A study of the relationship between wetland vegetation communities and water regimes using a combined remote sensing and hydraulic modeling approach

Wetland ecosystems are dynamic and diverse landscapes in not only their spatial scale, but also their hydrological conditions and vegetation communities. Understanding the relationships between vegetation distribution and hydrologic conditions can assist in more effective management of wetlands. As a key component of wetlands, wetland plant communities vary spatially with the type of climate (Piao et al. 2004; Liu et al. 2013), soil (Sanderson et al. 2008; Xu et al. 2013; Wang et al. 2014) and, in particular, the characteristics of water regime in terms of inundation duration (IDU), inundation depth (IDE), and inundation frequency (IFR) (Stromberg et al. 1996; Toogood & Joyce 2009). Water regime is a primary factor influencing the composition, productivity, stability, species diversity, and succession of a wetland vegetation community. This factor acts as an environmental sieve that interacts with life history characteristics of plant species (Van der Valk 1981; Gerritsen & Greening 1989).

Numerous studies of individual species have shown different responses to the effects of different aspects of water regimes (e.g., Denton & Ganf 1994; Rea & Ganf 1994; Van den Brink et al. 1995). For example, Casanova & Brock (2000) suggested that IDU, rather than IDE and IFR, is the most important aspect of the water regime in segregating vegetation communities. Todd et al. (2010) linked hydrological dynamics with vegetation distribution across Everglades National Park, indicating that IDU and IDE are the principal structuring variables to which individual communities respond.

Controlled experiments have been performed to identify key hydrologic requirements for species under specified conditions (Newbold & Nature 1997). However, other researchers have concluded that controlled experiments have limitations in their ability to predict species presence or abundance in field settings (Silvertown et al. 1999). Hence, there is a strong need for a typical ecosystem to reflect the relationship of vegetation with associated water regime.

The extensive wetland developed by the intra/inter-annual water-level fluctuations of Poyang Lake offers a reference site to study this relationship. The ecological and environmental conditions of Poyang Lake have been changing rapidly due to climate change and human activities, and the water level has declined significantly over the last decade (Zhang et al. 2012b). As the largest freshwater lake in China, Poyang Lake's hydrologic characteristics depend on both inflow from tributaries and outflow into the Yangtze River, which is affected by river discharge and the water level (Shankman & Liang 2003; Li & Zhang 2015). The construction of the Three Gorges Dam (TGD, the biggest concrete dam in the world), which began to impound water on 1 June 2003, has been suggested as a causal factor for the decline in the water level of Poyang Lake (Feng et al. 2013; Lai et al. 2014). Since 2003, the effects of the TGD pilot impoundment on the regional hydrology (Guo et al. 2012; Mei et al. 2015), environment (Zhao et al. 2013), and ecosystem (Yi et al. 2010; Fang et al. 2012) have begun to emerge. Changes in climate (Guo et al. 2008; Tao et al. 2014) and other human activities (de Leeuw et al. 2010; Ye et al. 2013) have also dried out the Poyang Lake wetland.

Consequently, the altered hydrology could result in a potential change in the plant community (Toogood et al. 2008). Any change will be mediated through species traits or attributes that establish niche and competitive abilities of particular species (Keddy 1992). The effects are greatest in shallow water, where even small changes in lake level can result in conversion of a standing water environment to an environment in which sediments are exposed to the air, or vice versa, resulting in death by flooding or in plant seeding-bank germination (Keddy & Reznicek 1986). For instance, Yu et al. (2010) suggested that the Ganjiang delta has expanded toward the main body of Poyang Lake, a movement that was obviously influenced by the variation in water level. In addition, species traits persist in successful adaptations to the environment and competition can be used to classify ecological groups of species (Boutin & Keddy 1993; Hu et al. 2015). Therefore, a better understanding of the patterns of vegetation distribution involved with the changed water regimes is essential for wetlands management.

The effects of variable hydrological processes on vegetation distribution have been investigated previously in the Poyang Lake wetland. However, most studies of hydrology and vegetation structuring have taken place in relatively narrow spatial scales, from a belt less than 1 km² (Xu et al. 2014) and a typical sub-lake of up to tens of km² (Wu et al. 2010), to a reserve of hundreds of km² (Zhang et al. 2012a). The Poyang Lake wetland is a highly dynamic and diverse area with abundant vegetation communities along different hydrological gradients. Previous methods, such as controlled experiments and field surveys, are not available that reflect the spatial heterogeneity of the hydrologic and vegetation conditions of the entire wetland over the same period in time. Different sampling dates and investigation methods lead to different conclusions (e.g., Ge et al. 2011). Even if some researchers have carried out studies on the entire Poyang Lake wetland, quantitative relationships between specific water regimes and the distribution of dominant communities have not been well documented (e.g., Zhang et al. 2013).

