Estimation of the Relationship Between Urban Landscape Pattern and Crop Yield by Remote Sensing Data and Field Measurement

Highlights

What are the main findings?

• The suburban crop yield decreased consistently with increasing ISA and decreasing forest coverage.
• The driving mechanisms of landscape patterns and diversity on crop yield differed across urbanization intensities.

What are the implications of the main findings?

• That crop yield was affected by urban landscape patterns and diversity.
• Our study is critical for landscape optimization to increase food production under rapid urbanization.

Abstract

Exploring how urban landscape patterns and diversity affect crop yields is critical for landscape optimization to increase food production under rapid urbanization. In this study, we used Landsat remote sensing data combined with field-measured crop yields to map the spatial distribution of yields in suburban Changchun, Northeast China, and to examine their relationships with urban landscape patterns and diversity indices. Our results showed that the urban landscape composition, such as impervious surface areas (ISA) or forest coverage, significantly affected crop yield, and the suburban crop yield decreased consistently with increasing impervious surface and decreasing forest coverage (p < 0.001). Additionally, crop yield exhibited a nonlinear increase as impervious surface edge density (ED_ISA) decreased, with a threshold identified at 200 m/ha. We also identified that the driving mechanisms of landscape patterns and diversity on crop yield varied across different levels of urbanization intensities. In the low-urbanization area (ISA coverage < 50%), the crop yield was mainly affected by the composition and pattern of the surrounding landscape, such as ISA or forest coverage, patch and edge density, and the largest patch index; In the medium-urbanization area (50% ≤ ISA coverage ≤ 80%), landscape diversity played a dominant role and had a strong positive effect on crop yield. In the heavy-urbanization area (ISA coverage > 80%), crop yield was mainly affected by indicators of the farmland itself, such as coverage, edge density, and the largest cropland patch index. These findings clarify the relationship between urban landscapes and crop yields, offering new insights into reconciling urban development with food security.

1. Introduction

The world is undergoing rapid urbanization [1,2], and more than half of the global population currently lives in urban areas. The global urbanization currently exceeds 50% and is projected to reach 68% by 2050 [3], which will have a profound impact on food production and security [4]. Urban expansion has led to the conversion of high-quality cropland surrounding cities into built-up areas, resulting in a substantial reduction in arable land [5]. Between 1986 and 2003, urban expansion in China occupied over 33,400 km² of previous cropland, representing 21% of the nation’s total cropland loss [6]. The process of urban expansion is characterized by the continual subdivision and encroachment of cultivated land [7]. This process causes cropland fragmentation, reduces landscape connectivity, and intensifies edge effects [8]. Furthermore, rapid urbanization has resulted in significant alterations in the landscape patterns surrounding agricultural land. Numerous impervious surface areas (ISA) have emerged around croplands, and green spaces adjacent to agricultural land have also been subdivided into smaller parcels, thereby increasing ecosystem fragmentation during rapid urbanization [5,9].

Grain production is one of the most important foundations of agroecosystem services [10]. Identifying the factors that influence crop yield is critical for ensuring the stability and security of food production systems [11]. Currently, scientists are increasingly considering the effects of natural and anthropogenic factors, such as climate, soil properties, and agricultural practices on crop yield. Temperature [12], precipitation [13], surface ozone concentration [14], soil type [15], fertilization [16], and tillage methods [17] are key factors that influence crop yield. Beyond these fundamental drivers, landscape patterns—particularly their composition and configuration—are increasingly recognized as key regulators of yield stability and ecosystem service provision in agricultural systems [18,19]. However, the effects of urban landscape composition and configuration on crop yield remains insufficiently understood. The conflict between urban expansion and food production is intensifying. How to balance urban development and maximize output is an important issue in current urban landscape planning. This study aimed to fill this knowledge gap and provide targeted management recommendations for urban landscapes under varying urbanization intensities.

