Anthropogenic and Climatic Factors Differentially Affect Waterbody Area and Connectivity in an Urbanizing Landscape: A Case Study in Zhengzhou, China

Abstract

Urbanization alters the distribution and characteristics of waterbodies, potentially affecting both the habitat availability and connectivity for aquatic wildlife. We used Landsat satellite imagery to observe temporal and spatial changes in open-water habitats in Zhengzhou, a rapidly growing city in central China. We classified open water into six categories: perennial rivers, seasonal rivers and streams, canals, lakes, ponds, and reservoirs. From 1990 to 2020, in 5-year intervals, we identified, counted, and measured the area of each kind of waterbody, and we used a model selection approach with linear regressions to ask which climate and anthropogenic drivers were associated with these changes. We also used Conefor software to examine how these changes affected the landscape connectivity for waterfowl. Over the study period, lakes and canals were the only waterbody types to show statistically significant changes in surface area, increasing by 712% and 236%, respectively. Changes in lakes and canals were positively correlated with the length of water pipeline in the city. The connectivity of waterbodies fluctuated over the same period, mirroring fluctuations in the perennial Yellow River. Ponds contributed very little to landscape connectivity, and the importance of reservoirs decreased over time. Conversely, canals played an increasingly important role in landscape connectivity over time. Counterintuitively, the connectivity of waterbodies increased in the built-up part of the city. Our results show that urbanization can have unexpected effects—both positive and negative—on the connectivity and area of open-water habitats. These effects are likely to be important for waterfowl and other aquatic organisms.

1. Introduction

Aquatic habitats such as lakes, rivers, and streams are rich in biodiversity, and urban waterbodies provide many ecological functions and services that cannot be replaced by other systems. Open water provides wildlife habitat and groundwater recharge [1,2,3], in addition to direct benefits to people such as flood control, irrigation water, heat island effect mitigation, and recreational activities [4]. The presence of an urban waterbody is also related to significantly higher housing values within 5 km [5]. However, growing human populations are usually associated with profound reductions in wetland density and proximity [6], and wetlands in rapidly urbanizing areas face increasing degradation and loss [7,8,9]. In 2014, a review of 189 reports of wetland change indicated that 54–57% of the world’s wetlands have been lost in the last few centuries [10]. A more recent study of 1250 inland Ramsar wetlands around the world found that 47% of the sites experienced wetland loss between 1980 and 2014. The study predicted a net loss of at least 6000 km² of wetland area by 2100 [11].

During the process of urbanization, a rapid expansion of built-up areas causes an increase in land-surface temperatures [12] and evaporation rates [13], as well as an urban precipitation deficit [14]. Construction, drainage, and the reshaping of urban land may further influence the distribution and type of waterbodies present in cities [15]. This can result in dramatic changes to the surface water and hydrology of urban landscapes, as seen in the Ganga basin in India [16] and Shenzhen, China [17]. These and other anthropogenic changes can impact the connectivity of waterbodies [18,19], water quality [20], and habitat for wildlife [21]. Climate changes, such as increasing temperature and altered precipitation, will also impact urban wetlands [22]. However, there are regional differences in waterbody changes and their drivers during urban development [23,24], which means that further research is warranted to understand global trends.

Aquatic and wetland birds such as ducks may be negatively impacted by urbanization via the loss of natural wetlands [25], although this remains an understudied area of research [26]. Development near wetlands appears to be altering hydrology, leading to habitat degradation and declining populations of several wetland-dependent bird species [27]. In one city, wetland birds declined faster than all other bird groups during a 10-year period of urbanization, due largely to loss of habitat and connectivity [28]. One possible mitigation effort is the construction of new aquatic habitats. Artificial wetlands in Cyprus provide habitat for many, although not all, waterbird species found in natural wetlands, including 11 threatened and near-threatened species [29]. Although constructed waterbodies cannot wholly replace natural ones as habitats for waterbirds [30], they can provide alternative and complementary habitats for waterbirds at all stages of life [31] and mitigate the adverse effects of natural wetland degradation [32], especially for migrating birds [33]. Constructed wetlands can also contribute to landscape connectivity [34], which is important for conservation of biodiversity [35].

Since the 1980s, China has experienced a rapid economic development and expansion of urban land. The drivers of this expansion are diverse and vary from one region of the country to another [36]. These changes are generally seen as benefiting urban residents but also bring socioeconomic and ecological issues [37]. For example, Mao et al. (2018) [38] found that from 1990 to 2010, China’s wetland area decreased by 2883 square kilometers due to urban expansion, which accounts for a 6.0% loss of the total wetland area. A different analysis showed a net wetland gain of 1548 km² between 2000 and 2015, but this net change hid considerable complexities in the types of wetlands gained and lost. In particular, natural wetlands decreased by 7562 km² during the study period [39]. The ecological value of urban waterbodies is particularly significant in northern China, where the climate is arid and surface water is scarce [40]. In recent years, a series of waterbody protection measures have been proposed in China, and some have been implemented by the government. Some of these measures are focused on water quality [41,42], but others focus on stemming habitat and biodiversity loss, particularly for waterfowl [43]. These protection measures can potentially counteract negative impacts of urbanization, but their long-term impacts are not yet clear.

