Next Article in Journal
GNSS/RNSS Integrated PPP Time Transfer: Performance with Almost Fully Deployed Multiple Constellations and a Priori ISB Constraints Considering Satellite Clock Datums
Next Article in Special Issue
Developing a New Vegetation Index Using Cyan, Orange, and Near Infrared Bands to Analyze Soybean Growth Dynamics
Previous Article in Journal
A Novel Forest EcoSpatial Network for Carbon Stocking Using Complex Network Theory in the Yellow River Basin
Previous Article in Special Issue
Comparing Different Spatial Resolutions and Indices for Retrieving Land Surface Phenology for Deciduous Broadleaf Forests
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Remote Sensing
Volume 15
Issue 10
10.3390/rs15102614
remotesensing-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Vegetation Dynamic in a Large Floodplain Wetland: The Effects of Hydroclimatic Regime

by
Lei Jing
1,2,3,
Qing Zeng
1,
Ke He
1,
Peizhong Liu
1,
Rong Fan
1,
Weizhi Lu
2,
Guangchun Lei
1,
Cai Lu
1,* and
Li Wen
4
1
School of Ecology and Nature Conservation, Beijing Forestry University, Beijing 100083, China
2
National Engineering Laboratory for Applied Technology of Forestry & Ecology in South China, Central South University of Forestry and Technology, Changsha 410004, China
3
Key Laboratory of Soil and Water Conservation and Desertification Combating of Hunan Province, Central South University of Forestry and Technology, Changsha 410004, China
4
Science, Economics, and Insights Division, Department of Planning and Environment, Lidcombe, NSW 2000, Australia
*
Author to whom correspondence should be addressed.
Remote Sens. 2023, 15(10), 2614; https://doi.org/10.3390/rs15102614
Submission received: 22 April 2023 / Revised: 13 May 2023 / Accepted: 14 May 2023 / Published: 17 May 2023
(This article belongs to the Special Issue Remote Sensing of Climate-Vegetation Dynamics and Their Effects on Ecosystems)
Browse Figures
Review Reports Versions Notes

Abstract

:
Floodplain wetlands are among the most dynamic ecosystems on Earth, featuring high biodiversity and productivity. They are also sensitive to anthropogenic disturbances and are globally threatened. Understanding how flow regime drives the spatiotemporal dynamics of wetland habitats is fundamental to effective conservation practices. In this study, using Landsat imagery and the random forest (RF) machine learning algorithm, we mapped the winter distribution of four wetland habitats (i.e., Carex meadow, reedbed, mudflat, and shallow water) in East Dongting Lake, a Ramsar wetland in the middle to lower Yangtze Basin of China, for 34 years (1988–2021). The dynamics of wetland habitats were explored through pixel-by-pixel comparisons. Further, the response of wetland habitats to flow regime variations was investigated using generalized additive mixed models (GAMM). Our results demonstrated the constant expansion of reedbeds and shrinkage of mudflats, and that there were three processes contributing to the reduction in mudflat: (1) permanent replacement by reedbed; (2) irreversible loss to water; and (3) transitional swapping with Carex meadow. These changes in the relative extent of wetland habitats may degrade the conservation function of the Ramsar wetland. Moreover, the duration of the dry season and the date of water level withdrawal were identified as the key flow regime parameters shaping the size of wetland habitats. However, different wetland vegetation showed distinct responses to variations in flow regime: while Carex meadow increased with earlier water withdrawal and a longer dry season, reedbed continuously expanded independent of the flow regime corresponding to the increase in winter rainfall. Our findings suggested that flow regime acts in concert with other factors, such as climate change and sand mining in river channels, driving wetland habitat transition in a floodplain landscape. Therefore, effective conservation can only be achieved through diverse restoration strategies addressing all drivers.

1. Introduction

Floodplain wetlands of large rivers are among the most productive ecosystems, with rich biodiversity, and are of high ecological and socioeconomic importance [1,2,3]. Floodplains are also highly dynamic fluvial systems of great ecological complexity driven by periodic inundations associated with the high and low flows of the parent rivers (flood pulses) [4]. In a floodplain with natural hydrological regimes, wetlands are characterized by their transience, being either vegetated (marshes and forest swamps), bare (mudflats), lentic (shallow lakes and lagoons) or lotic (channels) waters. The diverse hydrogeomorphic features of wetlands provide heterogenous habitats for a range of flora and fauna [5,6], and are fundamentally linked to regional biodiversity, having very high conservation value [7,8].
The ecological services and values of floodplain wetlands, particularly as wildlife habitats that support regional biodiversity, depend on their spatial and temporal environmental heterogeneity [2,9]. Floodplain biotas exhibit many adaptative mechanisms to take advantage of this heterogeneity [10,11,12]. Even relatively short episodes of wet and dry (days to weeks) can profoundly affect a floodplain plant community [13], which may have cascadic effects on many animals that inhabit a floodplain [14,15]. For example, in the middle to lower Yangtze floodplains, lush Carex meadows can develop in a newly exposed lakebed days after water level withdrawal [16]; and a study found that the wintering waterfowls changed foraging behaviors to trace newly geminated grass for maximum food intake [17].
Floodplain wetlands have long been recognized as sensitive to anthroponotic disturbances such as land reclamation and river regulation [18,19,20], and are globally threatened ecosystems [21,22]. For example, 80–90% of the floodplains in Europe are lost to agriculture and urban development [23]. In North America, nearly 90% of the floodplains are ‘cultivated’ and non-functional according to Erwin [24]. Most rivers in Australia’s Murry–Darling Basin are regulated by dams and weirs, resulting in extensive wetland decline and degradation [25,26]. The loss of floodplain wetlands in large Asian river basins, such as the Yangtze [27], the Mekong [28], and the Ganges [29], is also alarming and still proceeds at a high rate. In the context of accelerating global warming, continuously shrinking floodplains will be inevitable without effective management strategies [30].
Human disruptions to hydrological regimes, such as dam operations on the river channel and embankments within the floodplain, can substantially affect floodplain vegetation succession and the distribution pattern of vegetation patches [31,32,33]. In most cases, the flow regulation reduces the hydrological variation in floodplain wetlands, which can lead to vegetation encroachment on the floodplain, enabling the expansion of invasive vegetation communities with low species diversity, such as Phragmites monoculture [34,35,36]. Changes in the spatial pattern of vegetation communities aggravate the decline of the ecosystem functions of floodplains through habitat fragmentation and degradation [37], ultimately leading to the loss of regional biodiversity. To mitigate the detrimental impacts of anthropogenic activities, restoration measures are practiced worldwide, aiming to re-establish fundamental hydrological parameters, such as the magnitude, frequency, duration, timing, and rate of change in river flows, that drive ecosystem functions [38]. Quantitative analyses of the relationships between flow regime and distribution dynamics of a vegetation community are necessary to understand the interaction between physical and biological processes for effective ecological restoration [39]. However, these interactive mechanisms are difficult to quantify due to the inherent complexity of vegetation dynamics and of the numerous interactions and feedbacks among different processes, such as erosion and sedimentation, vegetation recruitment and colonization [40].
Extensive floodplains occur in the middle to lower Yangtze River [41,42], supporting regional and global biodiversity [43]. Recently, many studies have provided multiple lines of evidence to suggest that the accelerating loss of floodplain functions in the region are due to economic development and climatic change [44,45,46], prompting enormous conservation efforts to improve or restore floodplains and their associated aquatic habitats [47]. One priority is to re-establish the key elements of the flow regime that sustains the diverse wetland types [18] which support the variety of biota depending on floodplains in their life cycles [10,11,12], including fish [48], waterbirds [16] and others (e.g., mammals, reptiles and amphibians) [43,49]. In this context, knowledge of the quantitative relationships between flow regime and vegetation dynamics, such as how inundation timing and duration shape the pattern and condition of vegetation mosaic [33], is pressingly needed.
Spatiotemporal analysis of floodplain vegetation cover using a time series of satellite images is approved as cost-effective for wetland monitoring [50], especially for assessing the impacts of flow regulation on floodplain ecosystems [51,52], evaluating restoration projects, and inferring future changes [53,54,55]. Together with flow regime variables calculated from in situ hydrological gauge records and climate observations, it is possible to determine the driving factors of observed changes [56], providing scientific insights to design effective restoration strategies [38].
Many previous studies have mapped wetland vegetation cover changes in Dongting Lake [57,58,59,60]. However, these maps have only limited usability in quantifying the relationship between flow regime and wetland habitat dynamics due to the constraints of either spatial and/or temporal resolution, or insufficient number of years of data. In this study, we aimed to identify the key flow regime parameters that drive the temporal variation of the four major wetland types (i.e., Carex meadows, reed marshes, mudflats, and shallow waters) in East Dongting Lake, a Ramsar site and one the most prominent floodplains in the Yangtze Basin (Figure 1). Our approach involved creating a 34-year time series (1988–2021) of wetland maps using 383 Landsat images. With the time series of yearly habitat types, we then investigated how flow regime parameters affected the interannual variations of habitat using generalized addition modelling (GAM) [61]. Finally, the loss and gain of each wetland habitat during the study period was calculated at the pixel level by comparing the habitat distribution in 1988 and 2021, and the dominant signals of change for each class were identified [62,63]. This transition matrix analysis is of particular interest to restoration planning and practice as different habitats provide distinct ecological functions, for example, Carex meadows and mudflats are the main foraging ground for wintering geese [17], shallow waters are a key habitat of wintering swans and storks [41], and reed swamps provide roosting and nesting sites for colonial herons, glebes and egrets [64,65], but are avoided by many migratory waterbirds. Our specific research hypotheses were: (i) hydrological regime parameters act on habitat types differentially with less effect on perennial reed beds due to their persistence; and (ii) the long-term expansion of reedbed causes the variation of other habitats, especially Carex meadows.

