Introduction

Ecosystem services (ESs) refer to various ecological, social and economic benefits and values that humans directly or indirectly obtain from ecosystems (Daily 1997; Costanza et al. 1998; MEA 2005). In recent decades, the stability and function of ecosystems have been affected by human activities, posing a threat to the sustainable supply of ESs (Foley et al. 2005; Turner et al. 2007; Wang et al. 2022a). Therefore, it is imperative to restore and strengthen the effective management of ecosystems. Relatedly, landscape patterns and their spatial heterogeneity are considered crucial factors because they directly affect the type, quantity, and distribution of ESs (Lovett et al. 2005; Turner et al. 2013; Liu et al. 2018a). At present, many research results have been obtained on the influence of landscape patterns on individual ESs. For example, the influence of landscape pattern change on water-related ESs had significant category differences, annual variations and spatial differences (Qiu and Turner 2015; Li et al. 2021; Xia et al. 2021); in urban landscapes, forestland had a significant promoting effect on regulation services and support services, while construction land mainly had a negative effect (Wang et al. 2019); and compared with landscape configuration, the complexity of landscape composition had a stronger effect on crop yield (Nelson and Burchfield 2021). The above results show that landscape patterns mainly affect the supply of ESs by changing ecosystem structure and ecological processes.

However, there are complex and variable interactions and interdependence among ESs, which pose challenges to resource allocation and management decisions (Bennett et al. 2009; Howe et al. 2014). The relationships among ESs are usually divided into trade-offs, synergies and nonrelated. When the increase in one ES is accompanied by a decrease in another ES, it is called a trade-off; when two ESs simultaneously increase or decrease, it is called synergy; other relationships are nonrelated or are compatible (Li et al. 2013; Qiu and Turner 2013; Lee and Lautenbach 2016). Studies have shown that landscape patterns not only have impacts on individual ESs but also play an important role in the relationships among ESs (Rieb and Bennett 2020; Qiu et al. 2021; Lyu et al. 2022). For example, in urban landscapes, low connectivity is closely associated with a large trade-off between nutrient retention and carbon storage, as well as a large trade-off between habitat quality and pollinator abundance (Karimi et al. 2021); landscape fragmentation and land use intensity have a positive impact on the synergy between pollination and nitrogen retention (Lavorel et al. 2022). These results suggest that changes in ecosystem structure and ecological processes caused by landscape patterns may simultaneously affect multiple ESs, leading to trade-offs and synergies among ESs. Therefore, exploring the spatial–temporal heterogeneity in the influence of landscape patterns on trade-offs/synergies among ESs is helpful for understanding the generation and transmission processes of ESs, which is of great significance for multiple ES management.

Nevertheless, most of the above studies used regression analysis, correlation analysis and other global statistical methods to quantify the relationship between variables. The global model presupposes that the relationship between variables is homogeneous, ignoring the potential spatial nonstationarity, which makes it difficult to accurately evaluate the relationship between variables in local areas. The multiscale geographically weighted regression (MGWR) model can be used to determine the spatial heterogeneity in a relationship between variables and provide more scale-related information about different variables (Oshan et al. 2019). Therefore, this study introduced the MGWR model to determine the heterogeneous relationships between landscape patterns and ES trade-offs/synergies.

The Loess Plateau of northern Shaanxi (LPNS) is in the upper reaches of the Yellow River, which is a typical ecotone with water shortages, serious soil erosion, low vegetation coverage and prominent man-land interactions. Since the implementation of the Grain for Green Project (GFG) in 1999, the landscape pattern in the area has changed dramatically, and the ecological environment and the potential supply capacity of ESs have improved to some extent (Bao et al. 2016; He et al. 2023; Wang et al. 2023). However, the changes may also have led to significant changes in trade-offs and synergies among ESs. For example, Wang et al. (2022c) and Wu et al. (2022) showed that the number and intensity of trade-offs among ESs in northern Shaanxi increased from 2000 to 2015, and the synergy relationships decreased or even reversed into trade-offs during that period. Some scholars have also conducted in-depth exploration on the driving mechanism of ES trade-offs/synergies in the LPNS, and found that cultivated land, forest land and grassland in land use/land cover (LULC) are the main factors influencing of ES trade-offs/synergies. The above studies enriched the theoretical results on the driving mechanism of ES trade-offs/synergies in the LPNS and revealed the global relationship between ES trade-offs/synergies and drivers from a static perspective, but failed to further identify the differences in the spatial and temporal distribution of this relationship.