In this paper, we examined the impact of urbanization on the size, distribution, and connectivity of aquatic habitats in a rapidly growing region of central China. Using Landsat images from the United States Geological Survey (USGS), we examined the change in urban waterbodies over a 30-year period and investigated whether these changes were related to climate and/or anthropogenic drivers. We also examined changes in the landscape-level connectivity of aquatic habitats for waterfowl, and the contributions of different waterbodies to that connectivity. The goal of this study is to elucidate the underlying drivers of spatio-temporal waterbody changes and inform future water management and conservation efforts.

2. Materials and Methods

2.1. Study Area

Zhengzhou (34.7466° N, 113.6253° E) is the largest city of Henan province in China, with a total area of approximately 7446 km². Its population has nearly doubled over the last 30 years, from approximately 5.6 million in 1990 to 10.4 million in 2019 [44]. In 2018, Henan ranked first in urbanization growth rate among Chinese provinces; thus Zhengzhou provides a typical example of rapid urbanization in central China. Zhengzhou is considered a “prefecture-level” city and thus includes substantial rural areas as well as urban built-up areas; it contains six municipal districts, five county-level cities, and one county (Figure 1). The urban built-up area is approximately 1181 km² [45].

There are 124 rivers in Zhengzhou, with 29 large river basins belonging to the Yellow River or Huai River systems. The Yellow River system covers approximately one-quarter of the city’s total area and is the only perennial river in the area, while the Huai River system covers the remainder of the city. Two different water diversion projects have directly affected Zhengzhou’s waterbodies over the last 15 years: the South-to-North Water Diversion Project [46] and the Yellow River Diversion project [47]. The former project is the largest such project in Chinese history, and has the goal of transferring fresh water from southern China to northern China [48,49,50]. The latter project is much smaller in geographic scope and has the goal of drawing water from the Yellow River to enrich the Jialu River and Long Lake and Longzi Lake in Zhengzhou itself. Following these projects, in 2017, the government of Zhengzhou adopted the Wetland Protection Regulation, which required the forestry department to carry out a field survey of wetland resources. The survey mapped waterbodies and gathered ecological, geographical, ownership, management, and protection information for approximately 95% of waterbodies across the city. This manuscript is one outcome of that survey effort.

2.2. Satellite Image Acquisition and Processing

To evaluate the effect of urbanization on waterbodies in Zhengzhou, we observed changes in the landscape over the 30-year time period between 1990 and 2020. We chose 30 years as the study period because it covers the urbanization process of Zhengzhou, it is long enough to observe the effects of climate and anthropogenic factors on urban waterbodies, and data were available for this time period. We analyzed the landscape every 5 years during the 30-year period: in 1990, 1995, 2000, 2005, 2010, 2015, and 2020.

Seven Landsat satellite images were downloaded from EarthExplorer (USGS., https://earthexplorer.usgs.gov/, accessed on 20 July 2021), one for each year of the study (Figure 2). Landsat images have a 30 m resolution; this resolution imagery has been previously used for water identification [51,52,53]. Cloud-free images were selected from Landsat-5 TM (1990–2010), Landsat-7 ETM+ (2015), and Landsat-8 OLI/TIRS (2020), and reprojected into the World Geodetic System (WGS) 1984_UTM_49N coordinates. Each image was acquired in either May (2015, 2020) or June (1990–2010), to minimize the effects of seasonal variation in precipitation on the waterbody size [54]. Zhengzhou receives approximately 6 cm of precipitation in May and 9 cm in June [55,56].

In this paper, the Modified Normalized Difference Water Index (MNDWI) [53] was used to identify open water across Zhengzhou. MNDWI is similar to the Normalized Difference Water Index (NDWI) [57] but with the substitution of a middle-infrared (MIR) band, such as the Landsat TM band 5, which can be helpful in built-up and urban areas [53,58]. MNDWI uses the green and MIR bands to enhance open water features (Equation (1)). It also is sensitive to built-up features that are often correlated with open water in other indices.

MNDWI = (Green-MIR)/(Green + MIR)

(1)

Images were pre-processed using ENVI 5.3. We first calculated the MNDWI index, separated land and water using the band-of-interest thresholding tool, and performed a subset data extraction of water using the ROIS tool. All open water was then manually classified as one of the following seven waterbody types: perennial rivers, seasonal rivers and streams, lakes, canals, reservoirs, ponds, and aquaculture. Aquaculture sites were identified but excluded from further analysis because their value as bird habitats is hard to predict and can be negative [59,60], leaving six waterbody types in our analysis (

Table 1. Six classes of open waterbodies analyzed in this study.

