Effects of the Water-Sediment Regulation Scheme on the Expansion of Spartina alterniflora at the Yellow River Estuary, China

In recent decades, the invasion of saltmarsh plant Spartina alterniflora (S. alterniflora) over a large part of coastal wetlands in China, including the Yellow River Estuary (YRE) as a regional economic hub and global ecosystem services hotspot, has caused increasing concern because of its serious threats to native ecosystems. During the same period, local authorities have implemented a Water-Sediment Regulation Scheme (WSRS) in the Yellow River for flood mitigation and delta restoration purposes. The altered hydrological regime has resulted in unintended changes to estuarine ecosystem. However, the direct consequence of the WSRS on the expansion of S. alterniflora remains unclear. In this study, quantitative relationship between the inter- and intra-annual expansion patterns of S. alterniflora represented by relevant landscape metrics and indicators that quantify the concurrent variations of river and sediment discharges as the proxy of the WSRS impacts were analysed over the period of Year 2011–2018, and the analyses were performed on the YRE as a whole and on five different zones subdivided based on the invasion sequence. The results showed that there was no significant difference in the inter-annual area variation of S. alterniflora between the years with and without WSRS. Compared with the years without WSRS (2016–2017), the intra-annual (monthly) increment of the various landscape metrics [i.e., NP (number of patches), CA (class area), LPI (largest patch index) and AI (aggregation index] were found to be significantly higher in the initial stage of peak growing season (June-July) than in the mid- and late stages (July-September) in the years with WSRS (2011–2015, 2018) in the subregion located close to the south bank of YRE as the most prominent impact zone. In addition, F (mean flow), Ff (number of high flow pulses), Tf (Julian date of maximum flow) and D (duration of WSRS) were identified as the explanatory variables for the intra-annual vegetation landscape pattern changes, and their relative contributions to resultant changes were also assessed. Our results broaden the understanding of estuarine hydrological disturbance as a potential driver regulating the saltmarsh vegetation, and also have implications for S. alterniflora invasion control at estuaries under changing environment.


INTRODUCTION
Spartina alterniflora (S. alterniflora), an exotic species native to the East Coast of North America, was introduced to coastal areas of China in 1979 for promoting sediment deposition and erosion control (Li et al., 2009). Ever since then, S. alterniflora has gradually expanded to large areas of estuarine and coastal wetlands due to its high salt tolerance and strong reproductive capacity Qiu et al., 2020), and posed serious threats to local native ecosystems (Schwarz et al., 2016). The growth and expansion of S. alterniflora are dependent on a number of key physical factors in estuarine saltmarshes, such as surface elevation (Carey et al., 2017), inundation  and salinity (Elsey-Quirk et al., 2009). Subject to the combined action of terrestrial and marine forcings such as the river and sediment discharges, waves and tides, these physical factors per se are constantly evolving at the estuaries (Snedden et al., 2015;Gao et al., 2018;Han et al., 2018;Poormahdi et al., 2018). Furthermore, it is well established that frequent interventions of human activities such as upstream dam regulation can result in dramatic alterations of river and sediment discharges Gao et al., 2018). Therefore, it is necessary to elucidate the roles of the estuarine hydrological alteration to better understand invasion process of S. alterniflora at the estuaries that are crucial for wetland ecosystem management.
A number of studies have examined the interplay between the variations of river and sediment discharges and marsh vegetation on the intra-annual (seasonal) or inter-annual timescale (e.g., Kearney et al., 2011;Carle et al., 2015;Olliver and Edmonds, 2017;Hu et al., 2019). Amongst them, Kearney et al. (2011) analyzed the vegetation cover and marsh area at the freshwater diversion sites at the Louisiana's coastal wetlands from 1984 to 2005 using Landsat data, and showed that freshwater diversions have not increased vegetation and marsh coverage. However, the input of additional nutrients as a result of the freshwater diversion reduced marsh below-ground biomass and increased its vulnerability to storm damage. Carle et al. (2015) assessed the impact of the historical 2011 Mississippi River flood on the Wax Lake Delta and its vegetation community change using remote sensing and LIDAR data, and found a net delta growth and the conversion between lower-elevation and higher-elevation species in both directions across the delta due to the heterogeneous elevation change. Olliver and Edmonds (2017) confirmed the control of surface elevation and inundation period on the extent of vegetation communities through their analyses of the interand intra-annual changes in vegetation cover of Wax Lake Delta using remote sensing and field data.