Biodiversity concerns the variety of organisms in an ecosystem, while landscape diversity concerns the heterogeneity of land cover composition and spatial configuration [20,21]. Biodiversity, shaped by the composition, structure, and function of the surrounding landscape, is a key factor affecting agricultural ecosystems [22]. Numerous studies have explored how plant, insect, and crop diversity influence yield outcomes. Plant community diversity promotes yield stability, and pollinator diversity improves both the quality and quantity of crop yield by enhancing pollination efficiency, pest regulation, and disease suppression [20,23,24]. Crop diversity contributes to higher yields through complementary effects [25]. These effects include temporal and spatial optimization of nitrogen uptake, water absorption, root distribution, and the suppression of weeds, pests, and diseases [26,27,28]. Landscape diversity forms the basis of biodiversity as diverse landscape types support species across multiple trophic levels. Landscape diversity sustains biodiversity by regulating microclimates and promoting habitat diversity [29], thereby functioning as a pathway through which landscapes influence crop yields [21,30,31]. Although the relative importance of these effects compared to direct drivers necessitates context-specific evaluation, landscape diversity may be a significant determinant of yield resilience, especially in suburban settings marked by heterogeneous spatial configurations. However, research on the effects of landscape diversity on crop yield remains scarce, particularly in dynamic urban–rural fringe zones, where current understanding remains relatively limited, and most studies have focused on biodiversity while the impact of landscape diversity on crop yield remains poorly understood. Addressing this gap is essential to better understand how landscape diversity can enhance agricultural productivity in suburban areas.

Previous research has largely focused on national-scale studies encompassing vast agricultural areas. In contrast, this study focuses on suburban croplands, which are subject to pronounced land use and land cover changes. The characteristics of suburban cropland are that the landscape types are diverse and are more deeply influenced by the city [32,33]. The current research landscape lacks clarity regarding the influence of suburban cropland surroundings on in-field crop yield, particularly concerning the specific nature of this influence. Therefore, this study investigates how the composition, configuration, and diversity of urban landscapes surrounding croplands affect yield outcomes, and in what ways these effects manifest. Using remote sensing data to monitor agriculture is one of the significant advancements in agricultural technology. Remote sensing techniques have been extensively applied to crop monitoring and yield forecasting [34]. It also provides a non-destructive, cost-effective, and time-efficient approach for modeling crop growth [35,36]. Accordingly, this study focuses on Changchun, a major agricultural hub located in China’s “Golden Corn Belt”. Based on urban detailed land use and cover from high-resolution GF-2 satellite imagery and the spatial pattern of crop yield by integrating Landsat remote sensing with field-measured crop yield, we assessed the effects of landscape composition, configuration, and diversity on crop yield across different urbanization intensities. The objectives of this study were to (1) identify differences in suburban crop yield across different levels of urbanization intensity and landscape heterogeneity; (2) determine key factors and their thresholds influencing suburban crop yield, including urban landscape patterns and diversity; and (3) elucidate the possible pathways driving crop yield by urban landscape patterns and diversity. This study aims to help understand how urban landscape patterns and diversity affect crop yield, and provides valuable insights for urban planners to safeguard national food security from a landscape perspective.

2. Methodology

2.1. Study Area

This study was conducted in Changchun (43°42′–44°03′N, 125°09′–125°27′E) (Figure 1), which is an important national commercial grain production base in northeast China, located above the world-famous golden corn belt [37]. The city of Changchun experiences a temperate continental sub-humid climate with an average annual precipitation of 567 mm and an annual average air temperature of 4.8 °C [38,39], and the total population is 3.63 million. The urban area of Changchun has experienced rapid urbanization, from 34.0 × 10³ ha in 2000 to 47.5 × 10³ ha² in 2019 [40]. This urbanization process has led to the occupation of a large amount of high-quality cropland. Maize is the dominant crop in the study area, which is a globally important food crop and a major feed crop in China [41]. However, the yield of croplands around cities is threatened by urban expansion. Changchun is thus an excellent location for studying the impact of urban landscape patterns and diversity on suburban crop yield.

2.2. Heterogeneous Landscape and Urbanization Intensity Identification

Classification of land use and land cover (LULC): GF-2 remote sensing images acquired in August 2023 were used for LULC classification. The classification was conducted using the WGS 84/UTM Zone 51N. The spatial resolutions of the GF-2 images were 1 and 4 m for the panchromatic and multispectral bands (red, green, and blue), respectively. Image processing steps, including radiometric, atmospheric, and orthometric corrections and image fusion, were performed first using ENVI 5.3 (Exelis Visual Information Solutions) [42]. Then, using supervised classification in ENVI 5.3, the study area was classified into seven types of LULC (Figure 1): cropland, forest, grassland, water, impervious surface (ISA), bare land, and orchards. Sample points representative of various land use types were selected from the images and labeled; the training sample size for each land use type is more than 200. These samples were used to train the classifier, enabling it to discern distinctive features associated with different land use types. The resulting classification accuracy, measured according to the kappa coefficient, was 90.37%, thus satisfying the required level of accuracy.