2. Materials and Methods

2.1. Study Site

Dongting Lake (28°30′N–30°20′N, 111°40′E–113°10′E, Figure 1) is one of most prominent freshwater lakes in the middle to lower Yangtze Basin, and is renowned for its importance for regional biodiversity [43]. The lake is divided geographically and administratively into three sections: East Dongting Lake, South Dongting Lake, and West Dongting Lake, which are connected by main river channels. This study focused on the largest section, East Dongting Lake (hereafter referred to as EDT).
Influenced by the prevailing subtropic monsoon climate, EDT has distinct wet and dry seasons with an annual water level fluctuation of 12–14 m [57]. During the wet season (April to September), the water level at Chenglingji hydrological station (Figure 1) reaches a maximum level of up to 36 m, and the entire lake virtually becomes a single water body. During the dry season of October to March, the lake water level decreases to a minimum of ~19 m. With water levels receded, a vast lakebed is exposed, and Carex wet meadows develop rapidly, colonizing a large part of the lake [27]. The lake landscape is dominated by connected and disconnected vegetated mosaics separated by mudflats and shallow waters, providing wintering grounds for hundreds and thousands of migratory waterbirds [16]. However, this spatial heterogeneity that maintains the conservation function is at risk due to the de-coupling of the river–lake relationship [18], calling for effective restoration strategies.

2.2. Data Sources and Preparation

Wetland habitat mapping involves four major steps (Figure 2): extracting Landsat images and computing surface reflectance-derived spectral indices; preparing sampling points; building and verifying random forest models; and prediction.

2.3. Habitat Type

To confirm that the samples could be used to model habitat extent and distribution under all hydrological conditions, we selected two maps: the 1993 map representing the dry condition (mean January water level at Chenglingji was 20.05 m), and the 2013 map representing the wet condition (mean January water level at Chenglingji was 22.18 m). With these two maps, we produced a new map where the habitat was the same in both years, through raster overlaying. Using stratified random sampling of the new map, we produced a total of 4090 samples, representing Carex meadow (628), mudflat (336), reedbed (1567), and open water (1559).

2.4. Predictor Variables and Pre-Processing

As the study area is located in a humid subtropic zone with a monsoon climate, cloud contamination is an issue limiting the availability of quality space-borne optical imagery for land cover mapping [66], especially for investigating habitat dynamics using time series data [67]. Using a benign 15% cloud cover threshold, there were only 241 images for the study period, and there were large gaps for many years. However, as cloud cover is calculated for the entire scene in the USGS system, applying a threshold could exclude many quality images of the study area, which covered only a proportion of the theme. For example, the image for 1 April 2004 had 60% cloud cover, but the percentage for the study area was only a negligible 1.22, and of good quality for habitat mapping (Figure A1).
To minimize the gaps in the time series, we downloaded all Landsat images (Level 2, Collection 2, Tier 1, a total of 722) for the study period (January 1988 to March 2022) using the Google Earth Engine platform. Using the pixel quality band (i.e., band 18, USGS, 2021 and 2022), we masked out all cloud and cloud shadow pixels, and recalculated the cloud cover for the study area. Images with cloud cover of less than 15% were selected for further processing (Figure 2).
Many Landsat-based spectral indices have been developed [68], and the list is increasing [69]. Considering the distinct seasonal dynamics of the habitats (Figure 1), we selected vegetation indices that are more sensitive to vegetation phenology (Table 1). Further, the bare soil index (BSI) and normalized difference water index (NDWI) were included to maximize the differentiation of mudflat and water from vegetated areas.
We used the spline function in R (R Development Core Team 2019) to impute the permanent gaps in the time series of the above mentioned seven indices. The method by Fritsch and Carlson (1980) [77] was adopted to fit a piecewise monotonic cubic function for each index, which was then used to predict the missing data in the time series.
Six yearly statistics were derived from the filled raster time series: mean, range (max–min), and trend for both wet and dry season. The wet season included the calendar months of April to October (seven months) and the dry season from November to March next year (five months). For computing convenience, i.e., the calculations of yearly statistics of VI in wet and dry phases, we shifted the time series back by three months, so that the wet season was January to May, and the dry season was June to December. The trend was computed as the slope of the linear regression between the index value (y) and day of the year (x).

2.5. Random Forest Models for Wetland Habitat Classification

We used the random forest (RF) approach [78] to map the yearly distribution of the four habitat types. RF is a robust supervised machine learning classifier widely used in land cover and land use mapping [79,80]. It is among the top machine learning algorithms to discriminate both inland [81] and coastal wetlands [82,83]. The main advantages of RF include: (1) the classifier can effectively handle high data dimensionality and multicollinearity [80], which is a common issue in environmental data; (2) its algorithm is insensitive to overfitting; and (3) the computing cost is relatively low while performing comparably to the more computational demanding approaches such as convolutional neural networks [84,85].
Although RF and machine learning techniques in general are relatively robust to multicollinearity of predictor variables [86], highly correlated predictors could render lower model accuracy [87]. Therefore, we performed a VIF (variable inflation factor) test, and a VIF greater than 10 was excluded from model fitting [88].
The dataset was split into training (75%) and testing (25%) subsets using stratified random sampling. With the training dataset, the repeated (five times) 10-fold cross-validation procedure was used to fine tune the model to optimize its performance. To avoid overfit and increase prediction power, we used the train function in the R package “Caret” [89] to find the best model parameters. We defined a search grid with the two most important parameters (mtry—the number of variables that is randomly selected at each split; and ntree—the number of branches that will grow after each split) [90]. All other parameters such as the sampling theme and the minimal node size were kept as default. Using the testing dataset, we calculated the overall, producer, and user accuracy (OA, PA, and UA, respectively) of wetland classifications. We also evaluated the model performance with the Cohen’s kappa statistic, which considers the expected error rate. While kappa = 1 indicates prefect model prediction, kappa = 0 is interpreted as no agreement between observations and predictions.

2.6. Wetland Habitat Dynamics during 1989–2021

To explore the change in the extent and distribution of wetland habitats, we conducted a grid-by-grid comparison between habitat maps of the first (1988) and the last (2021) years: the two habitat maps were overlaid to produce a transition matrix of the four habitat types. The changes in habitat types were quantified in terms of gross gains, gross losses, and persistence, as well as net and swap changes. The swap change refers to gross loss of type in one area with a gross gain of the same type in another location, or vice versa [63]. With the class transition matrix, we further computed the loss-to-persistence ratio (i.e., loss/persistence) and gain-to-persistence ratio (i.e., gain/persistence) to assess the tendency of each wetland type to lose to, and gain from, other types, which was an approach developed by Braimoh (2006) [62].

2.7. Drivers of Wetland Vegetation Dynamics

The hydrological regimes were quantified with nine variables: date of rising (DOYR, day of year—DOY), rate of water level rising (ROR, m/day), date of withdrawal (DOYW, DOY), duration of low water season (DOD, days), rate of water level withdrawal (ROW, m/day), mean water level during high water season (MHH, m), mean water level during low water season (MHL, m), peak water level (m), and lowest water level (m). These hydrological metrics characterize the magnitude, timing, duration, and rate of change constituting the key flow components that regulate riverine ecosystems [91]. The raw daily water level records series (1988–2021) at Chenglingji was first smoothed using singular spectrum analysis (SSA) [92]. The hydrological regime variables were then derived from the smoothed water level curve. Total rainfall and mean temperature during dry season were calculated from climate records from the weather station at Yueyang.
We investigated the effects of hydrological regime and climate on habitat dynamics using generalized additive mixed models (GAMM) implemented in the R (http://cran.r-project.org; last visited 16 June 2022) package ‘mgcv’ [93], which accounts for complex nonlinear patterns [61]. To reduce the confounding effects of human activities, such as plantation and winter cropping, we restricted the investigation to the “natural area” (Figure 2), which is relatively free of levees and weirs, by manually inspecting the aerial photos. For each year, the predicted distributions of the habitats within this area were extracted. The GAMM modelled the proportions of meadow, mudflat and reed in response to hydrological regimes and climatic variables.
Instead of fitting three separate GAM, we included habitat type (Type) as a random effect in the model formula to investigate the habitat-specific effect of predictor variables:
gam (Y = Type + Year + X1 + X2 + s(Year, by = Type)+ s(X1, by = Type) + s(X2, by = Type)
where Y is the area percentage of each habitat type, Type is factor variable of the three habitat types, Year is the sampling year, and X1 and X2 are independent variables (i.e., hydrology or climate). We included Year in the model to account for overall trends of habitat and the autocorrelation of the time series data.
As the response variables are proportions (i.e., bounded by 0 and 1), we specified a beta distribution with a logit-link function. This link function also takes into account the observed and unobserved heterogeneity in the data [94]. Due to the small sample size (i.e., 34 years of data), we limited the number of covariates to three to keep the risk of overfitting low. Covariate selection was based on the generalized likelihood ratio test (GLRT) [61]. However, in cases of non-significant test (p > 0.05, models are not significantly different), both models were selected and reported on. Moreover, when fitting the GAMM, we added an additional penalty (i.e., the double penalty approach) to remove redundant covariates [95]. The additional penalty affects the functions in the null space of the original penalty. Thus, if all the smoothing parameters for a variable tended to infinity, this variable would be excluded from the final model [95]. Finally, we checked the model residuals for normality by visually inspecting the residual plots and autocorrelation using the Durbin–Watson test [96].