Based on the monitoring of mudflat elevation and marsh vegetation area at the Yangtze Estuary from 2013 to 2017, Hu et al. (2019) reported that the year with higher river and sediment discharges resulted in significantly higher rates of mudflat accretion and vegetation expansion, which suggested that a short-term increase in river and sediment discharges can stimulate the vegetation expansion due to the positive feedback between vegetation establishment and sedimentary processes. In addition, a number of studies have explored the influences of flow-sediment regime shift on landscape pattern changes of riparian plant species, such as cover of vegetation communities (Caruso et al., 2013), spatial distribution range (Mortenson and Weisberg, 2010) and landscape fragmentation degree (Shen and Ma 2020). By far, the majority of the existing studies focused on assessing the inter-annual change of marsh vegetation coverage and distribution in response to the variable river and sediment discharges using remote sensing data. However, direct relationship between the flow-sediment regime variables and the resultant expansion patterns of saltmarsh plants as represented by the relevant landscape metrics on both inter-and intra-annual timescales remain elusive.
The Yellow River Estuary (YRE) located in the northeast of Shandong Province, China, is a regional economic hub where a major oil field of China (the Shengli Oil Field) is located as well as a global ecosystem services hotspot hosting a key stopover site in the middle of the East Asian-Australasian flyway, which was designated as a Ramsar site in 2013 (Luo et al., 2018). S. alterniflora was purposely introduced to YRE for dike protection and erosion control in 1990, and has since spread to the tidal marshes along the elevation gradient (Ren et al., 2014). Meanwhile, the Yellow River Conservancy Commission (YRCC) initiated the Water-Sediment Regulation Scheme (WSRS) in the Yellow River in 2002 to create artificial flood pulses through dam regulation during a one-month period of the flood season every year to mitigate the siltation both in the Xiaolangdi Reservoir and the lower reaches, and restore ecohydrological conditions downstream . As the WSRS resulted in a significant increase in river discharge and sediment load during the operation period, it posed unexpected disturbances on estuarine and coastal ecosystems . Notably, Ren et al. (2014) found that the expansion of S. alterniflora at the Yellow River Delta was closely related to the fluctuating discharge and sediment load of the Yellow River, which was partly caused by WSRS. Wang et al. (2017) and Bi et al. (2019) demonstrated that coarser suspended sediments induced by WSRS tended to deposit locally at the estuary and stimulate rapid elevation accretion. Ren et al. (2019) studied the inter-annual expansion process of S. alterniflora at the Yellow River Delta, and reported a significant correlation between the expansion area and the improvement in water quality of the Yellow River. Nevertheless, until now, the effects of WSRS on the expansion of S. alterniflora at the YRE are still unclear.
In this study, we explore the effects of WSRS on the spatiotemporal expansion dynamics and distribution patterns of S. alterniflora at the YRE, through assessing the inter-and intra-annual changes of relevant landscape metrics between the years with WSRS (2011)(2012)(2013)(2014)(2015)2018) and the years without WSRS (2016-2017). As a case study, it also sheds some light on the broad question of the effects of flow-sediment regime shift caused by variable river discharge and sediment load on the colonization patterns of estuarine saltmarsh plants. The improved understanding on the invasion process of S. alterniflora can also inform invasion control under changing environment.

Study Area
The study area is located at the YRE wetland in Dongying City, China (119°09′E-119°19′E; 37°43′N-37°51′N, Figure 1), which has a typical semi-humid continental monsoon climate with mean annual temperature of 12.1°C and precipitation of 551.6 mm (the rainfall is mostly distributed between July and August) (Song et al., 2018). In 1996, the Yellow River shifted to the Q8 course before emptying into the Bohai Sea after an artificial avulsion, which resulted in the current YRE . The Yellow River Delta purposely introduced S. alterniflora near the No. 5 pile to the north of the Gudong oil exploration site in 1990, and S. alterniflora has since spread across the low marshes and spread out to the high marshes gradually (Ren et al., 2014). More recently, some field surveys and remote sensing image analyses have observed that there was a rapid expansion of S. alterniflora communities at the current YRE since 2011 (Ren et al., 2019).