The main aims of this research were (1) to quantify the change characteristics of the LPNS landscape pattern from 2000 to 2020; (2) to determine the spatial heterogeneity of trade-offs/synergies among various ESs in the LPNS from 2000 to 2020; and (3) to explore the spatial–temporal heterogeneity of the influence of landscape patterns on trade-offs/synergies. The results of this study can provide information for managers implementing spatially targeted ecological restoration measures in the future.

Materials and methods

Study area

The LPNS is in the middle of the Loess Plateau in China and the northern part of Shaanxi Province (107°30′ ~ 111°15′E, 34°10′ ~ 39°35′N), including the cities of Yulin and Yan’an, with a total area of 79967 km2 (Fig. 1). The terrain is high in the west and low in the east, and the area has a temperate arid and semiarid monsoon climate. The average annual precipitation is 400 ~ 600 mm, increasing from the northwest to the southeast. The average annual evaporation is 900 ~ 1200 mm, decreasing from north to south. The landforms are complex and diverse, with the northwestern part dominated by the wind-sand grass shoal area and the central and southern parts dominated by the loess hilly-gully area. Most of the study area is covered by loess, with low vegetation coverage, a vulnerable ecology, serious land degradation and soil erosion (Liu et al. 2020).

Fig. 1
figure 1

The Loess Plateau of Northern Shaanxi

Full size image

The LULC of the LPNS is dominated by cultivated land, forestland and grassland, the proportion of which has changed significantly since the implementation of the GFG in 1999 (Fig. 2). From 2000 to 2010, the scale of the GFG was large, the proportion of cultivated land decreased from 35.6 to 31.7%, and the area decreased by 3158 km2; the proportion of forestland increased from 13.8 to 15.3%, and the area increased by 1194 km2; the proportion of grassland increased from 43.7 to 46.0%, and the area increased by 1889 km2. From 2010 to 2020, the scale of the GFG gradually decreased, and the range of LULC change decreased. The proportions of cultivated land, forestland and grassland showed a slight decline with the advancement of urbanization, and the change fluctuated between 0.1% and 0.3%. However, the development of forestland and grassland in abandoned lands is becoming increasingly mature, and ecological benefits are gradually emerging. Given this context, the changes in landscape patterns and ESs in the study area provided excellent conditions for exploring the spatial heterogeneity relationships between landscape patterns and ES trade-offs/synergies to provide reasonable suggestions for the sustainable development of arid and semiarid ecotones.

Fig. 2
figure 2

Land Use/Land Cover for the Loess Plateau of Northern Shaanxi

Full size image

Data sources and processing

Based on the LULC data, topographic data, meteorological data, soil data, vegetation index and grain yield data in 2000, 2010 and 2020, the changes in typical ESs were evaluated to explore the spatial heterogeneity relationships between landscape patterns and ES trade-offs/synergies. All data were converted to the same spatial coordinate system and spatial resolution (1 km × 1 km) for the calculation of ESs. More details about the multisource data are shown in Table 1.

Table 1 Details and sources for the datasets used in this study
Full size table

Methods

Quantification of landscape patterns

Landscape metrics are important methods for quantifying landscape patterns. In this study, three levels of metrics were selected: shape index (SHAPE) and patch density (PD) at the patch level; percentage of landscape (PLAND) at the class level (PLAND only includes the three largest LULC types in the study area: cultivated land, forestland and grassland); and largest patch index (LPI), patch richness (PR), contagion index (CONTAG), connectance index (CONNECT) and interspersion and juxtaposition index (IJI) at the landscape level. The selection of these metrics mainly considered the following four factors: (1) starting from the integrity of landscape patterns, such as landscape diversity, heterogeneity and fragmentation (Su and Fu 2012; Tanner and Fuhlendorf 2018; Zhang et al. 2020a); (2) referring to the metrics used frequently in the literature (Lamy et al. 2016; Badora and Wróbel 2020; Chen et al. 2021; Ran et al. 2023); (3) priority should be given to metrics that are easy to understand and calculate to improve the repeatability of research; and (4) variance inflation factor (VIF) < 10 screening to reduce multicollinearity. All landscape metrics were calculated using FRAGSTATS v4.2 (McGarigal et al. 2012) with a spatial resolution of 1 km.