3. Results

3.1. Model Accuracy Assessment

All RF models had high overall accuracy and could be considered as “almost perfect” based on the kappa statistics [97] as the kappa coefficient of agreement was greater than 0.90 for all models. The lowest kappa (0.93) belonged to the 2002 model, and the model for 2014 had the highest kappa of 0.98 (Figure 3). Moreover, as the overall accuracy and kappa coefficient of agreement were comparable for training and testing, there was little evidence of overfitting.
At the class level, accuracies were also high (Figure 4). Overall, mudflat had the lowest producer’s and user’s accuracy. The variance for mudflat was also the highest (i.e., the largest standard deviation, Figure 4), and RF had a tendency to underestimate mudflats. The misclassification mainly came from two sources. The first was the confusion between water and mudflat, and to a lesser degree, between mudflat and meadow (Table 1). Reed and water had comparably high accuracy for both PA and UA (Figure 4), and PA and UA were 100% in many cases. The discrimination capacity for meadow was in the middle (Figure 4), and most of the errors were from the confusion between meadow and mudflat (Table 2). The high accuracy ensured the trustworthy habitat mapping of further analyses.

3.2. Spatial Extent of Habitat Transitions

The spatial extent of habitat loss, gain, and persistence during the study period of 34 years is presented in Figure 5. Overall, nearly 70% of the habitats mapped in 1988 remained the same in 2021 (Figure 5). However, the expansion of reed was extensive in some places, and localized erosion and lakebed build-up was also identified.
Transitions of habitat from one type to other types over the 34 year study period, including gains, losses, persistence, total changes, and swaps, are summarized in Table 3. The spatial extent of habitat loss, gain, and persistence during the study period of 34 years is presented in Figure 5. At lake scale, the total changes were dominated by mudflat (TC was 22.38%, 15.88%, 13.68% and 5.74%, for mudflat, meadow, water, and reed, respectively, Table 3). Mudflat, meadow and water had comparable swamp change, but reed had the lowest swamp change. While reed and meadow had similarly high relative persistence (84% of the land occupied by reed or meadow in 1988 remained as the same in 2021, Table 3), mudflat had the lowest relative persistence and 57% of the 15,320 ha mudflat area mapped in 1988 transferred to Carex meadow (4381 ha), water (3594 ha), or reedbed (748 ha) (Table 3). While reed had the least potential to change to other habitats, it also had the largest potential to gain.
Over the 34 years, mudflat was the only habitat that suffered net loss (5444 ha). A total of 8723 ha of the original mudflat was lost to meadow (4381 ha), water (3594 ha), and reed (748 ha). Although reed was the smallest habitat type in Dongting Lake, it had the most relative gain (the total area increased 53% from 4157 ha to 6187 ha, Table 3). Most of this gain resulted from expansion into meadow and mudflat. More than three thousand hectares of the original meadow was lost, mostly due to the encroachment of reed (1507 ha). However, this loss was overly compensated by expansion into mudflat (4381 ha).

3.3. Drivers of Vegetation Development in East Dongting Lake

Two models fitted the data equally well (p = 0.49, GLRT test): one had DOD (duration of dry season, hereafter referred to as duration model), and the other had DOYD (dates of water level withdrawal, hereafter referred to as date model) as covariate (Table 4). Although the date model had slightly lower AIC (Akaike information criterion) than the duration model, both models explained the majority of the deviance (98.6%) in the data and had a high adjusted R2 of 0.984. The two models produced similar response curves for year and rainfall (Table 4, Figure 6 and Figure 7). Note that the two flow regime variables (DOD and DOYD) were significantly negatively correlated (Pearson’s correlation coefficient = −0.69).
Despite the overall effects of year and rainfall not being significant, the response curves were distinct for the three habitat types. The meadow area did not vary significantly over the study period (i.e., both models estimated that the effect of year was zero). The reed area increased almost linearly over the period of 1988–2021, while the mudflat area was relatively constant before year 1998, after which it decreased almost linearly (Figure 6).
Both duration of dry season and date of water level withdrawal had no effect on the reed area (p = 0.465 and 0.302 for the DOS and DOYD, respectively, Table 4 and Figure 7). While the meadow area was positively related to the duration of dry season, it decreased with the date of water level withdrawal, and the effects were significant and more or less constant (i.e., straight response curves, Figure 7). The effect of the two hydrological regime variables on the mudflat area was highly nonlinear. The mudflat area decreased significantly with the day of the year when the water level started to decline. However, if the water level withdrawal began after late September (i.e., DOY > 270), the effect was positive but small (Figure 8). Mudflat had a similar nonlinear response curve to the duration of dry season (Figure 8), but the effect was insignificant (p = 0.068, Table 4).
The effects of dry season rainfall on meadow were small and insignificant while its effects on mudflat and reed were nonlinear and significant (Table 4 and Figure 8). The mudflat area decreased with the increase in dry season rainfall, although the decreasing slope was small at the two extreme ends (i.e., too dry or too wet, Figure 8). The reed area increased with rainfall until ~300 mm, after which the effect of rainfall was negative (Figure 8).

4. Discussion

4.1. Overall Evaluation of Mapping Approach

Medium resolution satellite imagery, such as the 30 m Landsat TM, ETM+, or OLI, is often preferred for mapping wetland vegetation at large scale [98]. Although high performance classification can be achieved using a single date satellite image [57,99], multitemporal data that capture the vegetation phenology are often more robust [100,101] due to the highly dynamic nature of the floodplain landscape [2,13]. This study retrospectively created a long-term time series (1988–2021) of yearly wetland vegetation cover in the EDT floodplain using multitemporal Landsat images. Our approach had three main advantages that contributed to the superior classification power (Figure 3 and Figure 4) despite the slight underestimation of mudflats and overestimation of water.
First, we maximized the number of available images used in vegetation indices computing. The study area is located in a subtropical monsoon region; therefore, high quality satellite images are limited if selected by the percentage of cloud cover over the whole scene [51,57,99,101]. A previous study using an image fusion of Landsat and MODIS (moderate resolution imaging spectroradiometer) data to address this issue achieved acceptable performance [67]. Recalculation of cloud cover with pixel quality assessment (band 18) greatly increased the number of useable images (i.e., a total of 364). The high frequency image collection justified the further procedures to extract the vegetation phenology for classification.
Second, lack of ground-truth samples is an obvious drawback for mapping historical vegetation distribution retrospectively. Many studies have used historical high-resolution satellite and/or aerial photos available from Google Earth or other platforms (such as Nearmap Australia, website accessed on 29 March 2022, https://www.nearmap.com/au/en) to compile a testing dataset and model training [101,102]. Here, we overlaid two historical vegetation maps representing dry and wet conditions more than 10 years apart to create stratified random samples. The approaches captured the characteristic wetland types for the whole study period and guaranteed the ability of the samples for the entire study period.
Finally, we extracted the essential phenological variables for classification. The four wetland covers had distinct signatures in terms of the 42 vegetation indices (Figure 2). For example, the NDVI values for reedbed increased in wet but decreased in the dry season, and Carex meadows had the opposite trend [27]. The combination of these phenological variables resulted in accurate and precise classification. Nevertheless, there were uncertainties between mudflat and water, both of which had very low vegetation indices values and no clear trends in wet or dry seasons. The inclusion of image texture features [101] might improve the discrimination of the two classes.

4.2. Drivers of Wetland Vegetation Dynamics

The mapping of the floodplain wetland types was not a primary objective of this study. The long-term time series of yearly maps was created to investigate how wetland habitats respond to river flow variation for designing effective restoration.
Overall, the duration of the low water season and date of water level withdrawal were found to be the primary drivers of wetland habitat dynamics. Many studies demonstrate that vegetation patterns in floodplains are driven by the hydroperiod, i.e., the duration and temporal pattern of inundation [57,58,91]. Spatial variations in the hydroperiod, which can result from minor differences in elevation (i.e., microtopographic pattern), often produce dramatic differences in vegetation across floodplains [103]. However, the effects of all other tested flow regime variables, such as the rate of water level withdrawal, mean water levels during high and low water seasons, and the lowest water level, were insignificant. The results were largely consistent with our previous findings [57], which found that the development of Carex meadow in the EDT floodplain was positively related to the date of water level withdrawal. However, these findings were at odds with Peng et al. [101], who reported the close correlations between the size of reedbed and Carex meadow area and water level at different seasons. The main reason for this disagreement might be the different focus areas. To reduce the confounding impacts of human activities on the floodplain, we excluded areas with reed plantation and winter cropping (Figure 2), whereas Peng et al. [101] investigated the entire lake.

4.2.1. Reedbed Is Highly Persistent and Has Rapidly Expanded during the Study Period

Throughout the study period, the majority of original reedbed patches remained (Table 3 and Figure 5), showing the greatest persistence to changes in flow regime (Figure 8). On the other hand, the total reedbed area increased 2337 ha (or 62%) by encroaching into Carex meadows (1507 ha) and mudflats (748 ha) over the 34 years. Expansion of reedbeds is a worldwide phenomenon that has biodiversity concern [104]. The gradual encroachment of reedbeds (Figure 6) into Carex meadows and mudflats will be a great threat to the hundreds and thousands of waterbirds such as geese and swans wintering in EDT [16,17].
Surprisingly, none of the flow regime variables had significant effects on the expansion of reedbed area (Figure 6 and Figure 7, Table 4). The lack of relationship with flow regime could be linked with the life form of Phragmites, which is a tall (up to 4 m) rhizomatous grass, enabling it to survive the maximum water depth at areas currently occupied by Carex meadows and mudflats. Moreover, Phragmites seeds spread copiously and clonally by a vigorous system of rhizomes and stolons [105], therefore, can proliferate other habitats through both clonal and sexual reproduction once recruited and established in wetland [106]. Other broader scale factors, such as the increase in winter rainfall associated with climate change [107], and the elevated atmospheric CO2 level (108), might be responsible for the continuous spread of this cosmopolitan species.