In 2002, the YRCC began to implement the Water-Sediment Regulation Scheme (WSRS) in the Yellow River to mitigate the siltation both in the lower river reaches and the Xiaolangdi Reservoir by flood pulses created by dam regulation. The WSRS has been operated every year since 2002, except 2016 and 2017, which resulted in significant increases in river and sediment discharges during a short period (∼20 days) from mid-June to early July Wang et al., 2017). Notably, the WSRS has posed substantial changes on estuarine and coastal ecosystems, such as shift of sedimentation dynamics and changes in delta morphology (Wang et al., 2010;Bi et al., 2014).
In this study, the entire YRE was subdivided into five subregions from a to e based on the sequence of the land formation and invasion (colonization) of the S. alterniflora ( Figure 2). Specifically, sub-region a and sub-region d were created by sediments from the Yellow River during the 2011 flood, where S. alterniflora began to spread on a large scale around the same period. Sub-regions b and c were subsequently formed in 2013, and sub-region b was colonized by S. alterniflora in 2014, while S. alterniflora did not emerge in sub-region c until 2016. Further away from the river mouth, sub-region e was formed in 2007, where S. alterniflora began to spread on a large scale in 2013.

Remote Sensing Data and Image Processing
The WSRS usually takes place in June-July every year since 2002, except 2016 and 2017, and S. alterniflora began to spread rapidly at the YRE since 2011 . Moreover, the peak growing season of S. alterniflora is from June to September, and it almost stops growing at the end of September at the YRE. According to previous studies, temporally adjacent dataset can yield a time series to estimate the effect of external disturbances to ecosystems (Walker et al., 2015). Considering its short return cycle (16 days) and moderate resolution, the Landsat satellite images were suitable for observing seasonal dynamics of S. Frontiers in Environmental Science | www.frontiersin.org June 2021 | Volume 9 | Article 642442 alterniflora over the study period. Therefore, 32 good-quality (i.e., cloud-free, low tide level) Landsat 7/8 remote sensing images of YRE with temporally adjacent mean 30 days (1 image per month) from June to September during 2011 to 2018 were acquired to examine the spatiotemporal patterns of S. alterniflora invasion in our study. These images were downloaded from the United States Geological Survey (USGS). Table 1 lists the features of the images used in this study. Following previous studies Mao et al., 2019), we chose maximum likelihood classifier to interpret S. alterniflora at the YRE based on its spectrum characteristics. Firstly, all images were reprojected into universe transverse Mercator world geodetic system-84 (UTM/WGS 84) coordinate system. Then, the Fast Line-of-sight Atmospheric Analysis of Hypercubes (FLAASH) model was applied to conduct atmospheric corrections in the ENVI 5.3 software (Harris Geospatial Solutions, Inc., United States). In order to improve classification accuracy, the panchromatic band with a spatial resolution of 15 m and the multispectral bands with a spatial resolution of 30 m for all Landsat 7/8 images were fused using Gram-Schmidt Pansharpening fusion method to exhibit higher sharpness and spectral quality. Additionally, the spectral enhancement methods of the tasseled cap transform and false color composition were used to initialize the data for efficient interpretation (Aiazzi et al., 2006). To further assess the spatial distribution of S. alterniflora, S. alterniflora and other land covers (i.e., water, mudflats, other vegetation patches such as Suaeda salsa, Tamarix chinensis and Phragmites australis) were identified as different classes and selected as training samples, and a supervised classification was carried out using the Support Vector Machine (SVM) (Mountrakis et al., 2011). Based on the field verification data and historical images, some misclassifications of unvegetated and vegetated habitats were further corrected.

