 1991), the reed extent and heterogeneity of the marsh vegetation decreased in the first decade after reclamation (Jans and Drost, 1995; Van Den Wyngaert et al., 2003; Van Eerden et al., 1997). In 1987, a water level drawdown was conducted in the western section of the wetland area to restore reed vegetation, which lasted until 1991, after which the area was inundated again. In the eastern section, which is hydrologically separated, no drawdown was undertaken, and the water level remained unaltered, to allow waterbirds to reside in the eastern part, while the western part underwent a multi-annual drawdown. Water level data was monitored on three different locations, one in the western drawdown section and two in the eastern non-drawdown section (in two separate lakes: Hoekplas and Keersluisplas). From 1980 till 2000, water level data was manually monitored twice a month using a monitoring well with a known reference height (m NAP). From 2000 onwards, data loggers were installed in the same monitoring wells and data was collected daily. In the period 2005-2015, data loggers in the drawdown section and the Hoekplas (non-drawdown section) were malfunctioning. Data was still automatically collected in the Keersluisplas in the non-drawdown section and there were some additional manual measurements. Keersluisplas was still hydrologically connected to the Hoekplas and therefore these measurements give a representation of the degree of water level fluctuations between summer and winter in this period.

Other than the difference in water level management, the western and the eastern section are similar regarding vegetation composition (before drawdown) and the presence of high nutrient concentrations in the clay sediment (Bouma et al., 2024). The Oostvaardersplassen organic clay consists mostly of microfossils, biological remains, organic colloid/tissue clothed silt (quartz mainly), with the presence of less than 10% clay minerals, such as illite and chlorite (Cheng et al. 2004). Yet, sediment elevation is lower in the western section due to the drawdown in the 1990s, which caused subsidence of the sediment. Yearly monitoring of sediment height using a dGPS informed us about land subsidence during the current drawdown.  30 years after the last water level drawdown, the same development was observed, a declined area of reed vegetation, and in 2020 another water level drawdown was induced to achieve the same goal of restoring the reed beds.

Figure 1. (A) Overview of the Netherlands and (B) Oostvaardersplassen nature reserve with delineated in orange the (western) drawdown section (drawdown from 1987 till 1991 and from 2020 till 2025). Delineated in blue is the (eastern) non-drawdown section. (C) Picture of a reed-dominated exclosure in the summer of 2023, surrounded by pioneer vegetation. (D) Locations of the 27 exclosures and paired control plots in the drawdown section. Camera icons indicate 9 of the 27 exclosures that are equipped with a camera trap. The arrow illustrates the sediment height gradient from high to low elevation. The yellowish colour illustrates the reed and willow vegetation before drawdown. (E) Overview of the monitoring during the first and second water level drawdown. The names indicate the methods used for monitoring (vegetation monitoring (green), vegetation classification using satellite imagery (brown), dGPS measurements (blue)). Additionally, the start of the field experiment, building of exclosures, is shown in yellow. The timeline with important events in the Oostvaardersplassen wetland is shown at the bottom of the line. The pictures illustrate some of the most important events (1 = start 1^st drawdown, 2 = introduction of red deer, 3 = start 2^nd drawdown, 4 = response during drawdown).

2.2 Vegetation classification using satellite imagery

2.2.1 Vegetation structural types former and present drawdown

To determine the vegetation development during the current drawdown (2020-2024) and compare it to the previous drawdown (1987-1991) when red deer where not present, we used satellite imagery to classify vegetation structures on a large scale (Figure 1E). Landsat satellite images (30x30 m) were obtained from Sentinel Hub EO Browser (Tiff 32-bit, high resolution, WGS 84) for the period 2017 till 2024 (June – September). 2019 is missing due to a lack of suitable images because of cloud cover. Clipped images were analysed in R. To quantify the cover of water, bare soil, Persicaria lapathifolia, Rumex maritimus, pioneer vegetation, new (developed) reed vegetation, old reed vegetation and woody vegetation, we informed a model (rpart (Therneau et al., 2022)) with ground-truthed vegetation data (2022, 2023 and 2024). Vegetation recordings (n = 57) were made across the entire former lake area which emerged upon drawdown (including control plots, see section 2.4). At each location, plant species and their cover were recorded (1x1 m). The vegetation recordings were clustered (based on kmeans in R) in distinguishable dominant habitat types (pale persicaria (Persicaria lapathifolia), golden dock (Rumex maritimus), pioneer vegetation and new (developed) reed). Additionally, using aerial photographs additional locations (without field data) were assigned to the habitat types of “old reed” and “woody vegetation”. Clustering was done for 2022 and 2023 combined (k=5) and separately for 2024 (k=3), due to inundated conditions that made it not possible to combine these into one clustering. These classes and their locations were used to inform the model based on the different band reflections of the Landsat images that correspond to known habitat types (for decision trees and predictions see Figure S3-S6). For the years 2017, 2018, 2020, 2021 and the entire eastern section (without specific field monitoring), knowledge on the field conditions and high-resolution aerial photographs (PDOK, https://www.pdok.nl/) were used to assign the correct vegetation types to each location and inform the model (for decision trees and predictions see Figure S3-S6). Total area cover of each habitat type was calculated from the number of pixels for each habitat type.

2.4 Field exclosure experiment

To study the effect of geese and red deer on vegetation development, 27 ungrazed plots were paired with grazed control plots. To exclude red deer and geese, we constructed exclosures (2.0 m high to exclude red deer from grazing over the wire, 1.5 m in diameter to exclude birds from flying in) with PVC rods and steel wire (mesh width of 15×15 cm to exclude geese from getting in and red deer from reaching through) in April 2022 (Figure 1C). Within each exclosure we sampled a plot of 1x1 meter to exclude edge effects. While red deer and geese could technically reach somewhat inside the exclosure through the mesh, no signs of grazing on plants inside exclosures were observed during the study period. Each paired control plot was located 10 m south-west of the exclosure and in parallel to the existing old reed border. To include differences in sediment height (elevation measured with a dGPS; Piper, TopCon, Appendix in manuscript) and potentially different vegetation types, the exclosures plus paired control plots were spread throughout the area in nine transects of three exclosures each at 20-100 (high elevation), c. 300 (intermediate elevation) and c. 600 meters (low elevation) from the reed border (Figure 1D). At the time of the installation of the exclosures, vegetation was only present in the exclosures and control plots closest to the reed border. Winter inundation of the plots occurred in 2022/2023 with 1.39 ± 1.64 cm water on high elevation, 5.83 ± 4.40 cm on intermediate elevation and 11.74 ± 5.09 cm on low elevation and in 2023/2024 with 1.59 ± 3.01 cm on high elevation, 3.83 ± 6.26 cm on intermediate elevation and 6.31 ±10.28 cm on low elevation. Although differences between locations were small, we did observe locations that were dry in winter and locations that were continuously inundated. In spring, a standing water level (no matter the depth) will inhibit germination (Meredino & Smith 1991) and through that germination and species survival. It has been shown that a delay in sediment exposure dates of 1 week (Grace 1987) may already impact recruitment from the seed bank and eventually vegetation composition. In this study area, the vegetation composition along the environmental gradient seems to be driven by environmental factors such as timing of drawdown (e.g. a combination of water level and microtopography), salinity or nutrient availability (Ter Heerdt et al. 2017). A previous study in our study area has shown that the seed bank composition itself did not differ along the environmental gradient (Bouma et al. 2024).

 2.4.1 Vegetation monitoring

At each plot, vegetation was monitored during peak standing crop in August 2022, August 2023 and July 2024. To monitor vegetation development a 1x1 m quadrant was placed in the centre of the exclosure and the control plot to visually estimate the species cover for each plant species. Total cover could exceed 100%, because of undergrowth. In each plot, height of reed was recorded for three individual stems.

 2.5 Herbivore species detection and quantification

Wildlife cameras (Bushnell Core no Glow) were employed continuously from August 2022 till August 2024 (Appendix in manuscript) on 9 out of the 27 exclosures, covering elevation gradients (high, intermediate, low) in three rows across the area (Figure 1D). Cameras were visited every 2-3 months to replace batteries and SD-cards. The camera traps were facing towards the old reed vegetation and were located at a height of approximately 150 cm to provide a good view of herbivores visiting the area close to the exclosure. The cameras were triggered by movement and took one picture each time, after which they became insensitive to movement for 10 s. Obtained pictures were loaded into the Agouti environment (Casaer et al., 2019). Agouti automatically groups images into bursts, which is defined as one event in which occurring (groups of) individuals are repeatedly observed (in this case this was defined as pictures taken within 120 s). Each burst was annotated manually with the present species and the count of individuals per species. The cumulative number of red deer or geese (both flying and standing) present at each location for a certain period was divided by the total number of days the camera was deployed in that period. Overall, we analysed 240.602 images that were converted into 32.823 bursts covering the period August 2022 to August 2024. Of these bursts, 6.292 showed an animal on the picture, this included the observation of red deer (52%) and geese (7%). The lower number of pictures with geese compared to red deer may be due to flocking of geese which can lead to a sudden peak in observance. Greylag geese (Anser anser) were the dominant goose species in the wetland area, followed by white-fronted goose (Anser albifrons). Other animals seen in the pictures at low occurrences were starlings (Sturnus vulgaris), foxes (Vulpes vulpes), yellow wagtail (Motacilla flava), reed warblers (Acrocephalus spec.) and white-tailed eagle (Haliaeetus albicilla).

 2.5 Data analyses

2.5.1 Satellite imagery

Vegetation recordings for ground-truthing of satellite imagery of 2022, 2023 and 2024 were clustered using kmeans (stats package; (R Core Team, 2023)), resulting in 5 clusters (estimated based on the Elbow Method) and nstart set at 25. The clustered data was used to train the model for recursive partitioning with the function rpart (package rpart; (Therneau et al., 2022)) for the period 2017 till 2024. The complete Landsat satellite images were predicted using the function ”predict” (stats package).

 2.5.2 Field exclosure experiment

Differences in vegetation composition between exclosure and control plots were assessed per year through non-metric multidimensional scaling (NMDS) with Bray-Curtis distances (vegan package; Oksanen et al., 2022). Subsequently, ADONIS analysis was performed to identify statistical differences between groups (exclosure/control).

We analysed how reed cover (independent variables) was impacted by herbivory (presence/absence), sediment elevation (high, intermediate, low), time (dependent variables) and their interactions with generalised linear mixed models corrected for zero inflation with tweedie family (link = log, glmmTMB package; Brooks et al., 2017). Furthermore, we analysed how pioneer vegetation and total cover (independent variables) were impacted by herbivory (presence/absence), sediment elevation (high, intermediate, low), time (dependent variables) and their interactions with generalised linear mixed models with tweedie family (link = log) (glmmTMB package; Brooks et al., 2017). Cover of cattail and woody vegetation was data-poor and therefore the interactions among the dependent variables were removed from the model. The model was run with a correction for zero inflation and with respectively the negative binomial function and the tweedie (link=log) function.  

 2.5.3 Herbivore presence

Counts of red deer and geese, obtained from the camera images, were tested for dependence on presence of water (November 2022 – May 2023, October 2023 – August 2024) and sediment elevation (high, intermediate, low) with generalised linear models corrected for zero-inflation with Poisson family distribution (package glmmTMB; Brooks et al., 2017).

Data were analysed in RStudio version 4.0.3 (R Core Team, 2023). All data are shown with their average ± Standard Deviations (SD), and in all hypotheses testing procedures the significance level was pre-set at α = 0.05.