4.2.2. Carex Meadow Has No Trend and Fluctuates with the Changes in Flow Regime

Although the mapped area of Carex meadow in 2021 was greater than that in 1988 (Table 3), the total area of Carex meadow varied greatly between years and showed no trend (Figure 6), which was consistent with the findings of a previous study [101]. The interannual variations of Carex meadow area in EDT were the balanced outcome of loss to reedbed and gain from mudflat. While loss to reedbed was permanent (i.e., once an area was replaced by reed, it remained as reed), the gain from mudflat was somehow transient (i.e., there were many pixels switching between Carex meadow and mudflat). In addition, the switching between Carex meadow and mudflat contributed more to the year-to-year changes in total Carex meadow area, and the majority of pixels that had high tendency to swap were located at the transient zone between the two habitats (Figure 5). A previous study found that the wintering geese preferred to forage in this transient zone [16], indicating its conservation importance.
The results of GAM indicated that the Carex meadow area increased with the duration of the dry season and decreased with the day of the year when the water level started to withdraw (Figure 8), which was consistent with both field observation and lab gemination experiment [108]. The importance of the water level withdrawal date might be related to the temperature requirement of the gemination of the species. In the middle Yangtze region, the rhizomes of the tufted perennial grass Carex spp. generally cannot regerminate when the air temperature is below 6 °C [109]. As the DOY of the water level withdrawal increases, the air temperature decreases, reducing the colonization of Carex meadows.

4.2.3. Mudflat Is the Most Vulnerable Habitat

Being the only habitat type to show significant decrease during the 34-year period (Figure 6), mudflat had the lowest persistence/total value (Table 3), and was vulnerable to being replaced by other habitat types. By inspecting the yearly habitat maps, three processes of loss were identified: (1) permanent replacement by reedbed; (2) irreversible loss to water (i.e., erosion); and (3) transitional swapping with Carex meadow. Moreover, the switch to and from mudflats was not symmetrical, with a higher tendency of shifting from mudflat to Carex meadow, resulting in the slow advance of the transient zone towards water. The accelerated erosion might be attributed to sand mining, a serious threat identified in many river systems of the Yangtze Basin [42].
Mudflats have long been recognized as important foraging habitat for shorebirds in coastal areas [110]. However, the ecological function of mudflats in freshwater floodplains are largely overlooked; and most restoration projects focus on vegetated areas and shallow waters. Nevertheless, they are also important resting and foraging habitats for a range of wintering waterbirds in the Yangtze region [65]. Many waterbirds, such as swans and storks, prey on organisms dwelling in mudflats [111]. The loss and vulnerability of mudflat in EDT should be closely monitored.

5. Conclusions

Restoration of floodplain wetland associated with large rivers is vital for global biodiversity. Understanding how flow regime shapes wetland vegetation cover is critical for effective restoration. In this study, we built a 34 year (1988–2021) time series map of wetland habitat using Landsat imagery and machine learning algorithm. The spatiotemporal dynamics of four broad habitats were further investigated using GAMM. Our analysis revealed the constant expansion of reedbeds and shrinkage of mudflats, which might degrade the conservation function of the Ramsar wetland. The GAMM results showed that the duration of dry season and the date when the water level started to withdraw were the key flow regime parameters acting on the extent of wetland habitats. Moreover, the responses of different wetland vegetation to flow variations were distinct. Specifically, while Carex meadow increased with an earlier water withdrawal and longer dry season, reedbed continuously spread out independent of flow regime.

Author Contributions

Conceptualization, L.J., L.W., Q.Z. and C.L.; data curation, L.W.; formal analysis, G.L. and L.W.; funding acquisition, W.L. and G.L.; investigation, L.J., K.H., P.L. and R.F.; methodology, L.J. and L.W.; project administration, W.L. and C.L.; software, L.W.; supervision, L.W.; validation, L.J., G.L. and C.L.; visualization, L.J. and C.L.; writing—original draft, L.J., Q.Z. and C.L.; writing—review and editing, L.W. and C.L. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by the National Key Research and Development Program of China, grant number 2017YFC0405300; the National Natural Science Foundation of China, grant numbers U20A2089 and 41971152; and the Natural Science Foundation of Hunan Province, China, grant number 2021JJ20012.

Data Availability Statement

Data will be made available on request.

Acknowledgments

The authors thank Changjiang Water Resource Commission for providing the hydrological data, and thank the staff of the East Dongting Lake Nature Reserve Authority for their help in the field investigation.

Conflicts of Interest

The authors declare no conflict of interest.

Appendix A

Remotesensing 15 02614 g0a1 550
Figure A1. Landsat 5 images obtained on 1 April 2004 (A) and 7 July 2010 (B) could be excluded if applying a threshold of 25% to select quality imagery. The cloud cover is 60% and 48%, respectively, for all themes (row 40 and path 123).
Figure A1. Landsat 5 images obtained on 1 April 2004 (A) and 7 July 2010 (B) could be excluded if applying a threshold of 25% to select quality imagery. The cloud cover is 60% and 48%, respectively, for all themes (row 40 and path 123).
Remotesensing 15 02614 g0a1