Ecosystem services assessment

Combined with the actual situation of water shortage, severe soil erosion and a vulnerable ecological environment in the study area and the availability of data, five typical ESs, water retention (WR), food supply (FS), habitat quality (HQ), soil retention (SR) and landscape aesthetics (LA), at a 1 km spatial resolution were calculated. The specific calculation methods are shown in Table 2 and 3.

Table 2 Methods and process for quantifying ecosystem services
Full size table
Table 3 Assessing indices of landscape aesthetics
Full size table

Quantitative measure of trade-offs and synergies among ESs

The trade-off/synergy degree (TSD) indicator can reflect the spatial pattern of the trade-offs/synergies among ESs (Zhao and Li 2022). Given this information, the normalization process was used to revise it to prevent meaningless samples caused by its original algorithm and simultaneously unify its value range from -1 to 1 to enhance the comparability of different trade-offs/synergies. It can be described as follows:

$${TSD}_{i-j}=\left\{\begin{array}{c}0 , ({\Delta ES}_{i,t2-t1}\times {\Delta ES}_{j,t2-t1}=0)\\ \sqrt{({{\Delta ES}_{i,t2-t1}}^{2}+{{\Delta ES}_{j,t2-t1}}^{2})/2}, ({\Delta ES}_{i,t2-t1}\times {\Delta ES}_{j,t2-t1}>0)\\ -\sqrt{({{\Delta ES}_{i,t2-t1}}^{2}+{{\Delta ES}_{j,t2-t1}}^{2})/2}, ({\Delta ES}_{i,t2-t1}\times {\Delta ES}_{j,t2-t1}<0)\end{array}\right.$$
(1)

where \({\Delta ES}_{i,t2-t1}\) and \({\Delta ES}_{j,t2-t1}\) refer to the variation in ES values for types \(i\) and \(j\) from \(t1\) to \(t2\), respectively, and are calculated as follows:

$${\Delta ES}_{t2-t1}={ES}_{t2}{\prime}-{ES}_{t1}{\prime}$$
(2)

where \({ES}_{t1}{\prime}\) and \({ES}_{t2}{\prime}\) represent the normalized values of \(ES\) at \(t1\) and \(t2\), respectively. To reflect the trend of ESs over time, the normalization process was performed simultaneously for the data from three periods, and the mathematical expression is as follows:

$${ES}_{t1}{\prime}=\frac{{ES}_{t1}-{\text{min}}({ES}_{t1},{ES}_{t2})}{{\text{max}}\left({ES}_{t1},{ES}_{t2}\right)-{\text{min}}({ES}_{t1},{ES}_{t2})}$$
(3)

where \({ES}_{t1}\) and \({ES}_{t2}\) represent the observed values of \(ES\) at \(t1\) and \(t2\), respectively. As shown in Eq. (1), when \({TSD}_{i-j}=0\), there is nonrelated between \({ES}_{i}\) and \({ES}_{j}\); when \({TSD}_{i-j}>0\), there is synergy between \({ES}_{i}\) and \({ES}_{j}\); and when \({TSD}_{i-j}<0\), there is a trade-off between \({ES}_{i}\) and \({ES}_{j}\). The absolute value of \({TSD}_{i-j}\) reflects the strength of the trade-offs/synergies.