Waterbody Type	Preview	Description
Perennial river		A stream or river (channel) that has continuous flow in parts of its stream bed throughout the year during years of normal rainfall. In this paper, the perennial river refers specifically to the Yellow River.
Seasonal rivers and streams		Seasonal rivers and streams are particularly affected by seasonality and partially dry up during the dry season due to insufficient recharge.
Lakes		Permanent freshwater lakes; seasonal/intermittent freshwater lakes (>8 ha *). Could include natural or constructed waterbodies.
Canals		Constructed canals and drainage channels.
Reservoirs		Constructed water storage areas and impoundments (generally > 8 ha *).
Ponds		Comparatively small constructed or natural shallow waterbodies (lentic ecosystems) including irrigation ponds, stock ponds, and small water tanks (generally < 8 ha *).

* The 8 ha cut-off between ponds and lakes was consistent with the classification used by the National Wetland Resources Survey.

). Manual classification relied largely on the map of wetland resources that was created during the aforementioned field survey of 2018; each identified waterbody was examined by eye, compared to the survey map, and then classified. When necessary, the classification and interpretation of waterbodies was also aided by information provided by the Zhengzhou Bureaus of Forestry and Water Resources.

2.3. Accuracy Assessment

We tested the accuracy of the open-water classification using stratified random sampling [61] with different waterbody types as the basis for stratification. We determined the sample size based on a 95% confidence level (α = 0.05) to create a confusion matrix [62,63]. We calculated the overall accuracy and kappa coefficient as well as the true positive rate (sensitivity) and the true negative rate (specificity) of each waterbody type. Gaofen-2 satellite images with 0.8 m resolution and timelapse imagery in Google Earth were the primary reference data sources.

2.4. Changes in Waterbody Distribution

To quantify the changes in waterbodies over time, we computed the following summary statistics for each type of waterbody in each 5-year period of the study: the total surface area of all waterbodies combined, maximum surface area of any individual waterbody, and total count of individual waterbodies. We focus our statistical analysis on total surface area but use the other summary statistics to provide additional insights about changes across the landscape. To test for significant changes in the total surface area of each waterbody type over time, we conducted a correlation analysis between the total surface area and time (i.e., year).

2.5. Potential Drivers of Waterbody Change

To evaluate potential drivers of waterbody change, we gathered a series of climate and anthropogenic variables that have been linked to waterbody changes in other studies: temperature [11], precipitation [64], potential evapotranspiration [65], human population density [66], the extent of built-up area [67], and the length of water-supply piping [68]. Each variable was measured repeatedly for each 5-year interval of the study.

Temperature and precipitation data for Zhengzhou were obtained from the China Meteorological Data Service Center (CMDC, https://data.cma.cn/en, accessed on 19 September 2021). These data included the average monthly temperature over the previous 12 months and average precipitation of the previous 6 and 12 months before the satellite image date. We calculated potential evapotranspiration using the SPEI package [69] in R [70]. The human population density, extent of built-up area, and length of water-supply piping in Zhengzhou were obtained from the Bureau of Statistics of Zhengzhou (http://tjj.zhengzhou.gov.cn/, accessed on 17 April 2021).

2.6. Linear Regression and Model Selection to Examine the Influence of Climatic and Anthropogenic Factors on Waterbody Change over Time

One priority of this study was to understand the drivers of waterbody change over time and, in particular, to contrast the relative importance of anthropogenic versus climatic drivers. However, several potential drivers were highly correlated over the time period (

Table 2. Correlation coefficients (lower half of matrix) and p-values (upper half of matrix) between possible drivers of waterbody change. Bolded correlation coefficients are significant at p ≤ 0.05. Variables include built-up area (BUA), length of water pipeline (LWP), population density (PD), potential evapotranspiration (PET), average monthly precipitation in the 6 months prior to satellite images capture (PRCP6), average monthly precipitation in the 12 months prior to satellite images capture (PRCP12), and annual mean temperature (AMT).