References

  1. Keddy, P.A.; Fraser, L.H.; Solomeshch, A.I.; Junk, W.J.; Campbell, D.R.; Arroyo, M.T.; Alho, C.J. Wet and wonderful: The world’s largest wetlands are conservation priorities. BioScience 2009, 59, 39–51. [Google Scholar] [CrossRef]
  2. Lewis, W.M.; Hamilton, S.K.; Lasi, M.A.; Rodríguez, M.; Saunders, J.F. Ecological Determinism on the Orinoco Floodplain: A 15-year study of the Orinoco floodplain shows that this productive and biotically diverse ecosystem is functionally less complex than it appears. Hydrographic and geomorphic controls induce a high degree of determinism in biogeochemical and biotic processes. BioScience 2000, 50, 681–692. [Google Scholar]
  3. Tickner, D.; Opperman, J.J.; Abell, R.; Acreman, M.; Arthington, A.H.; Bunn, S.E.; Cooke, S.J.; Dalton, J.; Darwall, W.; Edwards, G.; et al. Bending the curve of global freshwater biodiversity loss: An emergency recovery plan. BioScience 2000, 70, 330–342. [Google Scholar] [CrossRef] [PubMed]
  4. Junk, W.J.; Bayley, P.B.; Sparks, R.E. The flood pulse concept in river-floodplain systems. In Canadian Special Publications of Fisheries and Aquatic Sciences, Proceedings of the International Large River Symposium, Honey Harbour, ON, Canada, 14–21 September 1986; Dodge, D.P., Ed.; Department of Fisheries and Oceans: Honey Harbour, ON, Canada, 1989; Volume 106, pp. 110–127. [Google Scholar]
  5. Arthington, A.H.; Godfrey, P.C.; Pearson, R.G.; Karim, F.; Wallace, J. Biodiversity values of remnant freshwater floodplain lagoons in agricultural catchments: Evidence for fish of the Wet Tropics bioregion, northern Australia. Aquat. Conserv. Mar. Freshw. Ecosyst. 2015, 25, 336–352. [Google Scholar] [CrossRef]
  6. Pander, J.; Mueller, M.; Geist, J. Habitat diversity and connectivity govern the conservation value of restored aquatic floodplain habitats. Biol. Conserv. 2018, 217, 1–10. [Google Scholar] [CrossRef]
  7. Olson, D.M.; Dinerstein, E. The Global 200: A representation approach to conserving the Earth’s most biologically valuable ecoregions. Conserv. Biol. 1998, 12, 502–515. [Google Scholar] [CrossRef]
  8. Ward, D.P.; Pettit, N.E.; Adame, M.; Douglas, M.M.; Setterfield, S.A.; Bunn, S.E. Seasonal spatial dynamics of floodplain macrophyte and periphyton abundance in the Alligator Rivers region (Kakadu) of northern Australia. Ecohydrology 2016, 9, 1675–1686. [Google Scholar] [CrossRef]
  9. Wang, Y.; Jia, Y.; Guan, L.; Lu, C.; Lei, G.; Wen, L.; Liu, G. Optimizing hydrological conditions to sustain wintering waterbird populations in Poyang Lake National Natural Reserve: Implications for dam operations. Freshw. Biol. 2013, 58, 2366–2379. [Google Scholar]
  10. Adis, J.; Junk, W.J. Terrestrial invertebrates inhabiting lowland river floodplains of Central Amazonia and Central Europe: A review. Freshw. Biol. 2002, 47, 711–731. [Google Scholar] [CrossRef]
  11. Jackson, M.B.; Colmer, T. Response and adaptation by plants to flooding stress. Ann. Bot. 2005, 96, 501–505. [Google Scholar] [CrossRef]
  12. Parolin, P. Submerged in darkness: Adaptations to prolonged submergence by woody species of the Amazonian floodplains. Ann. Bot. 2009, 103, 359–376. [Google Scholar] [CrossRef] [PubMed]
  13. Thapa, R.; Thoms, M.C.; Parsons, M. The response of dryland floodplain vegetation productivity to flooding and drying. J. Arid Environ. 2016, 129, 42–55. [Google Scholar] [CrossRef]
  14. Doledec, S.; Statzner, B. Theoretical habitat templets, species traits, and species richness: 548 plant and animal species in the Upper Rhône River and its floodplain. Freshw. Biol. 1994, 31, 523–538. [Google Scholar] [CrossRef]
  15. Robinson, C.T.; Tockner, K.; Ward, J.V. The fauna of dynamic riverine landscapes. Freshw. Biol. 2002, 47, 661–677. [Google Scholar] [CrossRef]
  16. Guan, L.; Lei, J.; Zuo, A.; Zhang, H.; Lei, G.; Wen, L. Optimizing the timing of water level recession for conservation of wintering geese in Dongting Lake, China. Ecol. Eng. 2016, 88, 90–98. [Google Scholar] [CrossRef]
  17. Lei, J.; Jia, Y.; Wang, Y.; Lei, G.; Lu, C.; Saintilan, N.; Wen, L. Behavioural plasticity and trophic niche shift: How wintering geese respond to habitat alteration. Freshw. Biol. 2019, 64, 1183–1195. [Google Scholar] [CrossRef]
  18. Lu, C.; Jia, Y.; Jing, L.; Zeng, Q.; Lei, J.; Zhang, S.; Lei, G.; Wen, L. Shifts in river-floodplain relationship reveal the impacts of river regulation: A case study of Dongting Lake in China. J. Hydrol. 2018, 559, 932–941. [Google Scholar] [CrossRef]
  19. Nilsson, C.; Reidy, C.A.; Dynesius, M.; Revenga, C. Fragmentation and flow regulation of the world’s large river systems. Science 2005, 308, 405–408. [Google Scholar] [CrossRef]
  20. Tockner, K.; Pusch, M.; Borchardt, D.; Lorang, M.S. Multiple stressors in coupled river-floodplain ecosystems. Freshw. Biol. 2010, 55 (Suppl. S1), 135–151. [Google Scholar] [CrossRef]
  21. Dudgeon, D.; Arthington, A.H.; Gessner, M.O.; Kawabata, Z.I.; Knowler, D.J.; Lévêque, C.; Naiman, R.J.; Prieur-Richard, A.H.; Soto, D.; Stiassny, M.L.; et al. Freshwater biodiversity: Importance, threats, status and conservation challenges. Biol. Rev. 2006, 81, 163–182. [Google Scholar] [CrossRef]
  22. Opperman, J.J.; Galloway, G.E.; Fargione, J.; Mount, J.F.; Richter, B.D.; Secchi, S. Sustainable floodplains through large-scale reconnection to rivers. Science 2009, 326, 1487–1488. [Google Scholar] [CrossRef] [PubMed]
  23. Tockner, K.; Stanford, J.A. Riverine flood plains: Present state and future trends. Environ. Conserv. 2002, 29, 308–330. [Google Scholar] [CrossRef]
  24. Erwin, K.L. Wetlands and global climate change: The role of wetland restoration in a changing world. Wetl. Ecol. Manag. 2009, 17, 71–84. [Google Scholar] [CrossRef]
  25. Chen, Y.; Colloff, M.J.; Lukasiewicz, A.; Pittock, J. A trickle, not a flood: Environmental watering in the Murray–Darling Basin, Australia. Mar. Freshw. Res. 2020, 72, 601–619. [Google Scholar] [CrossRef]
  26. Kingsford, R.T.; Basset, A.; Jackson, L. Wetlands: Conservation’s poor cousins. Aquat. Conserv. 2016, 26, 892–916. [Google Scholar] [CrossRef]
  27. Guan, L.; Wen, L.; Feng, D.; Zhang, H.; Lei, G. Delayed flood recession in central Yangtze floodplains can cause significant food shortages for wintering geese: Results of inundation experiment. Environ. Manag. 2014, 54, 1331–1341. [Google Scholar] [CrossRef] [PubMed]
  28. Mahood, S.P.; Poole, C.M.; Watson, J.E.; MacKenzie, R.A.; Sharma, S.; Garnett, S.T. Agricultural intensification is causing rapid habitat change in the Tonle Sap Floodplain, Cambodia. Wetl. Ecol. Manag. 2020, 28, 713–726. [Google Scholar] [CrossRef]
  29. Singh, M.; Sinha, R. Evaluating dynamic hydrological connectivity of a floodplain wetland in North Bihar, India using geostatistical methods. Sci. Total Environ. 2019, 651, 2473–2488. [Google Scholar] [CrossRef]
  30. Xi, Y.; Peng, S.; Ciais, P.; Chen, Y. Future impacts of climate change on inland Ramsar wetlands. Nat. Clim. Chang. 2021, 11, 45–51. [Google Scholar] [CrossRef]
  31. Benjankar, R.; Jorde, K.; Yager, E.M.; Egger, G.; Goodwin, P.; Glenn, N.F. The impact of river modification and dam operation on floodplain vegetation succession trends in the Kootenai River, USA. Ecol. Eng. 2012, 46, 88–97. [Google Scholar] [CrossRef]
  32. Catford, J.A.; Downes, B.J.; Gippel, C.J.; Vesk, P.A. Flow regulation reduces native plant cover and facilitates exotic invasion in riparian wetlands. J. Appl. Ecol. 2011, 48, 432–442. [Google Scholar] [CrossRef]
  33. Sandi, S.G.; Saco, P.M.; Rodriguez, J.F.; Saintilan, N.; Wen, L.; Kuczera, G.; Riccardi, G.; Willgoose, G. Patch organization and resilience of dryland wetlands. Sci. Total Environ. 2020, 726, 138581. [Google Scholar] [CrossRef] [PubMed]
  34. Bevington, A.E.; Twilley, R.R.; Sasser, C.E. Deltaic floodplain wetland vegetation dynamics along the sediment surface elevation gradient and in response to disturbance from river flooding and hurricanes in Wax Lake Delta, Louisiana, USA. Geomorphology 2022, 398, 108011. [Google Scholar] [CrossRef]
  35. Johnson, W.C.; Volke, M.A.; Scott, M.L.; Dixon, M.D. The dammed Missouri: Prospects for recovering Lewis and Clark’s River. Ecohydrology 2015, 8, 765–771. [Google Scholar] [CrossRef]
  36. Shi, L.; Wang, Y.; Jia, Y.; Lu, C.; Lei, G.; Wen, L. Vegetation cover dynamics and resilience to climatic and hydrological disturbances in seasonal floodplain: The effects of hydrological connectivity. Front. Plant Sci. 2017, 8, 2196. [Google Scholar] [CrossRef] [PubMed]
  37. Knox, R.L.; Wohl, E.E.; Morrison, R.R. Levees don’t protect, they disconnect: A critical review of how artificial levees impact floodplain functions. Sci. Total Environ. 2022, 837, 155773. [Google Scholar] [CrossRef]
  38. Roni, P.; Hall, J.E.; Drenner, S.M.; Arterburn, D. Monitoring the effectiveness of floodplain habitat restoration: A review of methods and recommendations for future monitoring. Wiley Interdiscip. Rev. Water 2019, 6, e1355. [Google Scholar] [CrossRef]