Quantifying the Invasion of Spartina alterniflora
In order to quantitatively assess the landscape pattern change of S. alterniflora at the YRE, we reviewed relevant studies (Kelly et al., 2011;Almeida et al., 2016;Liu et al., 2017) and chose the landscape metrics with relevance to our study and easiness to interpret. These metrics mainly focus on composition and spatial configuration of landscape. Specifically, composition refers to the diversity and abundance of patch types within the landscape, and the composition metrics are the most basic and important measures of landscape patterns. In this study, we calculated CA (class area), NP (number of patches) and LPI (largest patch index) (McGarigal 2014). CA and NP are the metrics commonly used to measure landscape composition. LPI quantifies the percentage of total landscape area comprised by the largest patch, which is a simple measure of dominance. Spatial configuration is related to spatially explicit characteristics, which   gi max → gi P i × 100 AI reflects the level of aggregation of spatial pattern and provides a quantitative basis to correlate spatial pattern with processes that are typically class specific g i number of like adjacencies (joins) between pixels of patch type (class) i based on the single-count method max → g i Maximum number of like adjacencies (joins) between pixels of patch type (class) i (see below) based on the single-count method P i proportion of landscape comprised of patch type (class) i contains more complicated information about the landscape structural and functional features, and is closely related to landscape fragmentation degree (Leitão et al., 2012;Almeida et al., 2016). Here, AI (aggregation index) was chosen to assess landscape configuration, which is an effective index to measure the aggregation degree of spatial patterns (McGarigal 2014; Almeida et al., 2016). Finally, the monthly increment of the various landscape metrics (i.e., the difference of the landscape metrics between two adjacent months) was used to quantify the intra-annual difference of S. alterniflora invasion at the YRE. All landscape metrics are listed in Table 2, and they were calculated using Fragstats 4.2 software (McGarigal et al., 2002).

Hydrological Data Acquisition and Quantification With Flow-Sediment Regime Indicators
River and sediment discharges to the YRE were acquired from the records at the Lijin hydrological gauge station, which is the station closest to the YRE and located approximately 100 km upstream. Both daily runoff and sediment load data recorded at Lijin station from 2011 to 2018 have been made available by the YRCC of China.
The Indicators of Hydrologic Alteration (IHA) framework proposed by Richter et al. (1996) is one of the most widely used methods to evaluate the variations of flow regime and their impact on ecosystems. Wohl et al. (2015) broadened the flow regime concept into a more inclusive paradigm to also incorporate sediment regime. Xu and Li (2019) used the upgraded flow-sediment regime indicators of IHA to study the impact of flow-sediment regime changes on floodplain ecosystem. In this study, we adopted the IHA framework and focused on the indicators which are particularly relevant to the hydrological disturbances caused by the WSRS as well as the key life history stages of S. alterniflora (see Table 3 for a complete list of the chosen indicators).
Specifically, we chose the duration of annual WSRS as the time period for all hydrological indicators. As increases in runoff and sediment load over a short period had been demonstrated to affect the coastal ecosystem and saltmarsh plant growth (Hu et al., 2019), mean flow and sediment load during the WSRS period each year were selected. Moreover, the magnitude of flow and sediment load caused by the WSRS are far beyond the natural flow-sediment regime, and the hydrological extremes can most interfere with the phenology and growth of saltmarsh plants (Hu et al., 2019;Jones et al., 2019). As such, we chose indicators represent the magnitude and occurrence time of peak flow and sediment load. The 7 days/3 days/1 day maximum flow and sediment load were selected to best capture the peak value, and the Julian date of maximum flow and sediment load during the WSRS period each year represented the time when the peak flow and sediment load occurred. In addition, Kettenring et al. (2015) and Balke et al. (2014) found that a certain degree of frequency variation of sediment accretion/erosion and flooding can create windows of opportunity for invasion of exotic saltmarsh plants. Therefore, the number of high flow and sediment load pulses during the WSRS period each year were also selected. Besides, as the degree of fluctuation in flow and sediment regime have proven effects on establishment and survival of saltmarsh plants (Seabloom et al., 2001;Cao et al., 2018), we chose the number of flow and sediment reversals to assess the potential effects of rate of change of flow-sediment regime.
In summary, the above-chosen indicators include magnitude, timing, duration, frequency and rate of change of flow-sediment regime shift, which provides a holistic quantification of the variations of flow and sediment load that may impact the saltmarsh plants (Poff et al., 1997;Elsey-Quirk et al., 2019). These indicators were calculated using IHA software (Smythe Scientific Software, Boulder, CO, United States).