	Year	BUA	LWP	PD	PET	PRCP6	PRCP12	AMT
Year		0.0003	0.0009	<0.0001	0.0171	0.7383	0.4633	0.0004
BUA	0.9694		0.0002	<0.0001	0.046	0.8932	0.3306	0.0012
LWP	0.9538	0.9738		0.0004	0.0296	0.9376	0.3467	0.0002
PD	0.9862	0.9838	0.9656		0.0305	0.934	0.5092	0.0001
PET	0.8435	0.7631	0.803	0.8007		0.4852	0.3428	0.031
PRCP6	−0.156	0.0631	−0.0368	−0.0389	−0.3193		0.6625	0.7472
PRCP12	0.3345	0.434	0.4211	0.3028	0.4243	0.203		0.6363
AMT	0.965	0.9464	0.9722	0.9778	0.7993	−0.1506	0.2195

). The year was highly correlated (r > 0.95) with all anthropogenic variables and also significantly correlated with potential evapotranspiration and average monthly temperature. Given that the year is only a proxy for other changes within the city over time, it was excluded from further analyses. Correlations among anthropogenic variables were also extremely high (r > 0.96). Therefore, to avoid multicollinearity in our models, we removed the temperature and potential evapotranspiration as candidate variables and separated the remaining variables into two categories (climatic versus anthropogenic). Following this, climatic variables included six- and twelve-month precipitation, and anthropogenic variables included built-up area, length of water supply piping, and population density. There were no significant correlations between the remaining variables in different groups (Table 2).

Linear modeling was used to assess the drivers of waterbody change from 1990 to 2020. Each waterbody type was considered independently, with the total surface area of each waterbody type in each 5-year interval as the response variable. Predictor variables included the climatic and anthropogenic variables described above, each measured in 1990, 1995, 2000, 2005, 2015, and 2020. To ensure that highly correlated variables were excluded from the same model and to keep the number of predictor variables sufficiently low for our small sample size, we first included only one climatic or anthropogenic variable in each model. The best-supported variable for each category was determined using the Corrected Akaike Information Criterion (AICc) [71] to compare linear models with single predictor variables. Finally, the two best-supported (and uncorrelated) predictor variables were combined into a single linear model for each waterbody type. Statistical analyses were performed in RStudio.

2.7. Evaluating Landscape Connectivity

To evaluate changes in aquatic habitat connectivity over time, we used Conefor 2.6 [72]. Conefor is a frequently used software package for evaluating landscape- and patch-level connectivity in a variety of habitats and landscapes [73,74,75]. In our analyses, individual waterbodies were considered ‘patches’ and connections between patches were considered from the perspective of waterfowl (i.e., family Anatidae). Among waterfowl, one of the most well-studied species is the mallard (Anas platyrhynchos), a duck with generalist feeding habits that breeds in a wide variety of habitats and has broad distribution around the world [76,77,78,79]. This species has been shown to use small urban waterbodies [80] and was observed at many Zhengzhou waterbodies during the wetland resource survey described above; thus we used published mallard-related data to parameterize our connectivity models. Research has shown that the mallard’s average flight distance is 0.4 km [81], so we assumed that many ducks would travel this distance in a single flight event.

We used the waterbody maps created for each time period (1990, 1995, 2000, 2005, 2010, 2015, 2020), along with the Conefor ArcGIS extension, to create the ‘node’ (i.e., patch) and ‘connection’ input files for Conefor 2.6. Each year was analyzed separately. In Conefor, we set the connectivity distance to 0.4 km and the probability value to 0.5. We used the probability of connectivity index (PC) [82] to analyze the landscape- and patch-level connectivity changes over time for waterfowl in Zhengzhou. PC represents the probability that two randomly placed points within the landscape are accessible from each other (i.e., interconnected), and is calculated as follows in Equation (2):

$$\mathrm{PC}=\frac{{{\displaystyle \sum }}_{i=1}^n{{\displaystyle \sum }}_{j=1}^n{a}_i\times a_j\times p_{ij}^{*}}{A_L^2}=\frac{{\mathrm{PC}}_{num}}{A_L^2}$$

(2)

where ${\mathrm{a}}_{\mathrm{i}}$ and ${\mathrm{a}}_{\mathrm{j}}$ are the area (or other attributes) of patches $\mathrm{i}$ and $\mathrm{j}$, ${\mathrm{p}}_{\mathrm{ij}}$ is the probability of direct movement between patches $\mathrm{i}$ and $\mathrm{j}$, and ${\mathrm{A}}_{\mathrm{L}}$ represents the total landscape area, including habitat and non-habitat patches. PC ranges from 0 to 1; when PC equals 0, there are no habitat patches in the landscape, and when the entire landscape is covered by habitat, the PC value is equal to 1.

PC is a landscape-level attribute, but the proportion of connectivity attributed to each patch can be calculated by removing that patch and recalculating the landscape attribute, as shown in Equation (3) [83]:

$${\mathrm{dPC}}_k=100\times \frac{\mathrm{PC}-{\mathrm{PC}}_{remove,k}}{\mathrm{PC}}=100\times \frac{\Delta {\mathrm{PC}}_k}{\mathrm{PC}}$$

(3)

The resulting change in PC(dPC) measures the importance of that patch to landscape connectivity. dPC can be further divided into three parts according to a patch’s contribution to different aspects of connectivity. dPCintra measures within-patch connectivity and is a function of available habitat within the patch. dPCflux measures the area-weighted dispersal flux through a patch. Finally, dPCconnector measures the degree to which a patch connects other patches on the landscape [84].