Heterogeneous landscape classification: We selected a 1 × 1 km grid as the suburban cropland heterogeneous landscape analysis unit and selected all grids containing cropland for analysis [43]. Different combinations and configurations of various land uses within a landscape can have varying impacts on ecosystem service functions. We used cluster analysis and grouped landscape types into three to eight classes; based on the explanatory power comparison, a six-class division provided the best balance of accuracy and information sufficiency (Figure S1). Finally, the landscape categories were named according to the proportion of landscapes ranked in the top two categories (Table S1). The details are as follows:

(1)

Impervious surface-cropland landscape (CI, cropland-dominated, 40–60% cropland, and 20–40% impervious surface).

(2)

Impervious surface-cropland landscape (IC, impervious-surface-dominated, 50–80% impervious surface, and 1–10% cropland).

(3)

Impervious surface-cropland landscape (CIE, cropland-impervious surface equilibrium, 35–50% impervious surface, and 35–50% cropland).

(4)

Impervious surface-grassland landscape (IG, impervious-surface-dominated, 50–60% impervious surface, and 15–20% grassland).

(5)

Impervious surface-grassland landscape (GI, grassland-dominated, 30–40% grassland and 20–30% impervious surface).

(6)

Impervious surface-bare land landscape (BI, bare-land-dominated, 30–40% impervious surface, 40–60% bare land).

By applying these criteria and excluding landscapes that did not meet the standards, in conjunction with our chosen fish net scale, we were able to effectively assess landscape units in the suburban croplands of Changchun. Specific descriptions are presented in Figure 2.

Urbanization intensity: In our study, a proportion analysis was used to acquire the impervious surface areas (ISA) of Changchun City for each 1 × 1 km grid. Ultimately, the ISA coverage ranged from 0% to 100%. Urbanization intensity level was represented by the ISA value for each grid, which was used to determine the urbanization levels, categorized into low-urbanization areas (ISA < 50%), medium-urbanization areas (50% ≤ ISA ≤ 80%), and heavy-urbanization areas (ISA > 80%) [44].

2.3. Landscape Pattern and Diversity Metrics

We utilized a 1 m high-resolution land use map from GF-2 (2023) to calculate landscape metrics, including landscape composition, configuration, and diversity. We selected typical metrics to represent these three aspects based on the study advance on landscape ecology (above mentioned references), and our previous experience [18,45,46,47]. For landscape composition, we used the proportion of landscape (PLAND) for all land use types, as it directly reflects the available area for crop cultivation or competing land uses. In terms of landscape configuration, we selected the largest patch index (LPI), landscape shape index (LSI), and edge density (ED) to represent patch shape characteristics, all of which can alter microclimate and edge-mediated species interactions. Metrics such as connectivity (CONNECT), number of patches (NP), and patch density (PD) were chosen to reflect landscape fragmentation. The patch cohesion index (COHESION) was used to capture the degree of landscape aggregation, which can influence energy and material flows within cropland, as well as pest-enemy dynamics and pollinator movements. For landscape diversity, we employed the Shannon–Wiener diversity index (SHDI) and the Simpson diversity index (SIDI) to capture differences in land use diversity that can contribute to ecosystem resilience and yield stability (

Table 1. Description of landscape pattern and diversity index.

Landscape Metrics	Index	Expression	Description
Landscape composition	Proportion of Landscape (PLAND)	${\mathrm{PLAND}=\mathrm{P}}_{\mathrm{i}}={\displaystyle \frac{{\displaystyle \sum _{\mathrm{j}=1}^{\mathrm{n}}{\mathrm{a}}_{\mathrm{ij}}}}{\mathrm{A}}}\times 100$	Pi = proportion of the landscape occupied by patch type (class) i. aij = area (m2) of patch ij. A = total landscape area (m2).
Landscape configuration	Edge density (ED)	$\mathrm{E}\mathrm{D}={\displaystyle \frac{{\displaystyle \sum _{\mathrm{k}=1}^{\mathrm{m}}{\mathrm{e}}_{\mathrm{i}\mathrm{k}}}}{\mathrm{A}}}\times 10,000$	eik = total length (m) of edge in landscape involving patch type (class) i; includes landscape boundary and background segments involving patch type i. A = total landscape area (m2).
	Largest patch index (LPI)	$\mathrm{LPI}={\displaystyle \frac{\max ({\mathrm{a}}_{\mathrm{i}\mathrm{j}})}{\mathrm{A}}}\times 100$	aij = area (m2) of patch ij. A = total landscape area (m2).
	Landscape shape index (LSI)	$\mathrm{LSI}={\displaystyle \frac{25\mathrm{E}*}{\sqrt{\mathrm{A}}}}$	E* = total length (m) of edge in landscape; includes the entire landscape boundary and some or all background edge segments. A = total landscape area (m2).
	Number of patches (NP)	$\mathrm{NP}={\mathrm{n}}_{\mathrm{i}}$	ni = number of patches in the landscape of patch type (class) i.
	Patch density (PD)	$\mathrm{PD}={\displaystyle \frac{{\mathrm{n}}_{\mathrm{i}}}{\mathrm{A}}}\times \mathrm{10,000}\times 100$	ni = number of patches in the landscape of patch type (class) i. A = total landscape area (m2).
	Connectivity (CONNECT)	$\mathrm{CONNECT}=\left[{\displaystyle \frac{{\displaystyle \sum _{\mathrm{j}\mathrm{=}\mathrm{k}}^{\mathrm{n}}{\mathrm{c}}_{\mathrm{ijk}}}}{\frac{{\mathrm{n}}_{\mathrm{i}}({\mathrm{n}}_{\mathrm{i}}-1)}{2}}}\right]\times 100$	cijk = joining between patch j and k (0 = unjoined, 1 = joined) of the corresponding patch type (i), based on a user-specified threshold distance. ni = number of patches in the landscape of the corresponding patch type (class).
	Patch Cohesion Index (COHESION)	COHESION = $\left[1-{\displaystyle \frac{{\displaystyle \sum _{\mathrm{i}\mathrm{=}\mathrm{1}}^{\mathrm{m}}{\displaystyle \sum _{\mathrm{j}\mathrm{=}\mathrm{1}}^{\mathrm{n}}{\mathrm{P}}_{\mathrm{i}\mathrm{j}}^{*}}}}{{\displaystyle \sum _{\mathrm{i}\mathrm{=}\mathrm{1}}^{\mathrm{m}}{\displaystyle \sum _{\mathrm{j}\mathrm{=}\mathrm{1}}^{\mathrm{n}}{\mathrm{P}}_{\mathrm{i}\mathrm{j}}\sqrt{{\mathrm{a}}_{\mathrm{i}\mathrm{j}}^{*}}}}}}\right]\times {\left[1-{\displaystyle \frac{1}{\sqrt{\mathrm{Z}}}}\right]}^{-1}\times 100$	pij* = perimeter of patch ij in terms of number of cell surfaces. aij* = area of patch ij in terms of number of cells. Z = total number of cells in the landscape.
Landscape diversity	Shannon’s Diversity Index (SHDI)	$\mathrm{SHDI}=-{\displaystyle \sum _{\mathrm{i}\mathrm{=}\mathrm{1}}^{\mathrm{m}}({\mathrm{P}}_{\mathrm{i}}\times \ln {\mathrm{P}}_{\mathrm{i}})}$	Pi = proportion of the landscape occupied by patch type (class) i.
	Simpson’s Diversity Index (SIDI)	$\mathrm{SIDI}=1-{\displaystyle \sum _{\mathrm{i}\mathrm{=}\mathrm{1}}^{\mathrm{m}}{\mathrm{P}}_{\mathrm{i}}^2}$

). We analyzed a set of landscape metrics at two levels. At the class level, metrics including PLAND, ED, LPI, LSI, NP, and PD were calculated for each land use category. At the landscape level, COHESION, CONNECT, SHDI, and SIDI were computed to characterize overall spatial structure and diversity. All metrics were calculated using an integrated landscape pattern analysis approach that accounted for the entire mosaic of land cover classes. Detailed descriptions of the selected landscape metrics and their calculation methods can be found in the Fragstats 4.2 manual [48].

2.4. Spatial Estimation of Crop Yield

A stratified random sampling method was used based on a map of suburban croplands [49]. Suburban crop yield plot surveys were conducted between September and October 2024. Before formal investigation, a field investigation was conducted to ensure the scientific validity and rationality of the sample site selection. A total of 102 cropland plots were set up in 34 typical croplands in suburban areas, with sampling quadrats of 30 × 30 m (Figure S2). We used a high-precision global positioning system MG838GPS (UniStrong, Beijing, China) to pinpoint the location of each sampling plot with submeter precision. For each sampling plot, the number of effective maize ears was recorded, followed by counting the kernel numbers on 30 randomly selected ears. Subsequently, 20 representative ears were collected and brought back to the laboratory for measurement of kernel weight and moisture content. Finally, the unit area yield (t/ha) of maize was calculated for each plot. The following formula was used [50]:

$$CY(t/ha)=[CropWeight\times 10\times \left(100-MC\right)/\left(100-AdjustedMC\right)/\left(PlotArea\right)]$$

(1)

where CY is the crop yield, MC is the moisture content of the plot, the AdjustedMC is 14%, and the plot area is 900 m².

Meanwhile, EVI introduces the blue light band for atmospheric aerosol correction, reducing background noise and atmospheric interference, and making EVI have a better correlation with the absorption ratio of photosynthetically active radiation of crops, a factor crucial for the formation of final yield per unit area [51,52]. To establish the spatial forecasting model of crop yield, we utilized 30 m resolution Landsat 8 remote sensing imagery (dataset ID: LANDSAT/LC08/C02/T1_L2) equipped with OLI/TIRS sensors (Ball Aerospace & NASA Goddard Space Flight Center, Boulder/Greenbelt, CO/MD, USA), acquired from June to September 2024, to calculate the enhanced vegetation index (EVI) of each sample plot [53,54,55]. All the images involved in the analysis were automatically matched with the Landsat 8 path/row corresponding to the boundary of the study area through the spatial filtering function of the GEE platform. The actual effective images are based on the synthesis of all cloudless/low-cloud images from June to September 2024. This data product has been pre-processed by the United States Geological Survey and includes the necessary radiometric calibration and atmospheric correction steps, and can be directly used for quantitative remote sensing analysis. It is particularly noted that all the band data used in this study to calculate the enhanced vegetation index are directly derived from this surface reflectance product that has been adjusted for the atmosphere, thus ensuring the accuracy of the EVI calculation results and eliminating the uncertainty of whether they are based on apparent reflectance or surface reflectance. The following formula was used:

$$EVI=2.5{\displaystyle \frac{NIR-RED}{NIR+6RED-7.5BLUE+1}}$$

(2)

where NIR represents the near-infrared band, RED represents the red band, and BLUE represents the blue band.

A crop yield prediction model was developed using field measurements of crop yield data gathered from 72 plots (70% of the total) and the corresponding EVI parameter data from remote sensing images. Nonlinear regression was used to construct the models. Field measurements of the crop yield data for 30 plots (30% of the total) were employed to evaluate the trustworthiness and accuracy of the prediction models.

2.5. Data Analysis

A series of methods were employed to identify landscape patterns and diversity factors affecting crop yield, determine the significant influences of landscape diversity on yield under varying degrees of urbanization, and evaluate the direct and indirect effects of landscape diversity indicators. A random forest model [49] was employed to detect the landscape factors that exerted the strongest influence on crop yield at different intensities of urbanization. Using the Random Forest package, this model was configured with 500 decision trees (ntree = 500) to ensure the stability and reliability of the results. The number of variables available for splitting at each node (mtry) was set to the default value for regression. The importance of each predictor was quantified based on IncNodePurity. The node purity gain is based on the Gini index or variance reduction (regression problem), and it serves to quantify the contribution of each independent variable to the enhancement of node purity at the time of decision tree splitting [56]. All analyses were conducted in R version 4.2.1. Subsequently, linear regression models and correlation coefficients were applied to identify the potential associations between landscape diversity factors and yield. Finally, to explore the potential for complex relationships between urban landscape diversity and yield under different urbanization intensities, structural equation modeling (SEM) was employed to analyze the direct and indirect relationships between the variables and to detect potential relationships. The accuracy of the model was evaluated based on the chi-square degrees of freedom (${\chi }^2/df$) and the root mean square error of approximation (RMSEA) [57,58].

$$({\chi }^2/df)={\displaystyle \sum (\frac{(O-E{)}^2}{E}/df)}$$

(3)

O is the actual value, $E$ is the expected value, and $df$ is the number of degrees of freedom.

$$RMSEA=\sqrt{\left[\max ({\chi }^2/df)/df(N-1),0\right]}$$

(4)

where ${\chi }^2$ is the chi-square value, $df$ is its degrees of freedom, and $N$ is the sample size.

Preprocessing of the remote sensing data and yield model calculations were conducted using ENVI 5.6 (Harris Geospatial Solutions, Broomfield, CO, USA) and ArcGIS 10.5 (Esri, Redlands, CA, USA), and all statistical data analyses were performed using IBM SPSS version 26 (IBM, Armonk, NY, USA). All the landscape indicators used in this study conformed to normality with Kolmogorov–Smirnov (KS) p > 0.05 and passed the multicollinearity test (VIF < 10) (Figure S2).We used the Random Forest package (version 6.4-14) in R software version 3.6 (R Core Team, Vienna, Austria) and structural equation modeling based on the maximum likelihood method to establish the relationship between landscape patterns and crop yield under different urbanization intensities. We selected the landscape index within the fishing nets and correlated it with the production data within those nets. Then, we constructed models separately according to different levels of urbanization. The model was constructed using AMOS 25.0 (IBM, Armonk, NY, USA), and the final results were obtained by adjusting the model for relationships between the variables.

3. Results

3.1. Spatial Pattern of Crop Yield Across Different Urbanization Intensity

The total area enclosed by the fifth-ring road of Changchun covers 531.42 km², of which the suburban cropland within the fifth-ring road covers 57.25 km², accounting for 11% of the total area. The average area of the cropland patch was 2.81 ha, ranging from 0.001 ha to 382.79 ha. Suburban cropland is characterized by a fragmented distribution, with more cropland in the west than in the east, and more in the north than in the south (Figure 1). We utilized field-measured crop yield and the enhanced vegetation index (EVI) to establish a crop yield estimation model. Based on the modeling results, the EVI crop yield model demonstrated high accuracy (Figure 3b) in estimating crop yield. The validation results indicated a close similarity between the modeled and the plot-measured crop yields (R² = 0.92, root mean square error = 0.70) (Figure 3c). We then obtained the spatial distribution of the crop yield in suburban Changchun (Figure 3a). Overall, the maximum crop yield was 11.95 t/ha.

Crop yield varied significantly across different urbanization intensity and heterogeneous landscapes, showing a consistent decline with increasing urbanization intensity level. The highest yield was observed under low urbanization, at 5.98 t/ha (Figure 4a), whereas the lowest yield was recorded under heavy urbanization, at 4.08 t/ha. The crop yield under low urbanization intensity was 46.5% (Based on the average value) higher than that under heavy urbanization. Among the different landscape types, CI landscapes recorded the highest yield at 6.25 t/ha, whereas IC landscapes recorded the lowest yield at 4.06 t/ha (Figure 4b). Notably, the crop yield in the CIE landscapes was the second highest among all landscape types, at 5.95 t/ha.

3.2. Relative Importance of Landscape Pattern and Diversity on Crop Yield

Results revealed that impervious surface edge density (ED_ISA), impervious surface proportion (PLAND_ISA), and largest impervious surface patch index (LPI_ISA) were the most important factors influencing the yield of suburban croplands in Changchun (Figure 5a). Additionally, significant contributions were also observed from cropland metrics—including largest patch index (LPI_cropland), proportion (PLAND_cropland), and edge density (ED_cropland)—as well as forest metrics comprising edge density (ED_forest), patch density (PD_forest), largest patch index (LPI_forest), and proportion (PLAND_forest).

In particular, the influences of landscape patterns and diversity indicators on crop yields varied across different urbanization intensities. Under low urbanization, PD_forest, PLAND_ISA, and PLAND_forest significantly affected crop yield (Figure 5b). Under moderate urbanization, Shannon’s diversity indices (SHDI), ED_ISA, and LPI_ISA were the key determinants of crop yield (Figure 5c). Under heavy urbanization, ED_cropland, cropland patch density (PD_cropland), and LPI_cropland produced significant effects on yield (Figure 5d). Subsequently, a quantitative regression model was developed to clarify the relationship between the landscape factors and crop yield. Under low urbanization, crop yield increased significantly with the increase in forest patch density (R² = 0.14, p < 0.01) (Figure 6d) and forest proportion (R² = 0.21, p < 0.01) (Figure 6f). The crop yield decreased with the increase in the proportion of impervious surface (R² = 0.22, p < 0.01) (Figure 6e). Under moderate urbanization, the higher the landscape diversity, the higher the crop yield (R² = 0.40, p < 0.01) (Figure 6g), while the higher the edge density of impervious surface (R² = 0.21, p < 0.01) (Figure 6h) and the largest patch index, the lower the crop yield (R² = 0.16, p < 0.01) (Figure 6i). Under heavy urbanization, The higher the edge density of farmland (R² = 0.33, p < 0.01) (Figure 6j) and the patch density (R² = 0.32, p < 0.01) (Figure 6k), the higher the crop yield. The larger the largest patch index of farmland, the lower the crop yield (R² = 0.44, p < 0.01) (Figure 6l). The research results also revealed that some landscape factors have a nonlinear relationship with crop yields. Overall, when the ED_ISA ≤ 200 m/ha, crop yield declined rapidly with increasing ED_ISA. However, once ED_ISA > 200 m/ha, the rate of yield decline slowed considerably despite further increases in ED_ISA. (Figure 6a). Under low urbanization, the threshold of PD_forest for affecting crop yield was 40 patches/ha (Figure 6d).

3.3. Driving Mechanism of Landscape Pattern and Diversity on Crop Yield

The SEM indicated that the effects of landscape composition, configuration, and diversity on suburban crop yield varied across different urbanization intensities. Specifically, landscape composition influenced crop yield both directly and indirectly through landscape configuration and diversity. Overall, PLAND_ISA was negatively correlated with yield, both directly and indirectly, through its negative impact on LPI_ISA, SHDI, and SIDI (Figure 7a); the indirect effect was the main effect, which is − 0.11 (Figure 8a). PLAND_cropland positively influenced yield by increasing LPI_cropland. Under low urbanization intensity, PLAND_forest indirectly enhanced yield by influencing ED_forest and PD_forest. (Figure 7b). Higher PLAND_ISA values were associated with lower crop yields, with the direct effect being the main effect, which is 0.42 (Figure 8b). Under medium urbanization intensity, PLAND_cropland positively influenced yield by enhancing LPI_cropland and SHDI (Figure 7c), and the indirect effect was the main effect, which is 0.19 (Figure 8c). Under heavy urbanization, PLAND_cropland exhibited a direct positive correlation with yield, alongside an indirect negative effect on crop yield through ED_cropland and LPI_cropland. An increase in ED_cropland and LPI_cropland was negatively correlated with yield (Figure 7d), with the direct effect being the dominant factor, which is 0.15 (Figure 8d).

4. Discussion

Grain production plays a key role in agricultural ecosystem services. Urban expansion leads to significant changes in landscape patterns and diversity surrounding croplands during global urbanization, possibly directly or indirectly damaging ecological services and crop yields. Our study highlights the effects of urban landscape patterns and diversity with regard to crop yields and their possible driving mechanisms under rapid urbanization.