Statistical Analysis
A student's t test (Winter 2013) was used to analyze the significance of differences in inter-annual area variation of S. alterniflora invasion between the years with and without WSRS. A repeated-measures ANOVA (Keselman et al., 2001) was used to examine the significance of trend differences in intra-annual variation of S. alterniflora landscape patterns between the years with and without WSRS. All these statistical analyses were performed in SPSS 22 software (SPSS Inc., Chicago, IL, United States).
Furthermore, we employed generalized linear models (GLMs) using Gaussian errors to estimate how flow-sediment regime variations during the WSRS affect the landscape patterns of S. alterniflora. Spearman's rank correlation coefficients (r s ) with correlation values r s < |0.75| and variance inflation factors (VIF) < 4 (R package ''car'') were adopted as the criteria to reduce multicollinearity among the explanatory variables (Dormann et al., 2013;Fox and Weisberg 2018). The models were simplified using stepwise analysis (forward and backward) with the R package ''MASS'' (Crawley 2012). In order to assess the goodness-of-fit of the GLMs and obtain the optimal regression models, the adjusted determination coefficient (R adj 2 , a higher value indicates better fit), significance (p < 0.05 or 0.01) and Akaike Information Criterion (AIC, a smaller value indicates better fit) were used to compare multiple models (Burnham and Anderson 2004). Additionally, to further calculate the relative importance of all explanatory variables contained in the optimal GLMs and measure the individual contribution of each explanatory variable to the explained variables, the Lindeman, Merenda and Gold (LMG) metric was used with the R package ''relaimpo'' in the optimal models (Grömping 2006). We ranked relative importance values of each explanatory variable according to the LMG values to quantify their contribution to the intra-annual landscape pattern changes of S. alterniflora. All these modeling analyses were performed in R version 3.5.2, and the R Package "car" ''MASS'' and ''relaimpo'' were used to run the GLMs (R Core Team, 2017). In addition, the Durbin-Watson statistic (DW value: 0-4) was introduced to check the autocorrelation between residuals of optimal models (Penda et al., 2012). A value near 2 indicates non-autocorrelation between residuals, whereas values below (above) 2 indicate positive (negative) autocorrelation between residuals. Moreover, the normal probability plots of regression standardized residuals were introduced to check the normality between residuals of optimal models (Wisniak and Polishuk 1999). If the resulting plot is approximately linear, we can assume that the residuals are normally distributed. All these post-robustness tests were performed in SPSS 22 (SPSS Inc., Chicago, IL, United States). Figure 3 shows the evolution of S. alterniflora cover area and spatial distribution throughout the entire estuary and sub-regions in September each year from 2011 to 2018. In general, the S. alterniflora continued to invade from 2011 to 2018 with the total area increasing from 205.04 to 3,394.29 ha at the YRE, and the invasion of S. alterniflora followed sequence in different subregions. Specifically, the S. alterniflora expanded rapidly in subregion a and increased from 146.4 ha in 2011 to 910.2 ha in 2018. Sub-regions b and c, which were newly formed in 2013, were the youngest regions at the YRE. Due to the suitable environmental conditions, S. alterniflora quickly invaded and occupied 689.5 ha in sub-region b and 143.5 ha in sub-region c in 2018. Moreover, the S. alterniflora showed continuous expansion in sub-region d throughout the study period, and the area of S. alterniflora expanded from 58.6 ha to 527.4 ha. Sub-region e hosted the largest S. alterniflora communities in the south bank of YRE, and the area of S. alterniflora covered 1,123.67 ha in 2018. In addition, we also analyzed the significance of differences in inter-annual variation of S. alterniflora cover area in September between the years with WSRS (2011-2015, 2018) and without WSRS (2016-2017) for the entire YRE and all sub-regions. The results consistently showed no significant difference throughout the study area and study period.