We first examined PC in each 5-year interval to look for landscape-scale changes in connectivity over time. We subsequently examined differences in dPC among waterbody types to see if certain types of waterbodies contributed more or less to landscape-scale connectivity. The Yellow River was excluded from the dPC analyses because its large surface area and spread from east to west Zhengzhou caused it to dominate all indices and obscured changes in the other waterbody types.

As the data were not normally distributed, we used Kruskal–Wallis [85] and Dunn’s post hoc tests [86] to examine the differences in dPC values between different waterbody categories in each study year, with significance based on Bonferroni corrections. Non-parametric statistics and visualization were performed in RStudio, and we used ArcGIS 10.2.2 to map the dPC of each waterbody in each study period. Finally, for each study year, we extracted the top 100 waterbodies with respect to each aspect of connectivity (dPCintra, dPCflux, and dPCconnector) in a separate analysis of waterbody connectivity over time.

3. Results

3.1. Waterbody Classification Accuracy

The overall classification accuracy ranged from 87% to 96%, and the final kappa coefficients for the classification scheme ranged from 0.83 to 0.94 (

Table 3. The overall accuracy and kappa coefficients for the waterbody classification.

Year	Overall Accuracy	Kappa Coefficients
1990	93%	0.91
1995	92%	0.89
2000	93%	0.90
2005	95%	0.93
2010	87%	0.83
2015	89%	0.84
2020	96%	0.94

). Cohen [87] suggested that kappa values between 0.80 and 0.90 be interpreted as ‘strong’ and values above 0.90 as almost perfect agreement, so our classification scheme performed very well based on these results. Sensitivity and specificity are shown in Appendix A.

3.2. Waterbody Dynamic Changes

From 1990 to 2020, the number, size, and total surface area of different kinds of waterbodies changed substantially (Figure 3 and Figure 4). The only statistically significant changes were seen in the total surface area of lakes and canals, both of which increased over time (lakes: r = 0.87, p = 0.01; canals: r = 0.82, p = 0.02). The total surface area of lakes increased from 116 hectares in 1990 to 942 hectares in 2020, an increase of 712%. This change was clearly due to the increase in number of lakes and the addition of large lakes (Figure 3). The total surface area of canals increased 236% over the same time period, from 310 to 1042 hectares. The perennial river and the reservoirs both fluctuated in surface area over time, with peaks in the first year of the study and again in 2005. Ponds and seasonal rivers and streams showed relatively little change over the study period (Figure 3).

3.3. Driving Factors of Urban Waterbody Change

Over the 30-year study period, the total population of Zhengzhou increased almost linearly from 5.578 million to 10 million people and the built-up area and length of water pipeline followed a similar trend (Figure 5). These land-use change patterns are likely responsible for the observed concurrent increase in annual mean temperature and potential evapotranspiration (Figure 5). Precipitation fluctuated over the study period, with the 6-month precipitation highest in the first year of the study and the 12-month precipitation highest in 2005.

Among the climatic variables, six-month precipitation was the best predictor of the surface area of the perennial river, seasonal rivers and streams, and reservoirs (Table 4). Twelve-month precipitation was the best predictor of the surface area of lakes, canals, and ponds. However, there was little difference in AICc between the two precipitation variables for most waterbodies. Among the anthropogenic variables, population density was best supported for the perennial river, reservoirs, and ponds; extent of built-up area was selected for seasonal rivers and streams; and the length of water pipeline was the best supported for lakes and canals (Table 4).

Table 4. Selection of the most predictive climatic and anthropogenic variables for the total surface area of each waterbody type. Corrected Akaike Information Criterion (AICc) values are shown in the table and were used to select the predictor variables for the linear regression models shown in Table 5. The variables of AICc from low to high are indicated by ① ② ③.

Waterbody	Climatic		Anthropogenic
	6-Month Precipitation	12-Month Precipitation	Population Density	Built-Up Area	Length of Water Pipeline
Perennial river	140.3892 ①	141.5736 ②	142.8388 ①	143.4732 ③	143.0368 ②
Seasonal rivers and streams	103.0762 ①	110.3008 ②	112.5721 ③	111.8918 ①	112.3397 ②
Lakes	113.1882 ②	112.4519 ①	101.7744 ③	99.1148 ②	97.6410 ①
Canals	112.6251 ②	111.0877 ①	103.6491 ③	100.1516 ②	98.4526 ①
Reservoirs	121.6092 ①	122.2174 ②	120.8713 ①	122.0940 ③	121.3000 ②
Ponds	89.1690 ②	88.8273 ①	88.7068 ①	88.8032 ②	88.9787 ③

Table 4. Selection of the most predictive climatic and anthropogenic variables for the total surface area of each waterbody type. Corrected Akaike Information Criterion (AICc) values are shown in the table and were used to select the predictor variables for the linear regression models shown in Table 5. The variables of AICc from low to high are indicated by ① ② ③.