Before exploring the spatial heterogeneity relationships between landscape patterns and ES trade-offs/synergies, it is necessary to use global Moran’s I to detect the spatial dependence of trade-offs/synergies, and its value ranges from -1 to 1. After the significance test, Moran’s I > 0 indicates that the trade-off/synergy is positively correlated in space; Moran’s I < 0 indicates that the trade-off/synergy is negatively correlated in space; and Moran’s I = 0 indicates that the trade-off/synergy is randomly distributed in space. The greater the absolute value of Moran's I is, the stronger the correlation.

Identifying relationships between landscape patterns and ES trade-offs/synergies

Multiscale geographically weighted regression is an extension of geographically weighted regression (GWR), which eliminates the single bandwidth assumption and allows different processes to operate on different spatial scales. It is effective in estimating parameter surfaces with different levels of spatial heterogeneity (Fotheringham et al. 2017). The MGWR model can be presented as follows:

$${y}_{i}=\sum_{j=0}^{m}{\beta }_{{bw}_{j}}\left({u}_{i},{v}_{i}\right){x}_{ij}+{\varepsilon }_{i}$$
(4)

where \({y}_{i}\) indicates the estimated TSD value at sample \(i\); \({\beta }_{{bw}_{j}}\) indicates the regression coefficient of independent variable \(j\); \({bw}_{j}\) indicates the bandwidth of independent variable \(j\); \(\left({u}_{i},{v}_{i}\right)\) indicates the spatial coordinate of sample \(i\); \({x}_{ij}\) indicates the value of independent variable \(j\) at sample \(i\); and \({\varepsilon }_{i}\) indicates the random error term for sample \(i\).

Based on the actual performance of the MGWR model, the spatial resolution of the dependent variables (TSD) and the independent variables (landscape metric variation) involved in the calculation in this study was 3 km.

Results

Landscape pattern variation

Due to the implementation of the GFG, the LPNS experienced a decrease in cultivated land and an increase in forestland and grassland from 2000 to 2020, resulting in significant changes in the landscape pattern (Fig. 3), mainly in terms of increased landscape diversity and heterogeneity and reduced landscape complexity and fragmentation.

Fig. 3
figure 3

Spatial patterns and changes in landscape metrics in 2000—2010 and 2010—2020. (PLAND_C: percentage of cultivated land, PLAND_F: percentage of forestland, PLAND_G: percentage of grassland)

Full size image

Specifically, the changes in 2000—2010 were significant and mainly concentrated in the central part of the loess hilly-gully area. LPI and CONTAG showed an upward trend, while SHAPE, PD and CONNECT showed a downward trend, indicating that the dominant patches gradually expanded; in addition, the small and complex patches coalesced into relatively isolated and simple-shaped large patches, thus improving the fragmentation of the landscape. However, these results also caused some obstacles to the connectivity between landscapes. The increase in PR and IJI indicated that the landscape types in the area were enriched and more evenly distributed spatially, the landscape was diversified, and the heterogeneity was weakened. At the class level, cultivated land was concentrated from the northeastern and central southern areas to the central northern, western and southern areas, while forestland and grassland expanded significantly.

From 2010 to 2020, the change range of most areas in the study area was small and evenly distributed. In the northwestern area, PR, PD, and IJI showed an upward trend, while LPI, PLAND_G, SHAPE and CONTAG showed a downward trend. In essence, the erosion of grassland through the expansion of construction land (Fig. 1) led to fragmentation of the landscape while enhancing the diversity of the landscape and reducing heterogeneity. In addition, SHAPE in the southeastern mountains increased significantly, indicating that the reduction in the intensity of human activities caused by the GFG had a positive impact on the restoration of the ecological environment.

Ecosystem services variation

From 2000 to 2020, WR, FS, HQ, SR, and LA all showed significant upward trends in the study area. There was significant spatial–temporal heterogeneity in the trends during the first 10 years and the second 10 years (Fig. 4). Among the ESs, FS was mainly related to the distribution of cultivated land, while other ESs had a strong spatial correlation with forestland.