Spatiotemporal Dynamics of Spartina alterniflora Invasion
We further analyzed the trend differences in the intra-annual area variation of S. alterniflora invasion between the years with and without WSRS (Figure 4). The results showed that there was no significant difference for the entire YRE (p 0.248). However, when breaking up into the sub-regions, monthly increment of area in subregion d in the years with WSRS was significantly different from years without WSRS (p 0.037). In addition, in the years with WSRS, the monthly area increment of S. alterniflora was significantly higher in the initial stage of the peak growing season (June-July) than in the mid-and late stages (July-September), compared to the years without WSRS. In contrast to sub-region d, the intra-annual variation of S. alterniflora area in sub-region b (p 0.114) and sub-region e (p 0.318) showed no significant trend differences between the years with and without WSRS. It is noteworthy that S. alterniflora hardly spread in sub-region a since 2016, and S. alterniflora did not appear in sub-region c until 2016 (Figure 3). As such, there were not enough and accurate data to analyze the differences of S. alterniflora in years with and without WSRS, and thus we excluded these two subregions as well as sub-region b and e, and focus on sub-region d in our subsequent analysis.
We extended the above analysis to other landscape pattern metrics in sub-region d (Figure 5), and significant differences were also found in the monthly increment trends of NP (number of patches) (p 0.039), LPI (largest patch index, p 0.035) and AI (aggregation index, p 0.027) between the years with and without WSRS. This suggests that the intra-annual growth and colonization behaviors of S. alterniflora in sub-region d are very likely to be affected by the WSRS. Table 4 shows all 15 flow-sediment regime variables during the WSRS at the YRE. F (mean flow) was chosen as the basic variable for ease of interpretability and applicability in calculating the Spearman's rank correlation coefficients (rs) and variance inflation factors (VIF) for other variables (Richter et al., 1996). Based on the correlation analysis, the variables F f (rs 0.64), T f (rs 0.27), R S (rs 0.56) and D (rs 0.7) were selected to analyze VIF, and the associated VIF values (1.341, 2.643 and 1.933, respectively) all satisfied the criteria set to reduce multicollinearity (< 4). Finally, F (mean flow), F f (number of high flow pulses), T f (Julian date of maximum flow) and D (duration of WSRS) were retained after checking the multicollinearity among all explanatory variables using  Spearman's rank correlation coefficients and VIF. In order to assess the influence of WSRS on the S. alterniflora expansion, we analyzed the relationship between the identified key explanatory variables and the increment of the landscape metrics (i.e., CA, NP, LPI and AI) during annual peak growing season of S. alterniflora in the years with WSRS (2011WSRS ( -2015WSRS ( , 2018 in subregion d ( Table 5). In addition, the robustness of optimal GLMs were further tested to ensure the non-accidental character of the results, the DW value ( Table 5) and normal probability plots of regression standardized residuals ( Figure 6) showed that the obtained optimal models met the robustness requirements.

Spartina alterniflora Invasion in Relation to Variation of Flow-Sediment Regime Caused by Water-Sediment Regulation Scheme
In addition, we further analyzed the relative contribution of the identified explanatory variables in the optimal GLMs to landscape pattern changes of S. alterniflora (Figure 7; Table 5). Results showed that optimal models explained approximately 58% of CA (class area) change (R 2 0.583), 70% of NP (number of patches) change (R 2 0.696), 60% of LPI (largest patch index) change (R 2 0.604) and 65% of AI (aggregation index) change (R 2 0.648), suggesting that the identified flow-sediment regime variables were able to explain a large number of changes in intra-annual landscape pattern of S. alterniflora. Specifically, mean flow (F, p 0.032) was the most important explanatory variable for the change of class area (CA), and Julian date of maximum flow (T f , p 0.045) also had significant effect on the change of class area. Similarly, change of number of patches (NP) was mainly affected by mean flow (F, p 0.012) and Julian date of maximum flow (T f , p 0.019). The change of largest patch index (LPI) was mainly affected by number of high flow pulses (F f , p 0.033), followed by Julian date of maximum flow (T f , p 0.037). As for the aggregation index (AI), the mean flow (F, p 0.029), Julian date of maximum flow (T f , p 0.023) and duration of WSRS (D, p 0.047) were found to play significant role (Figure 7). Figure 8 further shows the correlation of key flow-sediment regime variables in the optimal GLMs with landscape pattern changes of S. alterniflora. Overall, when the runoff increased, the class area (CA) and aggregation index (AI) significantly increased. However, the number of patches (NP) was significantly lower under high runoff. In addition, when the occurrence time of high pulse was delayed, the number of patches (NP) decreased, whereas the class area (CA), largest patch index (LPI) and aggregation index (AI) of S. alterniflora increased. Furthermore, the aggregation index (AI) decreased with increasing duration of high pulses, and the largest patch index (LPI) decreased with increasing high pulse frequency.