Waterbody	Climatic		Anthropogenic
	6-Month Precipitation	12-Month Precipitation	Population Density	Built-Up Area	Length of Water Pipeline
Perennial river	140.3892 ①	141.5736 ②	142.8388 ①	143.4732 ③	143.0368 ②
Seasonal rivers and streams	103.0762 ①	110.3008 ②	112.5721 ③	111.8918 ①	112.3397 ②
Lakes	113.1882 ②	112.4519 ①	101.7744 ③	99.1148 ②	97.6410 ①
Canals	112.6251 ②	111.0877 ①	103.6491 ③	100.1516 ②	98.4526 ①
Reservoirs	121.6092 ①	122.2174 ②	120.8713 ①	122.0940 ③	121.3000 ②
Ponds	89.1690 ②	88.8273 ①	88.7068 ①	88.8032 ②	88.9787 ③

After identifying the best climatic and anthropogenic drivers for each type of waterbody, we combined them into a single linear regression model to explain the total surface area of each waterbody type. Models for the seasonal rivers and streams, lakes, and canals had high explanatory power (adjusted r² > 0.83), with the anthropogenic variable having a significant positive effect on the total surface area of lakes and canals and precipitation having a significant positive effect on the surface area of seasonal rivers and streams. Regression models for the perennial river, reservoirs, and ponds did not have high explanatory power or any significant variables.

Table 5. Linear regression for the total surface area of each type of waterbody (waterbody~ climatic variables + anthropogenic variables). Significant variables are highlighted with bold text. PRCP6 represents average monthly precipitation in the 6 months prior to satellite image capture; PRCP12 represents average monthly precipitation in the 12 months prior to satellite image capture; PD represents population density; BUA represents built-up area; LWP represents length of water pipeline.

Waterbody	Climate			Anthropogenic
	Variable	Coefficient	p-Value	Variable	Coefficient	p-Value	Adjusted R2 of Full Model
Perennial river	PRCP6	18.8420	0.1410	PD	−4.2270	0.3380	0.3106
Seasonal rivers and streams	PRCP6	2.8564	0.0066	BUA	0.6492	0.0951	0.8381
Lakes	PRCP12	−0.3556	0.6902	LWP	0.2022	0.0052	0.8466
Canals	PRCP12	0.5955	0.5213	LWP	0.1802	0.0085	0.8352
Reservoirs	PRCP6	3.5816	0.1920	PD	−1.5376	0.1500	0.3909
Ponds	PRCP12	0.1352	0.7640	PD	0.0477	0.6980	−0.3652

Table 5. Linear regression for the total surface area of each type of waterbody (waterbody~ climatic variables + anthropogenic variables). Significant variables are highlighted with bold text. PRCP6 represents average monthly precipitation in the 6 months prior to satellite image capture; PRCP12 represents average monthly precipitation in the 12 months prior to satellite image capture; PD represents population density; BUA represents built-up area; LWP represents length of water pipeline.

Waterbody	Climate			Anthropogenic
	Variable	Coefficient	p-Value	Variable	Coefficient	p-Value	Adjusted R2 of Full Model
Perennial river	PRCP6	18.8420	0.1410	PD	−4.2270	0.3380	0.3106
Seasonal rivers and streams	PRCP6	2.8564	0.0066	BUA	0.6492	0.0951	0.8381
Lakes	PRCP12	−0.3556	0.6902	LWP	0.2022	0.0052	0.8466
Canals	PRCP12	0.5955	0.5213	LWP	0.1802	0.0085	0.8352
Reservoirs	PRCP6	3.5816	0.1920	PD	−1.5376	0.1500	0.3909
Ponds	PRCP12	0.1352	0.7640	PD	0.0477	0.6980	−0.3652

3.4. Landscape Connectivity Analysis

While the PC did not consistently decrease over time, the connectivity was greatest at the beginning of the study period in 1990 (Figure 6). The connectivity briefly increased in 2005 before dropping off again but was still approximately 36% less than the PC value in 1990 (Figure 6).

In comparing the contribution of individual waterbodies to overall landscape connectivity, we see significant differences between waterbody types in every year except 2000 (p < 0.001 for Kruskal–Wallis tests, Figure 7, Appendix B). Ponds have significantly lower dPC than other waterbody types most years of the study. Reservoirs and lakes have the highest dPC in early years of the study, but canals have the highest dPC in later years. By 2020, the main urban area of Zhengzhou appears to contain the waterbodies with the highest contribution to connectivity within the city (Figure 8).