Fig. 4
figure 4

Spatial patterns and changes in the ecosystem services in 2000—2010 and 2010—2020. (FS: food supply, WR: water retention, SR: soil retention, HQ: habitat quality, LA: landscape aesthetics)

Full size image

As shown in Fig. 4, from 2000 to 2010, the implementation of the GFG and the increase in rainfall in the study area caused the WR in the southern mountainous area to have a significant upward trend, while the northwestern area also increased slightly. Due to the change in the proportion of cultivated land (Fig. 3), the high value areas of FS (92 ~ 459) were concentrated in the northwestern and central northern areas. The increase in forestland directly promoted the increase in HQ in the central and northeastern areas, but the spatial distribution was more dispersed; however, most areas showed a downward trend. The rising trend of SR was obvious, and the high value area (2.9 ~ 14.6) was mainly concentrated in the southeastern border area; the western loess tableland area decreased slightly. The LA in the study area increased significantly overall, while the declining trend in the southeastern mountainous areas was more obvious.

From 2010 to 2020, the GFG entered the advance stage, and the ecological benefits were gradually reflected. Due to the management of some unused land in the northwestern area, the vegetation was repaired to a certain extent, and the increase in WR strengthened. With the increase in agricultural demand and productivity, the increasing trend of FS was more obvious, and the high-value area expanded to the eastern area. At the same time, the central and southern areas were affected by the GFG, showing a certain degree of downward trend. The forestland and grassland in the central, western and northeastern areas gradually matured, which greatly enhanced the soil retention capacity. At the same time, due to the influence of altitude, SR increased significantly in the ridge zone from the central to the western area. The overall downward trend of HQ and LA increased, and the upward trend in the central area weakened.

Spatial–temporal patterns of trade-offs/synergies among ESs

A geographic grid was used to calculate the TSD at a 3 km spatial resolution to explain the spatial distribution and intensity of the trade-offs/synergies among the ESs, and the dominant relationship was determined based on the proportion of trade-off units and synergy units.

In general, the distribution of trade-offs/synergies among the ESs showed different degrees of heterogeneity in time and space (Fig. 5). The trend over time showed consistency; that is, the proportion of trade-off units showed an upward trend, and the proportion of synergy units showed a downward trend (Fig. 6). Specifically, the interactions between WR and other ESs were weak, and their trade-off and synergy units accounted for less than 50%, mainly based on synergies, concentrated in the northwestern wind-sand grass shoal area and the southern mountainous area. However, the spatial patterns of the trade-offs/synergies between FS and SR, HQ, and LA were opposite. From 2000 to 2010, FS—HQ was dominated by trade-offs, while FS—SR and FS—LA were dominated by synergies, mainly concentrated in the central and northern areas; during 2010—2020, the trade-offs/synergies in the eastern area were enhanced because of the significant increase in FS, and the synergistic relationship of FS—SR on the ridge zone was enhanced at the same time. In 2000—2010, HQ—SR and HQ—LA were dominated by synergies, mainly concentrated in the forestland in the central and southern areas. However, in 2010—2020, with the decreasing trend of HQ in the area becoming increasingly obvious, the synergistic relationship changed to a trade-off relationship, and the dominant relationships of HQ—SR and HQ—LA reversed. From 2000 to 2010, SR—LA was dominated by synergy, and the western and southern areas showed a trade-off trend. From 2010 to 2020, the degree of synergy in the central western area and the eastern border area increased with increasing LA, but the trade-off relationship in the northwestern area expanded due to the decline in LA.

Fig. 5
figure 5

Spatial patterns of the trade-offs/synergies among the ecosystem services in 2000—2010 and 2010—2020

Full size image
Fig. 6
figure 6

Proportion of the trade-offs/synergies among the ecosystem services in 2000—2010 and 2010—2020

Full size image

Table 4 shows that there were positive spatial autocorrelations in the relationships among the five ESs, and the autocorrelations showed a decreasing trend with time. Among them, SR—LA had the highest Moran’s I and the strongest spatial autocorrelation, while WR—FS had the lowest Moran's I and the weakest spatial autocorrelation.