  39. Jorde, K.; Burke, M.; Scheidt, N.; Welcker, C.; King, S.; Borden, C. Reservoir operations, physical processes, and ecosystem losses. In Gravel-Bed Rivers VI: From Process Understanding to River Restoration; Habersack, H., Piégay, H., Rinaldi, M., Eds.; Elsevier B.V.: Leinz, Austria, 2007; pp. 607–636. [Google Scholar]
  40. Entwistle, N.S.; Heritage, G.L.; Schofield, L.A.; Williamson, R.J. Recent changes to floodplain character and functionality in England. Catena 2019, 174, 490–498. [Google Scholar] [CrossRef]
  41. Wang, J.; Sheng, Y.; Tong, T.S.D. Monitoring decadal lake dynamics across the Yangtze Basin downstream of Three Gorges Dam. Remote Sens. Environ. 2014, 152, 251–269. [Google Scholar] [CrossRef]
  42. Wang, H.; Liu, X.; Wang, H. The Yangtze River floodplain: Threats and rehabilitation. In Fishery Resources, Environment, and Conservation in the Mississippi and Yangtze (Changjiang) River Basins; Yushun, C., Chapman, D., Jackson, J., Zhongjie, L., Kilgore, J., Phelps, O., Eggleton, M., Eds.; American Fisheries Society: Bethesda, MD, USA, 2016; Volume 84, pp. 263–291. [Google Scholar]
  43. Fang, J.; Wang, Z.; Zhao, S.; Li, Y.; Tang, Z.; Yu, D.; Ni, L.; Liu, H.; Xie, P.; Da, L.; et al. Biodiversity changes in the lakes of the Central Yangtze. Front. Ecol. Environ. 2006, 4, 369–377. [Google Scholar] [CrossRef]
  44. Du, Y.; Xue, H.P.; Wu, S.J.; Ling, F.; Xiao, F.; Wei, X.H. Lake area changes in the middle Yangtze region of China over the 20th century. J. Environ. Manag. 2011, 92, 1248–1255. [Google Scholar] [CrossRef] [PubMed]
  45. Feng, L.; Han, X.; Hu, C.; Chen, X. Four decades of wetland changes of the largest freshwater lake in China: Possible linkage to the Three Gorges Dam? Remote Sens. Environ. 2016, 176, 43–55. [Google Scholar] [CrossRef]
  46. Zhang, G.; Yao, T.; Chen, W.; Zheng, G.; Shum, C.K.; Yang, K.; Piao, S.; Sheng, Y.; Yi, S.; Li, J.; et al. Regional differences of lake evolution across China during 1960s–2015 and its natural and anthropogenic causes. Remote Sens. Environ. 2019, 221, 386–404. [Google Scholar] [CrossRef]
  47. Chen, T.; Wang, Y.; Gardner, C.; Wu, F. Threats and protection policies of the aquatic biodiversity in the Yangtze River. J. Nat. Conserv. 2020, 58, 125931. [Google Scholar] [CrossRef]
  48. Liu, X.; Wang, H. Effects of loss of lateral hydrological connectivity on fish functional diversity. Conserv. Biol. 2018, 32, 1336–1345. [Google Scholar] [CrossRef]
  49. Dudgeon, D. Requiem for a river: Extinctions, climate change and the last of the Yangtze. Aquat. Conserv. Mar. Freshw. Ecosyst. 2010, 20, 127–131. [Google Scholar] [CrossRef]
  50. Dronova, I. Object-based image analysis in wetland research: A review. Remote Sens. 2015, 7, 6380–6413. [Google Scholar] [CrossRef]
  51. Han, X.; Chen, X.; Feng, L. Four decades of winter wetland changes in Poyang Lake based on Landsat observations between 1973 and 2013. Remote Sens. Environ. 2015, 156, 426–437. [Google Scholar] [CrossRef]
  52. Mohammadi, A.; Costelloe, J.F.; Ryu, D. Application of time series of remotely sensed normalized difference water, vegetation and moisture indices in characterizing flood dynamics of large-scale arid zone floodplains. Remote Sens. Environ. 2017, 190, 70–82. [Google Scholar] [CrossRef]
  53. Chapple, D.; Dronova, I. Vegetation development in a tidal marsh restoration project during a historic drought: A remote sensing approach. Front. Mar. Sci. 2017, 4, 243. [Google Scholar] [CrossRef]
  54. Shuman, C.S.; Ambrose, R.F. A comparison of remote sensing and ground-based methods for monitoring wetland restoration success. Restor. Ecol. 2003, 11, 325–333. [Google Scholar] [CrossRef]
  55. Sims, N.C.; Colloff, M.J. Remote sensing of vegetation responses to flooding of a semi-arid floodplain: Implications for monitoring ecological effects of environmental flows. Ecol. Indic. 2012, 18, 387–391. [Google Scholar] [CrossRef]
  56. Shaeri Karimi, S.; Saintilan, N.; Wen, L.; Cox, J. Spatio-temporal effects of inundation and climate on vegetation greenness dynamics in dryland floodplains. Ecohydrology 2022, 15, e2378. [Google Scholar] [CrossRef]
  57. Jing, L.; Lu, C.; Xia, Y.; Shi, L.; Zuo, A.; Lei, J.; Zhang, H.; Lei, G.; Wen, L. Effects of hydrological regime on development of Carex meadows in East Dongting Lake, a Ramsar Wetland for wintering waterbirds. Sci. Rep. 2017, 7, 41761. [Google Scholar] [CrossRef]
  58. Jing, L.; Zhou, Y.; Zeng, Q.; Liu, S.; Lei, G.; Lu, C.; Wen, L. Exploring wetland dynamics in large river floodplain systems with unsupervised machine learning: A case study of the Dongting Lake, China. Remote Sens. 2020, 12, 2995. [Google Scholar] [CrossRef]
  59. Long, X.; Li, X.; Lin, H.; Zhang, M. Mapping the vegetation distribution and dynamics of a wetland using adaptive-stacking and Google Earth Engine based on multi-source remote sensing data. Int. J. Appl. Earth Obs. 2021, 102, 102453. [Google Scholar] [CrossRef]
  60. Yang, L.; Wang, L.; Yu, D.; Yao, R.; He, Q.; Wang, S.; Wang, L. Four decades of wetland changes in Dongting Lake using Landsat observations during 1978–2018. J. Hydrol. 2020, 587, 124954. [Google Scholar] [CrossRef]
  61. Wood, S.N. Generalized Additive Models: An Introduction with R, 2nd ed.; Chapman and Hall/CRC Press: New York, NY, USA, 2017; pp. 1–496. [Google Scholar]
  62. Braimoh, A.K. Spatial Analysis of Residential Land Use Change in Lagos, Nigeria; UNU-IAS Working Paper No. 144; United Nations University: Tokyo, Japan, 2006. [Google Scholar]
  63. Pontius, R.G.; Shusas, E.; McEachern, M. Detecting important categorical land changes while accounting for persistence. Agric. Ecosyst. Environ. 2004, 101, 251–268. [Google Scholar] [CrossRef]
  64. Okes, N.C.; Hockey, P.A.; Cumming, G.S. Habitat use and life history as predictors of bird responses to habitat change. Conserv. biol. 2008, 22, 151–162. [Google Scholar] [CrossRef]
  65. Rasool, M.A.; Hassan, M.A.; Zhang, X.; Zeng, Q.; Jia, Y.; Wen, L.; Lei, G. Habitat Quality and Social Behavioral Association Network in a Wintering Waterbirds Community. Sustainability 2021, 13, 6044. [Google Scholar] [CrossRef]
  66. Foga, S.; Scaramuzza, P.L.; Guo, S.; Zhu, Z.; Dilley, R.D.; Beckmann, T.; Schmidt, G.L.; Dwyer, J.L.; Joseph Hughes, M.; Laue, B. Cloud detection algorithm comparison and validation for operational Landsat data products. Remote Sens. Environ. 2017, 194, 379–390. [Google Scholar] [CrossRef]
  67. Chen, B.; Chen, L.; Huang, B.; Michishita, R.; Xu, B. Dynamic monitoring of the Poyang Lake wetland by integrating Landsat and MODIS observations. ISPRS J. Photogramm. 2018, 139, 75–87. [Google Scholar] [CrossRef]
  68. Bannari, A.; Morin, D.; Bonn, F.; Huete, A.R. A review of vegetation indices. Remote Sens. Rev. 1995, 13, 95–120. [Google Scholar] [CrossRef]
  69. Gómez, C.; White, J.C.; Wulder, M.A.; Alejandro, P. Historical Forest biomass dynamics modelled with Landsat spectral trajectories. ISPRS J. Photogramm. 2014, 93, 14–28. [Google Scholar] [CrossRef]
  70. Rouse, J.W.; Haas, R.H.; Deering, D.W.; Schell, J.A.; Harlan, J.C. Monitoring the Vernal Advancement and Retrogradation (Green Wave Effect) of Natural Vegetation; E73-10693; NASA: Greenbelt, MD, USA, 1973; p. 112. [Google Scholar]
  71. McFeeters, S.K. The use of normalized difference water index (NDWI) in the delineation of open water features. Int. J. Remote Sens. 1996, 17, 1425–1432. [Google Scholar] [CrossRef]
  72. Merzlyak, M.N.; Gitelson, A.A.; Chivkunova, O.B.; Rakitin, V.Y. Non-destructive optical detection of pigment changes during leaf senescence and fruit ripening. Physiol. Plantar. 1999, 106, 135–141. [Google Scholar] [CrossRef]
  73. Huete, A.R. A Soil-adjusted vegetation index (SAVI). Remote Sens. Environ. 1998, 25, 295–309. [Google Scholar] [CrossRef]
  74. Gitelson, A.A.; Merzlyak, M.N. Signature Analysis of Leaf Reflectance Spectra: Algorithm Development for Remote Sensing of Chlorophyll. J. Plant Physiol. 1996, 148, 494–500. [Google Scholar] [CrossRef]
  75. Rikimaru, A.; Miyatake, S. Development of Forest Canopy Density Mapping and Monitoring Model using Indices of Vegetation, Bare soil and Shadow. In Proceedings of the 18th Asian Conference on Remote Sensing (ACRS), Kuala Lumpur, Malaysia, 20–25 October 1997. [Google Scholar]
  76. Wang, C.; Chen, J.; Wu, J.; Tang, Y.; Shi, P.; Black, T.A.; Zhu, K. A snow-free vegetation index for improved monitoring of vegetation spring green-up date in deciduous ecosystems. Remote Sens. Environ. 2017, 196, 1–12. [Google Scholar] [CrossRef]
  77. Fritsch, F.N.; Carlson, R.E. Monotone piecewise cubic interpolation: Algorithms and software. In Proceedings of the SIAM Fall Meeting, Houston, TX, USA, 6 November 1980. [Google Scholar]
  78. Breiman, L. Statistical Modeling: The Two Cultures. Stat. Sci. 2001, 16, 199–231. [Google Scholar] [CrossRef]