Our analysis of the top 100 waterbodies that contribute most to landscape connectivity each year showed that seasonal rivers and streams are very important for all aspects of connectivity (

Table 6. Number of waterbodies in each category that are in the top 100 for each aspect of landscape connectivity (dPCintra, dPCconnector, and dPCflux). The perennial river was excluded to better understand the changes in connectivity of the other waterbody types in Zhengzhou.

dPC Index	Waterbody Type	1990	1995	2000	2005	2010	2015	2020
dPCintra	Seasonal rivers and streams	43	45	42	39	36	47	29
	Canals	7	6	8	7	12	13	26
	Lakes	3	7	5	9	12	13	20
	Reservoirs	46	37	40	45	40	27	25
	Ponds	1	5	5	0	0	0	0
dPCconnector	Seasonal rivers and streams	85	82	79	72	53	65	35
	Canals	9	13	20	16	37	29	57
	Lakes	0	0	0	3	0	1	8
	Reservoirs	6	5	1	9	10	5	0
	Ponds	0	0	0	0	0	0	0
dPCflux	Seasonal rivers and streams	66	71	71	58	51	63	40
	Canals	11	13	11	13	27	15	41
	Lakes	2	2	2	3	5	11	15
	Reservoirs	21	14	16	26	17	11	4
	Ponds	0	0	0	0	0	0	0

). However, their importance seemed to diminish over time. Reservoirs were also quite important for dPCintra and, to a lesser degree, dPC flux. Canals increased in importance over time for all aspects of connectivity but seemed to play the largest role in dPCconnector.

4. Discussion

Our study illustrates substantial changes in the surface water distribution and connectivity in Zhengzhou over a 30-year timespan, which were driven by both climatic and anthropogenic factors. These results highlight the high spatiotemporal variability that can characterize freshwater aquatic systems. This kind of variability has been seen in other Chinese cities [88,89] and in a rapidly urbanizing region in southwest Australia [90] over similar time periods. We found statistically significant increases in the total surface area of lakes and canals over time, while the area of reservoirs and the perennial river fluctuated but showed a non-significant decreasing trend from 1990 to 2020. The increase in area of lakes was accompanied by an increase in number of lakes and size of the largest lake, as several new lakes were created in the built-up areas of the city over the last two decades (Figure 4). Canals were also created as part of the two water diversion projects in Zhengzhou. The wetland resource surveys commissioned by the Zhengzhou government as well as some local news reports [91,92] show that these constructed waterbodies are used by many wildlife species, including waterfowl. This use by wildlife suggests that their construction could potentially offset other negative effects of urbanization. However, the increase in waterbody area we observed in the urban core is not the norm in most cities. For example, a study of 32 other major Chinese cities found that the total surface area of urban lakes decreased by 24% from 1990 to 2015 [93]. Therefore, we want to caution that urbanization in general is unlikely to have net positive effects on waterfowl or other wildlife.

We found that the length of water pipe was correlated with an increase in surface area of lakes and canals, and precipitation correlated with an increase in surface area of seasonal rivers and streams. Liu et al. (2020) [94] also found that anthropogenic variables had a stronger effect on lakes than natural variables, but they mainly found a negative effect on lake area change. This disparity may be explained by the fact that Liu et al. (2020) focused on natural lakes, whereas our “lake” category included constructed lakes as well. We likely see a positive effect of anthropogenic variables on the lake and canal area because they are correlated with human construction of these waterbody types. Seasonal rivers and streams, however, are highly dependent on precipitation for recharge [95], which may explain why climatic variables played a stronger role in determining the change in their surface area. The Yellow River Diversion Project has been drawing water from the Yellow River since 2009 [96], so we might expect to see decreases in the surface area of the Yellow River associated with anthropogenic variables. However, none of our predictor variables were significantly related to area of the Yellow River. On visual inspection, the fluctuations of the Yellow River resemble a superimposition of the 6-month and 12-month precipitation patterns, so it may be that we did not measure precipitation at the most relevant time scale for the Yellow River. It is also likely that the area of the Yellow River will decline over time due to the continued diversion of water.