To the best of our knowledge, remote sensing data with high spatial resolution provides a consistent and rapid measurement of vegetation conditions. Moreover, hydrodynamic/hydrologic models provide the most accurate method to identify the hydrologic conditions of a large area. Given this background, a combined approach of remote sensing and hydrodynamic modeling was applied to reveal the relationship between water regimes and the spatial patterns of communities in the entire Poyang Lake wetland. The objectives of this study were to: (1) identify the hydrological preferences of different plant communities over the entire Poyang Lake wetland and (2) evaluate the individual and combined effects of water regimes on patterns of community distribution.

Depending on the water level and other environmental conditions, wetland vegetation in Poyang Lake exhibits zonal distribution from the lake center to the shorelines. The primary vegetation zones include floating vegetation zone (e.g., Trapa bispinosa and Nymphoides peltata), submerged vegetation zone (e.g., Potamogeton malaianus, Vallisneria spiralis, and Hydrilla verticillata), emergent aquatic vegetation zone (e.g., Carex cinerascens and Phalaris arundinacea), semi-aquatic emergent tall vegetation zone (e.g., Phragmites communis, Triarrhena lutarioriparia, and Zizania latifolia), and mesophytic vegetation zone (e.g., Artemisia selengensis and Cynodon dactylon). Carex and Phragmites are the most widespread vegetation communities in the Poyang Lake wetland (Guan et al. 1987; Ge et al. 2011).

The primary Carex spp. (including Carex cinerascens, Carex argyi, and Carex unisexualis) can grow both in shallow water and wet soil. Another species – Phalaris arundinacea – is difficult to distinguish from Carex in remote sensing imagery. Therefore, both are classified into the Carex community in this study. The Carex community provides an ideal place for migratory birds to spawn, rest, feed, and avoid predators. It also supplies local residents with fuel, fertilizer, and grass. Phragmites is another important wetland community and usually mix with Triarrhena lutarioriparia and Artemisia selengensis. Phragmites communities provide food and habitat for wintering migratory birds, and their height also protects the migratory birds from human interference (Sang et al. 2014). The detailed composition of three typical communities within Poyang Lake wetland can be found in Table 1.

Owing to the flat terrain and fertile sediment, the alluvial plains maintain high plant species richness and diversity. Since the area of exposed grassland reaches a maximum in winter, the land surface conditions of the Poyang Lake wetland were derived using a December 24, 2013 Landsat 8 remote sensing image (LC81210402013358LGN00) with a spatial resolution of 30 m.

In the field survey, 259 test samples were collected in December 2013 coinciding with the satellite overpass (Figure 1). Sample sites were chosen mainly from the Poyang Lake National Nature Reserve (PLNNR), the Nanji Wetland National Nature Reserve (NWNNR), Kangshan, Baishazhou, and Duchang. In view of the spatial resolution of the Landsat 8 image (30 m), 10 types of land cover were defined: water, swamp, forest, mudflat, meadow, sparse grass, bare land, arable land, sand, and specified plant species (e.g., Carex cinerascens, Phalaris arundinacea, Phragmites communis, Triarrhena lutarioriparia, Artemisia selengensis, etc.). Considering the zonal distribution of the plant communities corresponding to the variation in elevation, field sampling was conducted perpendicular to vegetation zones to collect different species, and parallel to the shore of the lake to collect more samples from the same community. The individual community composition was assessed from a 1 × 1 m plot when the cover of the dominant plant species was greater than 50%. All sampling data were recorded by GPS for later image classification and accuracy assessment.

In this study, we applied object-based image analysis (OBIA) to examine the broad-scale composition of the general surface cover types of the Poyang Lake wetland. OBIA has been applied frequently and widely for image classification in wetlands and inundation systems (Desclée et al. 2006; Conchedda et al. 2008; Laba et al. 2010). Compared to pixel-based image analysis, OBIA adds object shape and context (e.g., neighborhood characteristics) to spectral and textural information in the analysis, which makes OBIA an exceptionally useful method for wetland studies (Johansen et al. 2007; Dronova et al. 2011).