Table 4 Spatial patterns of relationships among the ecosystem services
Full size table

Spatial–temporal heterogeneity of relationships between landscape patterns and ES trade-offs/synergies

Based on the relationships among the ESs, this study selected three representative ES pairs, namely, the trade-off-based FS—HQ, the synergy-based SR—LA, and the nonrelated-based WR—FS, to explore the heterogeneous relationships between landscape metrics and trade-offs/synergies through the MGWR model. According to the two parameters (Adj R2 and corrected Akaike information criterion (AICc)) provided by the model, the fitting effect of the model was characterized. The larger the Adj R2 and the smaller the AICc were, the better the fitting effect of the model. Table 5 shows that the landscape metrics had the best fitting effect with SR—LA, while the fitting effect with WR—FS was poor. Figure 7 shows that the absolute value of the MGWR standard residual \(\left|Residual\right|\) <2.5 in most areas of the study area, indicating that the results of the model were robust (Zhang et al. 2020b; Xue et al. 2022a).

Table 5 Fitting performance of the multiscale geographically weighted regression model
Full size table
Fig. 7
figure 7

Spatial distribution of the standard residuals

Full size image

The spatial nonstationary relationships between variables were described by the regression coefficients; that is, the positive regression coefficient and the increase in the independent variable or the negative regression coefficient and the decrease in the independent variable promoted the synergy between ESs, while the positive regression coefficient and the decrease in the independent variable or the negative regression coefficient and the increase in the independent variable promoted the trade-off between ESs. According to the results of the MGWR model, when the absolute value of the regression coefficient was large (\(\left|\beta \right|\) >0.1), the influence of the independent variable on the relationship between ESs was obvious. Figure 8 shows that there was spatial–temporal heterogeneity in the impacts of landscape patterns on trade-offs/synergies among ESs; that is, the same landscape metric had different influence directions and intensities on trade-offs/synergies in different periods and areas.

Fig. 8
figure 8

Quantitative effects of landscape metrics on depicting trade-offs and synergies through the multiscale geographically weighted regression model

Full size image

Specifically, from 2000 to 2010, WR—FS was dominated by the nonrelated factors and mainly affected by landscape metrics such as LPI, PLAND_C and PLAND_F, among which PLAND_C had the most significant impact. Both LPI and PLAND_C had positive effects, while PLAND_F had both positive and negative effects and showed certain spatial heterogeneity (Fig. 8a1, b1 and c1). FS—HQ was dominated by trade-offs and mainly affected by PLAND_C, PLAND_F, PLAND_G, PR and CONTAG. Among them, PLAND_F and PLAND_G had a greater impact (Fig. 8c2 and d2), and positive and negative effects coexisted, with strong spatial heterogeneity. SR—LA was dominated by synergy, which was mainly affected by PLAND_C, PR and SHAPE. Among them, PLAND_C had the strongest impact, and the positive and negative effects coexisted with its effects on SR—LA occurred at a smaller spatial scale, showing strong spatial heterogeneity (Fig. 8b3); the influence of PR and SHAPE was stronger (Fig. 8e3 and f3), with the former mainly having a positive influence and the latter mainly having a negative influence, and the spatial heterogeneity was weak.

From 2010 to 2020, WR—FS was mainly nonrelated and was affected by PLAND_C and PLAND_G (Fig. 8b4 and d4), positive and negative effects coexisted, the degree of influence was strong, and both showed strong spatial heterogeneity. The trade-off relationship between FS and HQ was enhanced, which was mainly affected by PLAND_C, PLAND_F and PLAND_G, and the positive and negative effects coexisted. Among them, the influence degree and spatial heterogeneity of PLAND_C were strongest (Fig. 8b5), and those of PLAND_F and PLAND_G were weakest (Fig. 8c5 and d5). The synergistic relationship between SR and LA weakened, which was mainly affected by LPI, PLAND_C and PR, and the degree of influence was weak. Among them, the positive and negative effects of LPI and PLAND_C coexisted (Fig. 8a6 and b6), and there was a certain spatial heterogeneity; PR was mainly positive (Fig. 8e6), and the spatial heterogeneity was weak.