The overall landscape connectivity, as measured by the PC index, fluctuated in Zhengzhou over time. The temporal pattern in connectivity appears similar to the pattern for the total surface area of the perennial river. The perennial river, i.e., the Yellow River, contains the majority of the water in Zhengzhou and spans the entire city from east to west. For this reason, the perennial river plays a dominant role in the overall connectivity of aquatic habitats. If the area of the Yellow River does decline over time in the future, as we speculated above, then that could have long-term and large-scale impacts on landscape connectivity. However, other kinds of waterbodies also contribute to landscape connectivity. Seasonal rivers and streams, for example, play an important role in connecting other waterbodies and supporting flux of waterfowl across the landscape, as seen by their high values of dPCflux and dPCconnector. This is likely because their linear form gives them proximity to many other waterbodies. Interestingly, we noticed increased connectivity values of waterbodies in the main urban district over time. This finding is counter to an analysis of 100 cities in the United States of America, in which the connectivity of waterbodies was found to decrease in the most urbanized areas [15].

Many of the changes we observed in Zhengzhou seem to be unique among urban areas. Some of these changes—in particular, the addition of artificial waterbodies and associated increase in connectivity in the most urban parts of Zhengzhou—can be attributed to specific local and regional planning and policies. For example, in 2000, an old military airport was removed from Zhengzhou and the municipal government proposed a new town in its place. A plan was developed by a Japanese architect that involved the creation of several new waterbodies in what is now the main urban area [97]. One of the new lakes became the waterbody with the greatest overall contribution to landscape connectivity in the region (Figure 8). Other changes in the landscape, particularly the canals, are attributed to the subsequent regional hydraulic projects [49]. As development in Zhengzhou was specifically planned to include open water, the patterns we detected in this system may be different from those seen in cities that did not intentionally include open water in their design.

Although the importance of wetlands has been recognized for many decades [98], urban wetlands have been relatively neglected compared to more natural habitats. Previous urban water system planning has been largely focused on meeting hydrological functions such as water supply and flood control [99,100], and most studies on waterbird protection have focused on areas such as nature reserves, coastal wetland landscapes [101], or more rural landscapes [102]. One exception is a study of Beijing, China, that examined habitat suitability and connectivity for the mallard duck (along with two other bird species; [80]), although only a single time period was examined. Our paper provides a novel contribution by exploring urban waterbird protection from the perspective of changes in the connectivity of urban water bodies over time. We hope to provide a new perspective on regional water system planning and waterbird conservation in urbanized areas and suggest that the overall landscape connectivity of the entire city or even region should be considered during planning. Water management strategies that disproportionately focus on certain areas within a landscape but ignore others can be problematic for migratory or otherwise highly mobile organisms such as waterbirds that require resources from multiple spatially distinct waterbodies [103]. The conservation or construction of waterbodies outside of the main urban areas may help contribute to a robust network of aquatic habitats region-wide, which may in turn help to increase the resilience of waterbird populations in the face of a changing climate [104].

Our results show both increases and decreases in aquatic habitat and connectivity over time, with changes driven by both anthropogenic and climatic factors. While we believe these results reflect real changes happening in Zhengzhou, they should be considered in light of some limitations of this study. In particular, we used only seven time periods for our analysis. A longer temporal extent might reveal when key changes in aquatic habitat began, and more frequent sampling periods could potentially reveal stronger relationships with climatic drivers. Moreover, the use of satellite imagery with higher spatial resolution could prove useful for monitoring smaller waterbodies and subtle shoreline changes in larger waterbodies. Future studies could also validate the classification of key waterbodies by using high-resolution GNSS receivers as well as more automated classification methods. Additional studies could consider the impacts of redistributing water resources at a larger scale (e.g., across the entire country) and how best to maintain and improve urban surface-water landscape connectivity. It would also be interesting to conduct a more qualitative analysis of the many complex social factors that affect land use and land cover change over time, as our variables did not explain the changes in all waterbody types. Finally, we suggest that future research could validate our connectivity analyses with fieldwork (e.g., placing GPS trackers on mallards) and expand the focus to other waterfowl or other taxa that use urban waterbodies. The work presented here is a first step toward understanding changes in aquatic habitats in Zhengzhou and we hope that it can assist in informing future urban planning strategies that benefit both people and wildlife.

5. Conclusions

Aquatic habitat availability and connectivity are critical to maintaining biodiversity and ecosystem services in urban landscapes. However, these aquatic habitats are often threatened by the process of urbanization. An in-depth analysis of the changes in aquatic habitats and their underlying drivers can help planners to propose efficient and targeted conservation and planning strategies. Using remotely sensed images from 1990 to 2020, we studied the spatial distribution, connectivity, and changes of urban waterbodies in a rapidly urbanizing area of central China. We found evidence of substantial changes in surface water distribution and connectivity in Zhengzhou over 30 years, driven by both climate and anthropogenic changes. Some categories of waterbodies significantly increased in surface area, while others fluctuated over time but showed decreasing trends. These changes in waterbody area caused fluctuations in landscape connectivity and, unintuitively, an apparent increase in connectivity in the urban built-up area. These results show that urbanization can have unexpected and potentially positive effects on the connectivity of surface water, although these results might not translate directly to positive effects on wildlife.