DISCUSSION
Many studies have analyzed the effects of marine forcings such as tides and waves on saltmarsh plants (Silinski et al., 2016;Carus et al., 2017). Moreover, some studies have shown that frequent hydrodynamic and sedimentation disturbances can restrict the saltmarsh plants growth (Balke et al., 2013;Aguiar et al., 2018). However, a general understanding of whether the landscape patterns of the saltmarsh vegetation are affected by the intraannual changes of river and sediment discharges is lacking. In this study, our results showed that the variations of estuarine river and sediment discharges caused by an annual hydrological regime regulation operation, i.e., the WSRS, significantly altered intraannual expansion patterns of S. alterniflora in part of the estuarine saltmarsh close to the mouth (sub-region d) (Figure 4 and Figure 5). Our study further analyzed the relative importance of identified flow-sediment regime variables to the change of intra-annual distribution patterns of S. alterniflora (Figure 7; Table 5), and demonstrated that the increase of high pulse magnitude and the delay of high pulse occurrence time can be beneficial to S. alterniflora invasion, whereas the increase of high pulse frequency and duration can mitigate S. alterniflora invasion (Figure 8).   River regulation such as WSRS results in biogeomorphic variation of estuarine saltmarshes in surface elevation, inundation and salinity (Han et al., 2018;Yang et al., 2020), and the resultant response of plant species to changing environments is complicated (Carle et al., 2015;Snedden et al., 2015). The artificial controls on water and sediment input may cause regime shifts in morphological evolution of estuaries and deltas, such as progradation of delta lobes and crevasse splays vs. avulsions (Nienhuis et al., 2018;Gao et al., 2019), which could further affect estuarine biogeomorphic environments (Xu et al., 2016). In previous studies, Keogh et al. (2019) and Nienhuis et al. (2018) have reported that the crevasse splays were conducive to generation of new land and variations of biogeomorphic conditions in estuaries, due to the water and sediment channeled along the crevasse splays. In sub-region d of the YRE, a crevasse splay was formed and developed since 2011, and S. alterniflora began to spread along the crevasse splay ever since then (Figure 9). The WSRS usually takes place in the initial stage of the peak growing season (June-July) of S. alterniflora, and the additional freshwater and sediment supply channeled along the crevasse splay from the Yellow River into sub-region d during the WSRS may contribute to the enhanced expansion and landscape pattern changes. Notably, Han et al. (2018) reported that compared with the regions only affected by tides, in the regions remained inundated all the time after annual WSRS including sub-region d, the soil salinity and soil moisture were significantly changed after the WSRS. Xie et al. (2019) further demonstrated from field survey that freshwater irrigation was beneficial to the S. alterniflora growth because of the reduced soil salinity and increased moisture content caused by freshwater input.
On the landscape scale, our results verified that the flowsediment regime changes caused by WSRS can interfere with the intra-annual expansion process of S. alterniflora in sub-region d. Compared with the years without WSRS (2016-2017), the monthly increment of S. alterniflora landscape metrics (i.e., CA, NP, LPI and AI) were significantly higher in the initial stage of the peak growing season (June-July), which was also the WSRS period, than in the mid-and late stages of the peak growing season (July-September) in the years with WSRS (2011WSRS ( -2015WSRS ( , 2018 (Figures 4, 5). In contrast to subregion d, there are no crevasse splays to supply additional freshwater and sediment from the Yellow River to saltmarshes during the WSRS in the other sub-regions, which may partly lead to the insignificant effects of the WSRS on the S. alterniflora expansion in these areas.