First, we rejected non-vegetation pixels by specifying a threshold value of the normalized difference vegetation index (NDVI) in ENVI 4.8 (ITT Inc.). Second, groups of pixels at desired scale, shape, and compactness criteria were segmented into objects in ENVI EX 4.8. In order to select the most suitable segmentation, we assessed the output sensitivity to multiple combinations of shape, scale, and compactness. Third, based on the sampling data and field survey experience, we classified the segments into categories of interest by supervised classification. Finally, individual plant species were merged into specified communities; spatial distribution and quantitative evaluation were conducted in ArcGIS 9.3 (ESRI Inc.). Classification uncertainty was examined by a fuzzy set-based accuracy assessment.

Lake water levels were obtained for five gauging stations (Hukou, Xingzi, Duchang, Tangyin, and Kangshan) from the Hydrological Bureau of Jiangxi Province and the Hydrological Bureau of the Yangtze River Water Resources Commission (Figure 1 and Table 2). The limited amount of observed data did not reflect the hydrological information of the entire wetland, especially for the dish-shaped pit groups (a large quantity of special geomorphic units) and the nature reserves due to the coarse spatial resolution. Accordingly, the water level information for the Poyang Lake wetland was simulated using a 2D depth-averaged hydrodynamic model, which was implemented previously by Li et al. (2014) using the MIKE 21 code (DHI 2007). The MIKE 21 code has been extensively applied in studying various water bodies around the world (Martinelli et al. 2010; Schoen et al. 2014).

The model covers an area of 3,124 km², which was determined by examining the historic lake surface during periods with high water levels. A 2D grid system with an unstructured triangular grid was adopted to capture the complex bathymetry of Poyang Lake (surveyed in 1998 and updated with new data obtained in 2000). The size of mesh elements varied from 70 to 1,500 m, resulting in a total of 20,450 triangular elements. The time step was set to 5 s to limit the Courant–Friedrich–Levy number for a stable solution.

Observed daily inflows from the five main rivers (monitored by seven gauging stations) were input into the lake model, specified as upstream boundary conditions. Meanwhile, observed daily water levels of Hukou were specified as the downstream boundary. Runoff from the ungauged catchment area was calculated using a simple linear extrapolation of the gauged runoff, and was added to the gauged inflows. For the periodically inundated wetlands, an option in MIKE 21 was used, which accounted for incidental rainfall at all model elements, whereas evaporation only applied to the wet elements representing the inundated area. For dry elements, it was assumed in the model that all rainfall was transformed to surface runoff. The numerical option of wetting and drying in the hydrodynamic model was well suited to the considerable variations in the lake water surface area in response to water-level fluctuations (Li et al. 2014).

In order to account for the high spatial heterogeneity of the Poyang Lake wetland, the interpretation precision of cover types was analyzed by a fuzzy set-based accuracy assessment. From the confusion matrix presented in Table 4, we can see that sparse grass and bare land were the most confusing classes, with omission portions accounting for 26.8% and 25.4%, respectively. The classification accuracy of the Carex community was 89.3%, which was higher than the Phragmites community (87.4%) and Artemisia community (81.5%). About 7.4% of the Phragmites community was wrongly classified as sand. The Artemisia community was often confused with bare land (accounts for 10.5%). Total accuracy of the classification was 84.2% (1,418/1,684), and the Kappa coefficient was 0.81. Assessment of the classification uncertainty indicated that the vegetation information derived from the Landsat 8 remote sensing image was evident in this study.

With fertile sediments and a relative lack of disturbance by human activity, alluvial plains provide an ideal environment for vegetation growth. Therefore, wetland vegetation communities are mainly distributed on the alluvial plains, such as the PLNNR in Wucheng and the NWNNR in Nanji (Figure 5). The Carex community was typically found in the delta front of big rivers and on the flat floodplains of the dish-shaped pit groups. A significant proportion of the Phragmites community was distributed along levees and on the higher plains of the dish-shaped pit groups. The Artemisia community was usually dispersed along river banks, and always mixed with the Carex community and Phragmites communities. Due to the complicated distribution and lower classification accuracy of the Artemisia community, it is not discussed in the following results.

In the CCA biplots (Figure 8), all variables were closely related to Axis 1, indicating an obvious water regime gradient along the first axis. A significant specific proportion (89%, p < 0.01) of the environmental axis could be explained by IDU. Meanwhile, more than 80% of the variance of vegetation composition could be explained by IDU (p < 0.01), suggesting a positive correlation between IDU and the richness of the Carex community, and a negative correlation between IDU and the richness of the Phragmites community. In contrast, IDU was significantly correlated with IDE (r = 0.64, p < 0.01).