Overall, the heterogeneity in the influence of landscape pattern on trade-offs/synergies among the ESs was mostly attributed to the different directions of changes in landscape metrics, which led to different forms of trade-offs/synergies. For example, from 2000 to 2010, PLAND_F in the central and northern parts of the study area had negative impacts on WR—FS and FS—HQ, while it showed a positive impact in the western area (Fig. 8c1 and c2). This finding might have been due to the increase in PLAND_F in the central and northern areas, which led to the decrease in FS and the increase in WR and HQ; a slight decrease in PLAND_F in the western area promoted an increase in FS and a decrease in WR and HQ (Fig. 4). Therefore, the increase and decrease in PLAND_F in different areas led to different types of trade-offs between WR—FS and FS—HQ. In this process, the relationships between landscape patterns and single ESs did not change. However, under the trade-off/synergy, the heterogeneity in the influence of landscape metrics on individual ESs might have been obscured by homogeneous regression coefficients. For example, from 2000 to 2010, the regression coefficients in the northwestern and southern areas were positive (Fig. 8b1 and b2), but the northwestern area was dominated by unused land. The increase in PLAND_C increased not only crop yield but also vegetation coverage to a certain extent, which had positive impacts on WR and HQ. The southern area was dominated by forestland and grassland, and the change in PLAND_C had negative impacts on WR and HQ. This interregional difference in impacts might have depended more on the difference in the landscape matrices.

Discussion

Relationships between landscape pattern and ES trade-offs/synergies

Based on the dominant relationship among the ESs, three groups of ES pairs were selected: nonrelated dominant (WR—FS), trade-off dominant (FS—HQ), and synergy dominant (SR—LA). The MGWR model revealed the complex spatial–temporal relationships between landscape patterns and ES pairs, providing a more detailed explanation. From the perspective of model parameters, the fitting effect of the model seemed to be related to spatial autocorrelation (Table 5). The larger Moran’s I was, the larger the Adj R2. This finding may indicate that the local regression model had a better fitting effect on variables with stronger spatial autocorrelation. Although the existing research results support this conclusion (Zhang et al. 2020b), more relevant studies need to be compared to verify it.

There was strong spatial–temporal heterogeneity in the response of the three group relationships to the landscape pattern. In 1999, the large-scale implementation of the GFG began in the LPNS. From 2000 to 2010, the LULC changed significantly, mainly in terms of the decrease in PLAND_C and the increase in PLAND_F and PLAND_G (Fig. 3). These changes led to an increase in vegetation coverage; enhanced the ability to intercept rainfall, block surface runoff and inhibit soil evaporation; and weakened the erosion from rainfall on the surface, thereby reducing soil erosion and promoting the trade-offs of WR—FS and FS—HQ and the positive synergy of SR—LA (Fig. 8b, c and d). From 2010 to 2020, the implementation rate of the GFG slowed significantly. The change in LULC was mainly observed in the slight increase and decrease in cultivated land and grassland at a smaller spatial scale (Fig. 3), and the change was widespread over a wide range of areas. Therefore, PLAND_C and PLAND_G had complex and significant effects on the relationships among ESs and showed strong spatial heterogeneity. The complexity may have been caused by many factors. For example, FS is usually closely related to the area of cultivated land. However, for ecotones, appropriate natural landscapes could increase biodiversity and species abundance, thereby increasing crop yield (Seppelt et al. 2016). Therefore, PLAND_G also had a certain role in promoting FS. Compared with that of grasslands, the soil reinforcement of crops is weaker, and the increase in PLAND_C is often accompanied by a decrease in SR. However, in some areas of the study area, sloping cultivated land was transformed into terraces through engineering measures (Xue et al. 2011), which reduced the slope of cultivated land and weakened the erosion effect of rainfall and runoff from soil (Arnáez et al. 2015), thus promoting the increase in SR.

At present, the ecological environment governance in the northwestern part of the study area is an urgent problem that needs to be addressed. Greater vegetation coverage can effectively slow the movement of surface runoff and increase the infiltration of water flow. Thus, expanding the area of forestland may be the most direct and effective solution. However, based on the strong evapotranspiration in this area, it must be considered that a greater canopy of forestland would result in stronger plant transpiration (Sun et al. 2006; Xue et al. 2022b). In contrast, the transpiration of grasslands is weaker, which can better enable water conservation and soil erosion reduction. Therefore, on the basis of ensuring a red line of cultivated land, appropriately expanding the proportion of grasslands in the landscape and adjusting the spatial configuration of cultivated land and grasslands may be feasible measures for promoting the coordinated development of food security and the ecological environment in the study area.

Difference in the influence between landscape composition and landscape configuration

Studies have shown that trade-offs/synergies among ESs may vary depending on landscape composition (Qiu et al. 2021) and landscape configuration (Rieb and Bennett 2020). Therefore, this study considered the differences between the two in the selection of landscape metrics. LPI, PLAND_C, PLAND_F, PLAND_G and PR were used to characterize landscape composition, and SHAPE, PD, CONTAG, CONNECT, and IJI were used to characterize landscape configuration. Comparing the regression coefficients, it was found that the influence of landscape composition on the interaction among Ess was significantly greater than that of landscape configuration, which was consistent with the results of existing studies (Lamy et al. 2016; Lavorel et al. 2022; Shi et al 2023).

In this paper, the distribution characteristics of the regression coefficients between landscape metrics and trade-offs/synergies were statistically presented in box plots, which visually reflected the differences in the effects of landscape composition and landscape configuration. Overall, compared with landscape configuration, the quartiles and extreme values of the regression coefficients corresponding to landscape composition showed a larger distribution range (Fig. 9). From the perspective of time, during 2000—2020, the impact of landscape composition on the nonrelated-dominated WR—FS and trade-off-dominated FS—HQ increased with time (Fig. 9a, b, d and e), and the impact on synergy-dominated SR—LA decreased over time (Fig. 9c and f). Among them, WR—FS and FS—HQ were mainly affected by PLAND_C, PLAND_F and PLAND_G, and SR—LA was more affected by PLAND_C and PR. Comparing the two periods, the impact of landscape configuration on the relationships among Ess was stronger in 2000—2010. CONTAG had a weak effect on FS—HQ and SR—LA, WR—FS had a slight response to SHAPE and IJI, and SHAPE had the most significant effect on SR—LA. The relationships among the Ess in 2010—2020 were not sensitive to changes in landscape configuration.

Fig. 9
figure 9

Box plots for regression coefficients. Notes: The landscape composition and configuration variables are marked in red and blue, respectively

Full size image

Overall, PLAND_C and PLAND_G from 2000 to 2020 had a greater impact on the three groups of relationships among Ess. This might have been due to the large proportion and changes in cultivated land, forestland and grassland in the LPNS, which had an important contribution to the change in the regional landscape pattern. As important landscape matrices, their spatial–temporal changes changed not only the size of the supply units of relevant Ess but also the transmission of ecological processes between landscapes through changes in landscape configuration, thus affecting the generation and Interactions of the eSs.

Conclusions

This study quantified the landscape pattern and five typical eSs (WR, FS, HQ, SR, and LA) in the LPNS and used the revised TSD indicator to map patterns of trade-offs and synergies among the eSs. The MGWR model revealed the spatial–temporal heterogeneity in the influence of landscape patterns on the trade-offs/synergies.

The results showed that the LULC change in the LPNS was significant from 2000 to 2010, mainly observed as a decrease in cultivated land and an increase in forestland and grassland, which led to the enhancement of landscape diversity and the weakening of landscape heterogeneity and fragmentation. HQ and LA increased significantly. From 2010 to 2020, the change range decreased, the spatial distribution was more uniform, and the landscape diversity and fragmentation in the northwestern area increased significantly. WR, FS and SR increased significantly. From 2000 to 2020, the trade-offs/synergies related to WR were weak. FS—SR, FS—LA and SR—LA were dominated by synergies. The proportion of trade-off units in all relationships had an upward trend in 2010—2020, and HQ—FS, HQ—SR and HQ—LA were dominated by trade-offs. Landscape patterns had complex impacts on trade-offs/synergies, and the same landscape variable might have the opposite impact on specific trade-offs/synergies in different periods and areas. The results of this study will inform managers in developing regional sustainable ES management strategies while advocating for more research to address ecological issues from a spatial–temporal perspective.