  79. Eisavi, V.; Homayouni, S.; Yazdi, A.M.; Alimohammadi, A. Land cover mapping based on random forest classification of multitemporal spectral and thermal images. Environ. Monit. Assess. 2015, 187, 291. [Google Scholar] [CrossRef] [PubMed]
  80. Rodriguez-Galiano, V.F.; Ghimire, B.; Rogan, J.; Chica-Olmo, M.; Rigol-Sanchez, J.P. An assessment of the effectiveness of a random forest classifier for land-cover classification. ISPRS J. Photogramm. 2012, 67, 93–104. [Google Scholar] [CrossRef]
  81. Mahdianpari, M.; Salehi, B.; Mohammadimanesh, F.; Brisco, B. An Assessment of Simulated Compact Polarimetric SAR Data for Wetland Classification Using Random Forest Algorithm. Can. J. Remote Sens. 2017, 43, 468–484. [Google Scholar] [CrossRef]
  82. Van Beijma, S.; Comber, A.; Lamb, A. Random forest classification of salt marsh vegetation habitats using quad-polarimetric airborne SAR, elevation and optical RS data. Remote Sens. Environ. 2014, 149, 118–129. [Google Scholar] [CrossRef]
  83. Wen, L.; Hughes, M. Coastal wetland mapping using ensemble learning algorithms: A comparative study of bagging, boosting and stacking techniques. Remote Sens. 2020, 12, 1683. [Google Scholar] [CrossRef]
  84. Liu, M.; Shi, J.; Li, Z.; Li, C.; Zhu, J.; Liu, S. Towards Better Analysis of Deep Convolutional Neural Networks. IEEE Trans. Vis. Comput. Graph. 2017, 23, 91–100. [Google Scholar] [CrossRef]
  85. Pouliot, D.; Latifovic, R.; Pasher, J.; Duffe, J. Landsat Super-Resolution Enhancement Using Convolution Neural Networks and Sentinel-2 for Training. Remote Sens. 2018, 10, 394. [Google Scholar] [CrossRef]
  86. Kotsiantis, S.; Kanellopoulos, D.; Pintelas, P. Data Preprocessing for Supervised Learning. Int. J. Comput. Sci. 2006, 1, 111–117. [Google Scholar]
  87. Senawi, A.; Wei, H.-L.; Billings, S.A. A new maximum relevance-minimum multicollinearity (MRmMC) method for feature selection and ranking. Pattern Recogn. 2017, 67, 47–61. [Google Scholar] [CrossRef]
  88. Segura, C.; Caldwell, P.; Sun, G.; McNulty, S.; Zhang, Y. A model to predict stream water temperature across the conterminous USA. Hydrol. Process. 2015, 29, 2178–2195. [Google Scholar] [CrossRef]
  89. Kuhn, M. Caret package. J. Stat. Softw. 2008, 28, 1–26. [Google Scholar]
  90. Probst, P.; Wright, M.; Boulesteix, A.-L. Hyperparameters and Tuning Strategies for Random Forest. Wiley Interdiscip. Rev. Data Min. Knowl. Discov. 2019, 9, e1301. [Google Scholar] [CrossRef]
  91. Poff, N.L.; Allan, J.D.; Bain, M.B.; Karr, J.R.; Prestegaard, K.L.; Richter, B.D.; Sparks, R.E.; Stromberg, J.C. The natural flow regime. BioScience 1997, 47, 769–784. [Google Scholar] [CrossRef]
  92. Golyandina, N.; Korobeynikov, A.; Zhigljavsky, A. SSA Analysis of One-Dimensional Time Series. In Singular Spectrum Analysis with R. Use R! Springer: Berlin/Heidelberg, Germany, 2018; pp. 31–120. [Google Scholar]
  93. Wood, S.N.; mgcv: Mixed GAM Computation Vehicle with Automatic Smoothness Estimation. R Package Version 1.8-23. 2018. Available online: https://cran.r-project.org/web/packages/mgcv/mgcv.pdf (accessed on 16 June 2022).
  94. Cribari-Neto, F.; Zeileis, A. Beta Regression in R. J. Stat. Softw. 2010, 34, 1–24. [Google Scholar] [CrossRef]
  95. Marra, G.; Wood, S.N. Practical variable selection for generalized additive models. Comput. Stat. Data Anal. 2011, 55, 2372–2387. [Google Scholar] [CrossRef]
  96. Durbin, J.; Watson, G.S. Testing for serial correlation in least squares regression, 1. Biometrika 1950, 37, 409–428. [Google Scholar]
  97. Landis, J.R.; Koch, G.G. The measurement of observer agreement for categorical data. Biometrics 1977, 33, 159–174. [Google Scholar] [CrossRef]
  98. Huylenbroeck, L.; Laslier, M.; Dufour, S.; Georges, B.; Lejeune, P.; Michez, A. Using remote sensing to characterize riparian vegetation: A review of available tools and perspectives for managers. J. Environ. Manag. 2020, 267, 110652. [Google Scholar] [CrossRef]
  99. Mu, S.; Li, B.; Yao, J.; Yang, G.; Wan, R.; Xu, X. Monitoring the spatio-temporal dynamics of the wetland vegetation in Poyang Lake by Landsat and MODIS observations. Sci. Total Environ. 2020, 725, 138096. [Google Scholar] [CrossRef]
  100. Borro, M.; Morandeira, N.; Salvia, M.; Minotti, P.; Perna, P.; Kandus, P. Mapping shallow lakes in a large South American floodplain: A frequency approach on multitemporal Landsat TM/ETM data. J. Hydrol. 2014, 512, 39–52. [Google Scholar] [CrossRef]
  101. Peng, H.; Xia, H.; Shi, Q.; Chen, H.; Chu, N.; Liang, J.; Gao, Z. Monitoring spatial and temporal dynamics of wetland vegetation and their response to hydrological conditions in a large seasonal lake with time series Landsat data. Ecol. Indic. 2022, 142, 109283. [Google Scholar] [CrossRef]
  102. Murray, N.J.; Keith, D.A.; Simpson, D.; Wilshire, J.H.; Lucas, R.M. Remap: An online remote sensing application for land cover classification and monitoring. Methods Ecol. Evol. 2018, 9, 2019–2027. [Google Scholar] [CrossRef]
  103. Hamilton, S.K.; Kellndorfer, J.; Lehner, B.; Tobler, M. Remote sensing of floodplain geomorphology as a surrogate for biodiversity in a tropical river system (Madre de Dios, Peru). Geomorphology 2007, 89, 23–38. [Google Scholar] [CrossRef]
  104. Ailstock, M.S.; Norman, C.M.; Bushmann, P.J. Common reedbed Phragmites australis: Control and effects upon biodiversity in freshwater nontidal wetlands. Restor. Ecol. 2001, 9, 49–59. [Google Scholar] [CrossRef]
  105. Clevering, O.A. Between-and within-population differences in Phragmites australis. Oecologia 1999, 121, 447–457. [Google Scholar] [CrossRef] [PubMed]
  106. Kettenring, K.M.; Mock, K.E.; Zaman, B.; McKee, M. Life on the edge: Reproductive mode and rate of invasive Phragmites australis patch expansion. Biol. Invasions 2016, 18, 2475–2495. [Google Scholar] [CrossRef]
  107. Saintilan, N.; Asbridge, E.; Lucas, R.; Rogers, K.; Wen, L.; Powell, M.; Colloff, M.J.; Rodriguez, J.F.; Saco, P.M.; Sandi, S.; et al. Australian forested wetlands under climate change: Collapse or proliferation? Mar. Freshw. Res. 2021, 73, 1255–1262. [Google Scholar] [CrossRef]
  108. Yuan, S.; Yang, Z.; Liu, X.; Wang, H. Water level requirements of a Carex hygrophyte in Yangtze floodplain lakes. Ecol. Eng. 2019, 129, 29–37. [Google Scholar] [CrossRef]
  109. Eller, F.; Lambertini, C.; Nguyen, L.X.; Brix, H. Increased invasive potential of non-native Phragmites australis: Elevated CO2 and temperature alleviate salinity effects on photosynthesis and growth. Glob. Chang. Biol. 2014, 20, 531–543. [Google Scholar] [CrossRef]
  110. Burger, J.; Niles, L.; Clark, K.E. Importance of beach, mudflat and marsh habitats to migrant shorebirds on Delaware Bay. Biol. Conserv. 1997, 79, 283–292. [Google Scholar] [CrossRef]
  111. Jia, Y.; Jiao, S.; Zhang, Y.; Zhou, Y.; Lei, G.; Liu, G. Diet shift and its impact on foraging behavior of Siberian crane (Grus leucogeranus) in Poyang Lake. PLoS ONE 2013, 8, e65843. [Google Scholar] [CrossRef] [PubMed]
Remotesensing 15 02614 g001 550
Figure 1. Landsat images show the extent of East Dongting Lake in wet ((A), 3 August 2022) and dry ((B), 29 April 2022) seasons. The stratified random samples of four major habitats were drawn from the overlaps of two habitat maps (1993 and 2010). (C) Map showing the location of Dongting Lake in China and its spatial relation with the Yangtze and the Three Gorges Dam (TGD). (D) Map showing the study area (shaded East Dongting Lake) situated in the lower section of Dongting Lake, and the thick blue line is the Yangtze River.
Figure 1. Landsat images show the extent of East Dongting Lake in wet ((A), 3 August 2022) and dry ((B), 29 April 2022) seasons. The stratified random samples of four major habitats were drawn from the overlaps of two habitat maps (1993 and 2010). (C) Map showing the location of Dongting Lake in China and its spatial relation with the Yangtze and the Three Gorges Dam (TGD). (D) Map showing the study area (shaded East Dongting Lake) situated in the lower section of Dongting Lake, and the thick blue line is the Yangtze River.
Remotesensing 15 02614 g001
Remotesensing 15 02614 g002 550
Figure 2. Flowchart illustrating wetland habitat mapping using Landsat image time series. VI = vegetation index, VIF = variable inflation factor; and RF = random forest.
Figure 2. Flowchart illustrating wetland habitat mapping using Landsat image time series. VI = vegetation index, VIF = variable inflation factor; and RF = random forest.
Remotesensing 15 02614 g002
Remotesensing 15 02614 g003 550
Figure 3. Testing overall accuracy (mean = 0.98) and kappa (mean = 0.97) of the 34 wetland habitat classification models.
Figure 3. Testing overall accuracy (mean = 0.98) and kappa (mean = 0.97) of the 34 wetland habitat classification models.
Remotesensing 15 02614 g003
Remotesensing 15 02614 g004 550