This study is among the first to link vegetation distribution and hydrologic conditions at a spatial scale larger than 3,000 km². The Carex community has been shown to favor more hydric environments with longer IDU and deeper IDE, compared with the Phragmites community. The differences in hydrological preferences between the two major communities observed in this study were consistent with previous studies (Hu et al. 2010; Sang et al. 2014; Xu et al. 2015). We know that fluctuations in water level affect plant establishment from the seed bank by stimulating or inhibiting germination, modifying oxygen availability in the soil (and the subsequent concentrations of nutrients and toxic substances), desiccating aquatic plants or inundating terrestrial plants, and changing light availability with changes in depth (Mitchell & Rogers 1985). IDU and IDE affect the distribution of individual communities due to species tolerance of anoxia, as low oxygen availability reduces respiration and growth in nonadapted roots and can eventually lead to the death of root meristems (Laan et al. 1991; Van den Brink et al. 1995). Under these conditions, some microorganisms use electron acceptors other than oxygen for respiration, resulting in the formation of potentially phytotoxic metal ions such as Fe²⁺ (Laan et al. 1989) and Mn²⁺ (Waldren et al. 1987). Moreover, the accumulation of harmful gases, such as ethylene, can also damage plant organs or at least limit plant growth (Visser et al. 1997). Additionally, IFR can affect species richness when an intermediate frequency creates establishment opportunities for species and prevents competitive exclusion (Bornette et al. 1994) consistent with the intermediate disturbance hypothesis (Connell 1978).

For the three factors examined by CCA, IDU alone had a significant effect on the distribution of plant communities. In contrast, IDE and IFR were less important in this study. Anoxic conditions may develop once a site is inundated. These conditions vary little with depth, and may gain additional discriminatory power. Furthermore, due to the low slope of the delta in the Poyang Lake wetland, slight changes in elevation can play large roles in the influence of IDE on vegetation distribution. The frequency of flood disturbances within grasslands varies in a narrow range (generally 1 to 3 times), under the influence of the subtropical monsoon. This may partially contribute to the limited correspondence between community composition and IFR. Researchers have diverged in their assessments of individual water regimes. For instance, Zweig & Kitchens (2009) showed that IDE was the primary mechanism in driving vegetation community change in the Everglades ecosystem. In contrast, Huang et al. (2013) found that IDE and IFR were not as important as the IDU of flooding. It is noteworthy that water transparency played a key role in the action of IDE. Studies have shown that vegetation types also affect the performance of individual water regimes. Additionally, environmental stresses and human-caused disturbances that may be occurring at multiple scales often obscure or alter the relationships of individual species or communities to hydrologic conditions (Schueler 1994; Gwin et al. 1999).

IDU, IDE, and IFR all affected the distribution of plant communities in certain ways, although for the PLNNR, only IDU was significant when the individual effect of water regimes was examined by CCA. IDU was highly correlated with IDE in this study. However, IFR is usually closely linked with IDE and IDU (high frequency of short shallow floods versus frequency of long deep floods) and thus the effects of IFR can be difficult to isolate (Casanova & Brock 2000). The distribution of plant communities under natural, fluctuating water regimes is a consequence of hydrology changes and their interactions with a suite of other variables, including: nutrients, soil characteristics, competition among species, pathogen activity, fire regimes, and biota (Zweig & Kitchens 2008).

Human activities, especially the construction of TGD, have made the Poyang Lake wetland drier in the 21st century. The proposal for a Poyang Lake Project (PLP, a control structure at the Lake outlet to manage dry season discharge) could provide an effective way to balance the operation of the TGD in the upper Yangtze River and water storage in Poyang Lake (Wang et al. 2015). Operation of PLP will undoubtedly change the water level fluctuations and hydrological processes of the lake, and subsequently affect wetland vegetation, water quality, and the migration of aquatic species. For instance, the operation of PLP will raise the water level of Poyang Lake in the dry season. Sustained high water levels will increase the depth and duration of inundation over the wetlands. As a result, longer IDU and deeper IDE may lead to the degradation of floating and submerged vegetation at lower elevations, and the succession of emergent aquatic vegetation to mesophytic and semi-aquatic vegetation at higher elevations. The hydrological preferences of the two major vegetation communities indicated in this study provide a series of practical references for the management of hydro-engineering projects.

This study was financially supported by the National Natural Science Foundation of China (41371062), National Basic Research Program of China (973 Program) (2012CB417003), and the Collaborative Innovation Center for Major Ecological Security Issues of Jiangxi Province and Monitoring Implementation (JXS-EW-00). The authors express their gratitude to the Hydrological Bureau of Jiangxi Province and the Hydrological Bureau of the Yangtze River Water Resources Commission for providing data.