As for the specific roles of the flow-sediment regime shifts in S. alterniflora invasion, our results shed some light on the relative importance of the identified flow-sediment regime variables (Figures 7, 8). The distribution pattern of S. alterniflora is related to the gradient of the various biogeomorphic factors (e.g., elevation, inundation, salinity etc.), which are subject to physical disturbances such as flooding (Kettenring et al., 2015). Our results show that when the runoff increased, the grow and expansion of S. alterniflora were enhanced and the degree of fragmentation decreased as indicated by the increasing class area (CA) and aggregation index (AI) and decreasing number of patches (NP). This may be attributed to the increased soil moisture and reduced salinity due to the WSRS, which stimulated the S. alterniflora expansion (Han et al., 2018). Furthermore, delayed disturbance was shown to produce more benefits to S. alterniflora growth and expansion through the analysis with the delay of high pulse occurrence time (Figure 8). Notably, Richter et al. (1996) indicated that the timing of extreme flood conditions provides measure of environmental disturbances, which may be intimately linked to the key life-cycle phases of plants, and the degree of physical disturbances (i.e., hydrodynamic forces and associated sedimentation dynamics) defines the lethal thresholds of seedling establishment (Hu et al., 2015;Cao et al., 2018). The initial stage of the peak growing season is the key period of S. alterniflora seedling establishment at the YRE, and the delay of high pulse occurrence time, which created disturbance-free or lowdisturbance periods for seedling establishment and recruitment of S. alterniflora, can be beneficial to S. alterniflora growth and expansion (Balke et al., 2011). On the contrary, the effects of the increase in high pulse frequency and duration tend to be negative, e.g., the former led to increased fragmentation as manifested by the decreasing largest patch index (LPI), and the latter led to decreasing aggregation index (AI). This intricate interplay is consistent with the previous findings that the survival and growth of S. alterniflora seedlings were closely related to the life-history-dependent stress tolerance and hence the adaptability of the plant species to external disturbances (Balke et al., 2013;Balke et al., 2014;Bruckner et al., 2019).

CONCLUSION
Our study explored the effects of the annual Water-Sediment Regulation Scheme on S. alterniflora expansion at the Yellow River Estuary as a case study of the broad question of the effects of flow-sediment regime shift caused by variable river discharge and sediment load on the colonization patterns of estuarine saltmarsh plants. In line with previous studies (Ren et al., 2019), our analysis based on remote sensing images confirmed that the S. alterniflora had a dramatic expansion at the YRE since 2011. The results showed that compared with the years without WSRS (2016-2017), the intra-annual (monthly) increment of S. alterniflora landscape metrics (i.e., CA, NP, LPI and AI) were significantly higher in the initial stage of the peak growing season (June-July) than in the mid-and late stages of the peak growing season (July-September) in the years with WSRS (2011WSRS ( -2015WSRS ( , 2018 in the subregion located close to the south bank of YRE as the most prominent impact zone. Furthermore, our study identified the key flow-sediment regime variables affecting the intra-annual landscape pattern changes of S. alterniflora, namely, F (mean flow), F f (number of high flow pulses), T f (Julian date of maximum flow) and D (duration of WSRS). Our study further analyzed the relative importance of identified flowsediment regime variables to the change of intra-annual distribution patterns of S. alterniflora, and demonstrated that the increase of high pulse magnitude and the delay of high pulse occurrence time can be beneficial to S. alterniflora invasion, whereas the increase of high pulse frequency and duration can mitigate S. alterniflora invasion. Our results broaden the understanding of estuarine hydrological disturbance as a potential driver regulating the saltmarsh vegetation at the landscape scale, and also have implications for S. alterniflora invasion control at estuaries under changing environment.
It is worth pointing out that there are limited data in the years without WSRS (2016-2017), which may affect the accuracy of statistical results. In addition, S. alterniflora only started spreading in the YRE from 2011, by then the WSRS has been implemented for Frontiers in Environmental Science | www.frontiersin.org June 2021 | Volume 9 | Article 642442 9 years. As such, we cannot establish a rigorous control group to compare the inter-annual landscape pattern changes of S. alterniflora before and after the implementation of WSRS. In spite of this, our study divided the entire YRE into five sub-regions based on the sequence of the land formation and invasion (colonization) of the S. alterniflora and explored the expansion characteristics of S. alterniflora in different regions in the years with/without WSRS. Furthermore, our study focused on the change of vegetation landscape patterns through remote sensing image based spatial analysis, and future studies can be extended to examine the underlying mechanisms of the vegetation landscape changes in response to changing environmental factors, such as hydrodynamic disturbance, sediment load and nutrients supply, induced by the WSRS.

DATA AVAILABILITY STATEMENT
The original contributions presented in the study are included in the article/supplementary materials, further inquiries can be directed to the corresponding author.