Figure 4. Producer’s accuracy (PA) and user’s accuracy (UA) of the four classes for model training (left) and testing (right). Columns are the mean of 34 RF models, and bars are the standard deviation.
Figure 4. Producer’s accuracy (PA) and user’s accuracy (UA) of the four classes for model training (left) and testing (right). Columns are the mean of 34 RF models, and bars are the standard deviation.
Remotesensing 15 02614 g004
Remotesensing 15 02614 g005 550
Figure 5. Maps of habitat transition showing that most of the habitat (nearly 70%) remained the same during the study period of 1988–2022. Map (A) shows the locations of reed and meadow expansion, as well as the places of erosion (i.e., becoming water). Map (B) shows that a large part of the original mudflat was lost, and the places of lakebed build-up (i.e., water replaced by other habitats).
Figure 5. Maps of habitat transition showing that most of the habitat (nearly 70%) remained the same during the study period of 1988–2022. Map (A) shows the locations of reed and meadow expansion, as well as the places of erosion (i.e., becoming water). Map (B) shows that a large part of the original mudflat was lost, and the places of lakebed build-up (i.e., water replaced by other habitats).
Remotesensing 15 02614 g005
Remotesensing 15 02614 g006 550
Figure 6. Modelled response curves to year from the duration model (upper panel) and date model (lower panel). Black lines are the estimate, and the grey shaded areas are 95% confident intervals. The Y-axis is the modelled partial response of habitat area to year. Note the zero effect on meadow (the modelled partial effect is a flat line).
Figure 6. Modelled response curves to year from the duration model (upper panel) and date model (lower panel). Black lines are the estimate, and the grey shaded areas are 95% confident intervals. The Y-axis is the modelled partial response of habitat area to year. Note the zero effect on meadow (the modelled partial effect is a flat line).
Remotesensing 15 02614 g006
Remotesensing 15 02614 g007 550
Figure 7. Modelled response curves to duration of dry season (upper panel) and date of water level withdrawal (lower panel). Black lines are the estimate and the grey shaded areas are 95% confident intervals. The Y-axis is the modelled partial response of habitat area to date of water level withdrawal. Note that the modelled partial effect of both DOD and DOYD on reed is a flat line.
Figure 7. Modelled response curves to duration of dry season (upper panel) and date of water level withdrawal (lower panel). Black lines are the estimate and the grey shaded areas are 95% confident intervals. The Y-axis is the modelled partial response of habitat area to date of water level withdrawal. Note that the modelled partial effect of both DOD and DOYD on reed is a flat line.
Remotesensing 15 02614 g007
Remotesensing 15 02614 g008 550
Figure 8. Modelled response curves to dry season rainfall from the duration model (upper panel) and date model (lower panel). Black lines are the estimate and the grey shaded areas are 95% confident intervals. Note that the effect on meadow is small and insignificant (Table 4).
Figure 8. Modelled response curves to dry season rainfall from the duration model (upper panel) and date model (lower panel). Black lines are the estimate and the grey shaded areas are 95% confident intervals. Note that the effect on meadow is small and insignificant (Table 4).
Remotesensing 15 02614 g008
Table
Table 1. Spectral indices for random forest classifier and their respective characteristics.
Table 1. Spectral indices for random forest classifier and their respective characteristics.
NameRelevanceFormulaReference
NDVI (normalized difference vegetation index)NDVI measures photosynthetically active biomass in plants. It is the most highly used index to monitor plant development dynamics. N I R − R e d N I R + R E D [70]
NDWI (normalized difference water index)NDWI detects open water features. G r e e n − N I R G r e e n + N I R [71]
PSRI (plant senescence reflectance index)PSRI is sensitive to the ratio of carotenoids to chlorophyll in plants, indicating the process of vegetation senescence. R e d − G r e e n N I R [72]
SAVI (soil adjusted vegetation index)SAVI is used to mitigate the impact of soil brightness to correct NDVI in areas where vegetative cover is low. N I R − R e d N I R + R E D + 0.5 × 1.5 [73]
GNDVI (green normalized difference vegetation index) GNDVI is a modification of NDVI to detect wilted or aging plants and to measure nitrogen content in leaves. It is suitable to monitor vegetation with dense canopies or at different maturity stages. N I R − G r e e n N I R + G r e e n [74]
BSI (bare soil index)BSI is used to differentiate bare soil and other land cover types. Due to the high contrast between bare soil and vegetation, BSI provides a continuum ranging from high vegetation cover to exposed soil. R e d + S W I R 2 − N I R + B l u e R e d + S W I R 2 + ( N I R + B l u e [75]
NDPI (normalized difference phenology index)NDPI improves the spring greening-up phenology monitoring capacity in snow-contaminated or low vegetation cover areas. N I R − 0.74 × R e d + 0.26 × S W I R 1 N I R + 0.74 × R e d + 0.26 × S W I R 1 [76]
Table
Table 2. The testing confusion matrix of the worst model for mudflat (year 1989), with producer’s accuracy (PA) and user’s accuracy (UA).
Table 2. The testing confusion matrix of the worst model for mudflat (year 1989), with producer’s accuracy (PA) and user’s accuracy (UA).
ClassMeadowMudflatReedWaterSumPA (%)
Meadow14780015594.8
Mudflat855006387.3
Reed002990299100
Water214020321992.7
Sum15777299203736
UA (%)93.671.4100100 95.7 (OA)
Table
Table 3. Summary of habitat transition (percentage of the total lake area) in East Dongting Lake during the period of 1988–2021. The shaded part is the transition matrix with year 1988 in columns and year 2021 in rows.
Table 3. Summary of habitat transition (percentage of the total lake area) in East Dongting Lake during the period of 1988–2021. The shaded part is the transition matrix with year 1988 in columns and year 2021 in rows.
MeadowMudflatReedWaterPersistenceGainLossNCSCTCP/TG/PL/P
Meadow29.648.170.980.9629.6410.115.764.3511.5315.880.840.340.19
Mudflat1.5612.300.244.3112.306.1216.27−10.1512.2322.380.430.501.32
Reed2.811.397.100.237.104.431.313.122.625.740.840.620.18
Water1.396.700.0922.1222.128.185.502.6811.0013.680.800.370.25
NC = net change, SC = swap change, TC = total change, P/T = persistence/total area in 1988, G/P = gain/persistence, L/P = loss/persistence. Lake area = 53,632 ha.
Table
Table 4. Summary of the two fitted GAMM for wetland habitat variation in East Dongting Lake during 1988–2021.
Table 4. Summary of the two fitted GAMM for wetland habitat variation in East Dongting Lake during 1988–2021.
Duration Model Date Model
Parametric TermEst.SEz-Valuep * Est.SEz-Valuep
Intercept−1.2640.453−2.7930.005Intercept−2.0050.491−4.0880.000
DOD−0.0010.001−2.1850.029DOYD0.0020.0012.4640.014
Smooth termedfRef.dfChi.sqp edfRef.dfChi.sqp
s(Year): Meadow #0.00090.0000.690s(Year): Meadow0.00090.0000.730
s(Year): Mudflat2.140958.7780.000s(Year): Mudflat2.110955.9540.000
s(Year): Reed0.84496.6330.010s(Year): Reed0.85797.3680.008
s(Hab)1.99924895.3120.000s(Hab)1.99924859.3470.000
s(DOD): Meadow1.759911.3360.000s(DOYD): Meadow0.915911.8510.000
s(DOD): Mudflat1.34293.2630.068s(DOYD): Mudflat1.77497.5120.010
s(DOD): Reed0.00090.0000.465s(DOYD): Reed0.00090.0000.302
s(Rain): Meadow1.16192.5140.108s(Rain): Meadow0.59290.8090.239
s(Rain): Mudflat1.795917.7020.000s(Rain): Mudflat1.743916.9020.000
s(Rain): Reed2.184915.6080.001s(Rain): Reed1.993910.8010.004
R2 (adj) 0.984 R2 (adj) 0.984
edf = effective degree of freedom; Ref.df = reference degree of freedom; Chi.sq = Chi square. DOD = duration of dry season (days); DOYD = date of water level withdrawal (day of year); and Rain = total precipitation in winter. # s = smoothing term; * Bold indicated significant, and p = 0.000 means < 0.00001.
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Jing, L.; Zeng, Q.; He, K.; Liu, P.; Fan, R.; Lu, W.; Lei, G.; Lu, C.; Wen, L. Vegetation Dynamic in a Large Floodplain Wetland: The Effects of Hydroclimatic Regime. Remote Sens. 2023, 15, 2614. https://doi.org/10.3390/rs15102614

AMA Style

Jing L, Zeng Q, He K, Liu P, Fan R, Lu W, Lei G, Lu C, Wen L. Vegetation Dynamic in a Large Floodplain Wetland: The Effects of Hydroclimatic Regime. Remote Sensing. 2023; 15(10):2614. https://doi.org/10.3390/rs15102614

Chicago/Turabian Style

Jing, Lei, Qing Zeng, Ke He, Peizhong Liu, Rong Fan, Weizhi Lu, Guangchun Lei, Cai Lu, and Li Wen. 2023. "Vegetation Dynamic in a Large Floodplain Wetland: The Effects of Hydroclimatic Regime" Remote Sensing 15, no. 10: 2614. https://doi.org/10.3390/rs15102614

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop