Introduction

Natural wetland refers to the natural, permanent, or temporary marsh, peatland, or water area, static or flowing fresh water, brackish water, and salt water body, including the water depth of not more than 6 m at low tide; meanwhile, it also includes the coastal areas of rivers and lakes adjacent to wetlands, coastal areas, and islands within the scope of wetlands or seawater with a water depth of no more than 6 m at low tide (Sarkar, 2021). Natural wetlands in the semiarid regions have been regarded as environmentally, economically, and ecologically (Darrah et al., 2019; Li et al., 2021; Lovvorn & Crozier, 2022) valuable resources because they have important effects on regulating climate, conserving water resources, and maintaining biodiversity (He & Wang, 2019). However, anthropogenic disturbances have altered natural wetlands in recent centuries through a number of processes (e.g., fire, agricultural reclamation, and urban development) (Qu et al., 2018), which has promoted land use/land cover change (LUCC), including the conversion of wetlands to cultivated land and construction of impervious urban buildings (Do & Joo, 2015; Evans et al., 2017). Liu et al. (2016a) constructed the Hemeroby index as a human disturbance index to study the changes of the wetland landscape pattern and its response to anthropogenic interference in Xiaosanjiang Plain from 1955 to 2010. With the policy implementation of the 10 billion kg grain project in Jilin Province, a large number of other lands have been developed as reserve cultivated land resources (Wang et al., 2021). Natural wetlands have not only been altered by human activities but also by climate change (Rogers et al., 2019). A large number of previous studies have reported that climate change altered the wetlands through changing the hydrological conditions of the wetland (Gong et al., 2020; Li et al., 2020; Liang, 2020); for example, Zhang et al. (2018) used a simple isotope-based model to study the effects of evaporation changes on wetlands a semiarid marsh wetland in northeastern China. Especially in semiarid regions, decreased mean annual precipitation deteriorates wetland hydrology and makes the wetlands dry easily, leading to the continuous degradation and loss of wetland landscapes (Chen et al., 2019; Lovvorn & Crozier, 2022). Previous studies on landscape patterns mainly applied landscape dynamic models and landscape metrics based on remote sensing and Geographic Information System (Gyenizse, 2021; Hua et al., 2019; Suir & Sasser, 2019; Zhao et al., 2017). Landscape metrics are mostly used to quantify wetland landscape structural and pattern changes and include landscape composition metrics (e.g., evenness, diversity, and proportional abundance) and landscape configuration metrics (e.g., shape index, patchiness index, connectivity index, and fractal dimension) (Evans et al., 2017; Silva et al., 2021). In addition, changes in the wetland landscape pattern are spatially heterogeneous because landscape patterns are the results of various ecological processes on different scales (Liang & Li, 2018). The effect of different scales has been studied extensively on landscape pattern changes by using a moving window method. Thompson-Ambriz et al. (2020) found a scale-dependent dissimilarity between the use of extensive estuarine and small inland brackish wetland landscape by using ordination techniques and generalized linear models. Thompson-Ambriz et al. (2020) analyzed the scale effects on wetland bird assemblages to landscape patterns. Zhang et al. (2014) concluded that it is necessary to select an appropriate scale for the analysis of landscape patterns because scale changes affect the quantitative analysis. The moving window refers to cutting off a square quadrat according to a certain window size and moving step length along the transect, then extract the landscape in the quadrats (Cao et al., 2018). Compared with traditional methods based on overall landscape characteristics, using the moving window method to analyze the landscape pattern can be implemented on the quantization and spatial visualization of regional landscape metrics (Cao et al., 2018; Hoffmann et al., 2016). For instance, Lv et al. (2019) studied the wetland landscape evolution and its relation to human disturbance in the Xiong’an New Area of China based on the moving window method. Xu et al. (2011) discussed that the percentage reduction and area reduction of natural wetlands were respectively quantified within different wetland cover zones using the moving window analysis. Thus, landscape metrics combined with the moving window method can demonstrate the dynamic changes in landscape patterns and reveal the internal structural characteristics of a landscape (Gallardo-Cruz et al., 2018).

Western Jilin Province is a semiarid region of China, which is an area sensitive to climate change and that has intensive human disturbance. The increased temperature, less precipitation, and strong sandstorm have led to desertification and salinization, threatening the regional ecological environment (Wan et al., 2021). Natural wetlands experienced great changes in this area in recent decades, including impacts from various abiotic and biotic environmental factors (e.g., precipitation, temperature, soil, topography, and anthropogenic activities) (Lei et al., 2017; Yu et al., 2019). Among these environmental factors, anthropogenic disturbance involved in LUCC was considered to be the primary factor driving landscape changes. Because Wang et al. (2021) found that cultivated land and artificial surface have been increasing in recent years when studying the land use change in the west of Jilin Province. Particular precipitation conditions in this semiarid region were important to wetland landscape changes because they can provide important water sources (Zhang et al., 2019). Furthermore, studies showed that precipitation had become the main factor affecting wetland dynamic changes (Chen et al., 2019). However, the mechanism promoting spatial variability in wetland landscapes is still unclear because information regarding the correlation of wetland landscape patterns with anthropogenic disturbances and mean annual precipitation is lacking.

In this study, we investigated the spatial differences in landscape metrics using a moving window method and examined their relationships with anthropogenic disturbance and mean annual precipitation in a semiarid region using SPSS. Specifically, our research questions included the following: (1) How change the wetland landscape did during the period 1986–2019; (2) How diverse the change in wetland landscape features is; (3) What are the relationships between the heterogeneity of wetland landscape with anthropogenic disturbances and mean annual precipitation?

Materials and methods

Study area

The study area is located in western Jilin Province, China, extending from 123° 09′ to 124° 22′ E and 44° 57′ to 45° 46′ N (Fig. 1). The area is 4,689,731 ha in size and is semiarid with a humid monsoon climate. Average relative humidity ranges from 60 to 65%, annual precipitation ranges from 350 to 450 mm (Liu et al., 2019), and evaporation ranges from 1600 to 2000 mm (Liu et al., 2016b). Water systems in the study area are mainly the Songhua River and Nen River. This region belongs to the eastern ecotone between agriculture and prairie in northern China, and therefore, wetlands in this region play an important role in preventing rapid desertification and salinization. There are 4 national natural reserves, namely, the Xianghai National Natural Reserve, Chaganhu National Natural Reserve, Momoge National Natural Reserve, and Dabusu National Natural Reserve. Wetlands referred to rivers, lakes, and marshes in our study.

Fig. 1
figure 1

Study area, western Jilin Province, China

Full size image

Land cover and climate data

Landsat TM/ETM + /OLI images at three time slices (1986, 2000, and 2019) were used to interpret land cover/land use maps of western Jilin Province at a spatial resolution of 30 m. These images were captured from mid-May to mid-September and were downloaded from the USGS Center for Earth Resources Observation and Science (http://glovis.usgs.gov) and Geospatial Data Cloud (http://www.gscloud.cn/). We corrected all images by removing radiometric and atmospheric effects using the radiometric calibration and FlAASH atmospheric correction methods provided in ENVI 5.2. We have performed some multispectral compositions. We performed 6, 5, 4 multi-band synthesis of OLI image and 5, 4, 3 multi-band synthesis of TM image. We applied object-oriented classification to classify the land cover/land use by eCognition8.9 (Du et al., 2019) The object-oriented classification method has been proven to be more accurate than traditional pixel-based methods (Balasubramaniam et al., 2018). The culmination of the ENVI program and input environmental variables (e.g., the DEM and soil types) led to a greater accuracy (Man et al., 2016). The resolution of DEM is 30 m and the resolution of the soil type is 1:1,000,000. We classified the images into 9 land cover/land use types, including paddy field, dry land, forest, grassland, bare land, wetlands (e.g., river, lake, and marsh), residential land, industrial land, and mining land. We verified the interpretation results by selecting random verification points from the high-score images from Google Earth combined with field investigation. The number of random verification points was selected according to the area of each land use type. The overall classification accuracy was above 90%. Thus, we consider the interpretation accuracy to meet the demands of our study. The Projected Coordinate System and Datum of our study are Krasovsky_1940_Albers and D_Krasovsky_1940, respectively.

We obtained monthly precipitation data of 7 meteorological stations (1961–2019) across western Jilin Province through the China Meteorological Administration, Meteorological Data Center (http://data.cma.cn/site/index.html).

Calculation of landscape metrics, Hemeroby index, and mean annual precipitation

To assess landscape pattern characteristics comprehensively and to reduce information redundancy (Yang et al., 2001; Zhao, 2003), we selected class metrics including class area (CA), percentage of landscape (PLAND), number of patches (NP), patch density (PD), edge density (ED), aggregation index (AI), and perimeter area ratio-mean (PARA-MN) from a number of metrics based on land use practices (Liu et al., 2016a). We calculated wetland class metrics two times using Fragstats 4.2 for the past three decades (1986–2019) with equations (Table 1) both from holistic and regional scales (Mcgarigal & Marks, 1995; Wu, 2000).

Table 1 Equations of landscape metrics
Full size table

First, we calculated the overall landscape metrics for the study area. Second, we applied a moving window (Campagnaro et al., 2017) with a length of 30 km and started in the upper left corner of the study area, because landscape metrics changes are relatively small at the scale of 30 km (Zhang et al., 2014), which can reflect landscape pattern changes effectively for the study area. We moved a grid every time and calculated landscape metrics at a 30-km resolution, resulting in 76 grids across our study area. The reason why we adopted the scale of 30 km was that we set the size of the moving window according to the relationship between the study area and scales in previous studies (Li et al. 2014; Liu et al., 2016a; Zhang et al., 2014). We assigned landscape metrics value to the grids (Liu & Guo, 2009) and finally formed a spatial differentiation map (Ke et al., 2010) of the landscape metrics in western Jilin Province during the period (1986–2019) using Kriging interpolation (Liu et al., 2016a).

We calculated the Hemeroby index (HI) in each grid according to the HI value determined mathematically (Liu et al., 2016a) and by land cover/land use (Liang et al., 2016; Moiwo et al., 2010) (Table 2). The Hemeroby index was introduced by (Jalas, 1955) as an integrative measure of the effects of all anthropogenic disturbances on ecosystems (Celka, 2011). We used the Hemeroby index to measure the degree of anthropogenic disturbance on wetland landscape changes.

Table 2 Hierarchy of land cover/land use types with respect to HI
Full size table

We also calculated mean annual precipitation (MAP) using monthly precipitation at a scale of the grid. Here, we mainly study the relationship between mean annual precipitation and wetland landscape pattern in the semiarid region. We referred to previous studies that precipitation was the most important factor of wetland change in western Jilin Province in recent 10 years, so we did not take the mean annual temperature and annual evaporation into consideration in our study.

We used Spearman’s rank correlation coefficient to calculate the landscape metrics, Hemeroby index, and mean annual precipitation.

Spearman’s rank correlation coefficient is defined as the Pearson correlation coefficient between grade variables. For samples with sample size n, N original data are converted into hierarchical data, and the correlation coefficient ρ by:

$$\rho =\frac{{\sum }_{i}\left({x}_{i}-\overline{x }\right)\left({y}_{i}-\overline{y }\right)}{\sqrt{{\sum }_{i}{\left({x}_{i}-\overline{x }\right)}^{2}{\sum }_{i}{\left({y}_{i}-\overline{y }\right)}^{2}}}$$
(1)

Data analysis

We used overall landscape metrics to analyze wetland landscape changes for western Jilin Province during the period (1986–2019).

We used the ordinary Kriging method and assumed that the data changes were normally distributed. We interpolated CA, NP, PLAND, PD, ED, PARA_MN, AI, HI, and MAP based on the Kriging interpolation method according to their values in the 76 windows respectively to identify the spatiotemporal differences of the wetland landscape composition and configuration, anthropogenic disturbance, and mean annual precipitation.

We used the Spearman correlation method to calculate the correlation coefficient of wetland landscape metrics with the HI and MAP, respectively. Figure 2 shows the procedure of the total research methods and the corresponding flowchart.

Fig. 2
figure 2

Flowchart of the study

Full size image

Results

Wetland landscape changes in western Jilin Province

Our results showed that CA and PLAND decreased over the past few decades (1986–2019) (Table 3). NP, PD, ED, and PARA_MN decreased slightly during the first period (1986–2000) and increased remarkably during the second period (2000–2019) (Table 3). AI increased slightly then decreased dramatically at the year 2000 (Table 3).

Table 3 Dynamic changes in wetland landscape metrics from 1986 to 2019
Full size table

Change percentages of landscape metrics showed that values of landscape metrics during the first period (1986–2000) were generally lower than those during the second period (2000–2019) (Table 3). Change percentages for PARA_MN exceeded 100% from 1986 to 2019 (Table 3). Change percentages for NP, PD, and PARA_MN exceeded 100% during the second period (2000–2019) (Table 3).

Spatiotemporal differences of wetland landscape metrics, HI, and MAP

Changes in wetland landscape composition metrics (e.g., CA, NP, PLAND, and PD) were identified, as high PLAND values were found in the north of the study area during the period 1986–2000 (Fig. 3) but then declined noticeably. High PLAND values were observed in the northeast, and two low-value areas were observed in the southeast and northwest of the study area during the first period (1986–2000) (Fig. 3). Areas with low PLAND values increased and were mainly scattered around the margin of the study area by 2019. Changes in CA followed the same pattern as PLAND. The value of PD in the west was higher than that in the east. Areas with high PD values decreased in the north during the period 1986–2000. Areas with high PD values disappeared by 2000 and increased significantly by 2019 (Fig. 3). Changes in NP trends were similar to those observed for PD.

Fig. 3
figure 3

Spatiotemporal differences of a natural wetland landscape. Note: CA class area, NP number of patches, PLAND percentage of landscape, PD number of patches, ED edge density, PARA-MN perimeter area ratio-mean, AI aggregation index

Full size image

Changes in wetland configuration metrics (e.g., ED, PARA_MN, and AI) were observed; areas with high ED values were decreasing during the past few decades (1986–2019). PARA_MN in the west and east was higher than that in the central region, and the relatively high area increased minimally during the first period (1986–2000) (Fig. 3). High PARA_MN values covered the whole study area by 2019. High AI values mainly appeared in the central-eastern region, and low AI values appeared northwest and south of the study area from 1986 to 2019 (Fig. 3). Spatial patterns of high AI values at 1986 and 2000 were similar, but these areas increased distinctly by 2019 compared with the first period (1986–2000) (Fig. 3).

HI values in the middle were higher than those in the study area margins during the period 1986–2019. Relatively high HI values expanded continually during the first period (1986–2000), and the highest class of HI values became two smaller regions by 2019 (Fig. 4).

Fig. 4
figure 4

Spatiotemporal differences of Hemeroby index (HI) and mean annual precipitation (MAP)

Full size image

MAP decreased constantly, and its value in the west was lower than that in the east of the study area during the period 1986–2019. The regions of MAP lowest values increased remarkably by 2019 (Fig. 4).

Correlations of wetland landscape metrics with the HI and MAP

Results from correlation analyses showed that wetland composition metrics (e.g., NP and PD) and configuration metrics (e.g., AI and ED) were positively correlated with the HI in 2019 (p < 0.01). Wetland composition metrics (e.g., NP and PD) and configuration metrics (e.g., ED) had positive correlations with the HI (p < 0.01) during the period (1986–2019). In addition, the composition metrics of the wetland landscape (e.g., CA and PLAND) had positive correlations with the HI in 1986 and 2000 (p < 0.05) (Table 4).

Table 4 Spearman correlation analysis of landscape metrics with the HI and MAP
Full size table

Results from correlation analyses also showed that wetland landscape composition metrics (e.g., NP and PD) had negative correlations with MAP (p < 0.01) during the past few decades (1986–2019). Configuration metrics of the wetland landscape (e.g., PARA_MN and ED) were correlated with MAP in 2019.

Discussion

Wetland landscape conditions in western Jilin Province

Temporal changes in the CA and PLAND metrics suggested that wetland area decreased continually during the period 1986–2019, as with the results of a previous study that the total area showed a sharp decrease from 1950 to 2012 (Zhang et al., 2019). It was likely a result of the reclamation of a large area of wetlands to cultivation due to the household contract responsibility system that started in the early 1980s, along with rural productivity and severe overgrazing and reclamation. It was almost the rice that changed the natural wetland because of the demand for food and the abundant organic matter in the wetland that was conducive to the growth of crops (Liu et al., 2014). CA decreased, while NP and PD increased on the whole, which suggested that wetland landscapes in western Jilin Province gradually fragmented. Previous studies showed that wetland losses and fragmentation in western Jilin Province resulted in serious environmental problems in recent decades (Yu et al., 2016). Studies also suggested that wetland fragmentation was extensive due to irrational land use/land cover activities (Bian et al., 2004). Our results were consistent with previous studies (Wang et al., 2006; Yu et al., 2016). Landscape metrics could be used as a reference method to monitor changes in wetland landscapes over time.

However, Liu et al. (2014) revealed that the wetland area in western Jilin Province increased, likely because they studied wetland dynamic changes in not only natural wetland (e.g., rivers, lakes, and marshes) but also artificial wetland (e.g., paddy field) for the period 1985–2010. Artificial wetlands exerted an important effect on wetland biodiversity and human well-being (Gerba et al., 2015). Thus, investigating wetland landscape changes may need to consider artificial wetlands as well.

Spatiotemporal differences of wetland landscape metrics, HI, and MAP

High values of PLAND and CA were observed in the northeast due to the establishment of Momoge National Natural Reserve in 1981 (Jiang et al., 2007). Areas with high values of PLAND and CA declined distinctly because intensive anthropogenic disturbance and warming climate led to wetland degeneration in the Momoge National Natural Reserve (Bian et al., 2004; Chen et al., 2006). PD and NP values were higher in the west than in the east because the western region of the study area is a main food production area. High PD and NP values covered the entire area by 2019 because large areas of natural wetlands were encroached into paddy field and dry land expansion due to policies such as the building of the western granary in Jilin Province (Chen, 2008). Specifically, agricultural reclamation increased the number of wetland patches but decreased wetland area, which led to substantial changes in the PD values (Wang et al., 2011). Additionally, high values of PARA_MN were distributed across the entire area by 2019. This finding was most likely due to a lower mean annual precipitation, which caused the water table of the wetland to decrease and further shrank the wetland area. Moreover, if the shape of the wetland patch changed slightly, and the patch area decreased, PARA_MN values would increase notably. PARA_MN in the west and east was higher than that in the middle, which meant that the complexity of wetland patches on both sides of the study area was higher than that in the central region. Wetland patches across the whole study area are affected in a complex manner by anthropogenic disturbance by 2019 (Liu et al., 2014). The observation that AI values were smaller in the southwest than in the northeast suggested that wetland landscape fragmentation in the west was severe, as Chaganhu National Natural Reserve and Momoge National Natural Reserve in the northeast exerted a positive effect on protecting wetlands from anthropogenic disturbance.

Spatiotemporal differences of the HI revealed that anthropogenic disturbance in the middle was higher than on the margins and that high HI values became two areas by 2019, likely because wetlands were affected by surrounding landscapes as the land cover/land use with the largest reclamation interest (Rooney & Wilson, 2012). Land cover/land use changes showed that dry land and paddy fields increased dramatically during the period 1986–2019. Food policy has led to wetlands being converted to agricultural land (Liu et al., 2016b). Songyuan and Baicheng City in the central region of western Jilin Province are important commodity grain bases; therefore, HI values in the central region were higher than those in the study area margins. MAP was higher in the east than in the west, as the general precipitation pattern for China shows a decrease in MAP from the southeast coast to the northwest, which reflects the differential impacts of the monsoon (Chen et al., 2013; Zhao et al., 2011).

Wetland landscapes interacted with the surrounding landscapes (Rooney & Wilson, 2012; Tscharntke et al., 2010). Anthropogenic disturbance by land use/land cover changes had considerable effects on the composition and configuration metrics, which caused the lands to become homogeneous and simple (Gowda et al., 2012). Anthropogenic activities resulted in a simplified land use/land cover pattern and greater landscape homogeneity along with agricultural development. From this perspective, landscape heterogeneity should be considered as an important factor for studying wetland landscape pattern changes.

Relationships of wetland landscape metrics with the HI and MAP

The composition metrics (e.g., CA, NP, PLAND, and PD) had positive correlations with the HI from 1986 to 2000, likely because intensive agricultural activities had converted the wetlands into farmland, which not only changed the wetland distribution area but also changed the distribution extent of wetland patches (Xu et al., 2012; Yan et al., 2010). Our results that wetland spatial configuration varied with different anthropogenic disturbances was consistent with previous studies (Evans et al., 2017). This finding reconfirmed the conclusions of Sun et al. (2012), which suggested that the composition metric NP and the configuration metric ED had positive correlations with the HI and Chen et al. (2014), which demonstrated that the configuration metric ED and the composition metric PD had positive correlations with the HI. Our results also found that the HI had a positive correlation with the AI by 2019 and was not correlated with PARA_MN during the period 1986–2019, which differed from previous studies. The AI positively correlated with the HI by 2019 because wetland patch numbers increased sharply along with anthropogenic disturbance. An increase in the number of patches increases the common length of wetland patch at a grid scale, resulting in an increase in the AI at a grid scale. Specifically, the reason for this result might be that our study calculated the landscape metrics of natural wetlands and analyzed the impact of anthropogenic disturbance on the landscape metrics at different periods, whereas the previous studies used overall landscape metrics and analyzed the relationship between anthropogenic disturbance and landscape metrics at one period (Liu et al., 2016a). Thus, the effects of anthropogenic disturbance on wetland landscape metric had periodic and regional characteristics. Additionally, previous studies mainly adopted buffer analysis to study the relationship between human activities and landscape patterns (Lü et al., 2004; Zeng et al., 2001), whereas only the correlation analysis method was used to reveal the landscape patterns affected by anthropogenic disturbances, which may be uncertain. Our findings are nonetheless consistent with previous studies that anthropogenic disturbance effects were mainly reflected by the change in landscape structure using buffer analysis (Zeng et al., 2001).

Results from correlation analyses indicated that MAP had negative correlations with the composition metrics (e.g., NP and PD) and with configuration metrics (e.g., ED), likely because sufficient precipitation could change the wetland patch numbers through altering the water table of wetlands located in semiarid regions (Hess et al., 2003; Watts et al., 2015). This result differed from some previous studies that reported wetland landscape metrics had no significant correlation with precipitation (Li et al., 2009). It is likely that this result arose because our study was located in a semiarid region and the wetland landscape was heavily dependent on MAP (Moreno et al., 2007), in contrast to patterns observed in regions with abundant water supplies. Water supplies may alter the wetland proportional abundance that caused the changes in wetland composition metrics (Evans et al., 2017). Thus, combined with the correlation between landscape metrics and anthropogenic disturbance, we should take into account the impact of anthropogenic disturbance on the shape and fragmentation of wetland landscape when formulating wetland protection policies.

Implications of the study findings and limitations

This methodology may be good for monitoring and assessing wetland complexities and distributions and could be an additional approach to guiding wetland restoration projects, specifically, by prioritizing areas on the landscape where anthropogenic disturbances are causing the greatest impact. Our study has significant implications for wetland conservation and management and anthropogenic food policies and planning of western Jilin Province. Our findings were subject to a number of limitations. Initially, we did not consider that some projects, including a water diversion project, a project connecting rivers and lakes, and wetland restoration projects, might affect wetland landscapes and areas (Zang et al., 2014; Zhang et al., 2017). Second, some judgments on how to discriminate disturbance types and quantify the HI were subjective, based on expert knowledge and experience (Holland et al., 2016). Thus, specific indices based on objective characteristics are needed to better reflect the effects of human activities. In addition, we did not take the edge effect (based on the project extent) into account. Differences in scales between variables (e.g., DEM and soil data) limited the image interpretation accuracy to a certain extent. Nevertheless, our findings were generally consistent with previous studies and will help facilitate the monitoring and assessment of wetland landscape fragmentation in the future.

Conclusions

In conclusion, the wetland landscape has become fragmented in western Jilin Province during the past period (1986–2019). Anthropogenic disturbance had a positive impact on wetland landscape composition metrics (e.g., NP and PD) and configuration metrics (e.g., ED and AI) in 2019. Mean annual precipitation had a negative impact on wetland landscape composition metrics (e.g., NP and PD) and configuration metrics (e.g., ED) by changing the wetland water source during 1986–2019. Our study has significant implications for monitoring and assessing wetland complexities and distributions and guiding wetland restoration projects.