4.1. Urbanization Plays a Key Role in the Changes in Suburban Crop Yield

Our findings demonstrated a significant negative relationship between urbanization intensities and crop yield, which was characterized by a continuous decline in crop yield as urbanization intensity increased (Figure 4a). Our results are consistent with the general understanding of the disruptive impacts of urbanization on agricultural systems [59,60,61] but provide a novel quantification of the effects of urbanization intensity on crop yield. In contrast to prior studies that predominantly focused on analyzing the effects of climatic or administrative regions on crop yield [18,21,30], the present study systematically investigated crop yield responses across different urbanization intensities. These findings highlight the critical role of impervious surface areas in shaping agricultural productivity. Impervious surface areas disrupt ecosystem processes by reducing water infiltration, increasing surface runoff [62], and exacerbating urban heat island effects [63]. Such disruptions may reduce the availability of essential resources such as water and nutrients for crops, while impervious surface area expansion disrupts the surrounding non-natural habitats of croplands. This, in turn, affects the biodiversity within the farmland, contributing to the observed decline in crop yield [64]. In addition, we found that crop yield differed significantly across different heterogeneous landscapes, being the highest in impervious surface-cropland landscapes dominated by cropland (CI) landscapes and the lowest in impervious surface-cropland landscapes dominated by impervious surface (IC) landscapes (Figure 4b). Interestingly, the landscapes produced by impervious surface-cropland landscapes with a cropland-impervious surface equilibrium (CIE) also showed a relatively higher crop yield. This may be attributed to the interplay between landscape patterns and diversity. Previous research has shown that landscapes with moderate complexity exhibit over 20% increase in maize production [30]. This may be because appropriately configured landscape diversity can provide suitable habitats and food resources for pollinating insects and predators, thereby enhancing biodiversity. By promoting pollination, it improves crop quality and yield, while simultaneously preventing yield reduction through pest control [65]. Therefore, we recommend that during the development of medium-sized cities, moderate landscape diversity should be maintained, particularly in areas surrounding suburban farmlands, to achieve a balance between food production and urban development. This means that balanced landscapes frequently promote biodiversity, which is beneficial for enhancing ecosystem services such as pollination, pest regulation, and soil health [66,67]. Therefore, achieving a reasonable balance between urban development and landscape heterogeneity is important in the planning and management of urban areas for suburban cropland and food security.

4.2. Key Landscape Pattern and Diversity Factors Affecting Crop Yield

We found that the main factors affecting the suburban crop yield were impervious surface edge density (ED_ISA), the relative proportion of the ranking by the ED_ISA which is 21.5%, impervious surface proportion (PLAND_ISA; 20.26%), and largest impervious surface patch index (LPI_ISA; 19.76) (Figure 5a). The dominance of large patches of impervious surfaces, as indicated by high LPI, further exacerbates the loss of ecosystem diversity and functionality. Additionally, increased ED_ISA may signify fragmented landscapes, potentially hindering agricultural efficiency and increasing vulnerability to external disturbances. We also found that when the density of impermeable water surface edges (ED_ISA) reaches 200 m/ha, the crop yield is affected and decreases due to the density of impermeable water surface edges (Figure 6a). This finding strongly aligns with the core “pattern-process” principle in landscape ecology [68], which posits that ecological processes, such as agricultural production, often respond nonlinearly to changes in landscape pattern, with state shifts occurring at critical thresholds. Our results provide empirical evidence that an ED_ISA of 200 m/ha serves as one such pivotal threshold, triggering a state shift and delineating two distinct response regimes: a high-sensitivity regime and a low-sensitivity regime. From a land-use planning perspective, this threshold effectively defines an “ecological security baseline,” underscoring the necessity for stringent regulatory interventions before urban expansion surpasses this critical limit. Previous studies have focused on large areas of conventional croplands, but insufficient attention has been paid to suburban croplands. Our results highlight the intricate ways in which urbanization-driven changes in landscape composition and configuration affect agricultural production.

In addition, the varying impacts of landscape composition, configuration, and diversity on crop yield across urbanization intensities reflect the intricate interplay between ecological dynamics and anthropogenic pressures. In this study, in low-urbanization areas, forests emerged as key contributors to crop yield (Figure 5b), crop yield increased significantly with the increase in forest patch density (R² = 0.14, p < 0.01), likely because of their capacity to deliver crucial ecosystem services, including water regulation, microclimate stabilization, and biodiversity support [69,70]. These services enhance agricultural productivity by creating favorable growing conditions and buffering environmental fluctuations. Xin’s research found that forest proportion has a positive effect on maize yield [18]. This is in line with our results, and we highlight the benefits of more forests and scattered woodlands in croplands in less-urbanized areas. Impervious surface areas (ISA) also play a dominant role in reducing crop yield (R² = 0.22, p < 0.01) because urban expansion is typically scattered, leading to fragmented croplands and disrupted ecological processes [9]. Studies have shown that impervious surface negatively affects soil acidification, and urban refuse affects nutrient cycling [32]. This is particularly significant in less-urbanized regions, where agriculture relies heavily on local soil quality. Additionally, the expansion of impervious surface during early urbanization stages is associated with higher edge effects, which intensify habitat fragmentation and reduce connectivity. This counteracts the flow of pollinating insects and thus negatively affects yield [71]. The current study showed that when forest patch density (PD_forest) reached 40 patches/ha, the crop yield will experience a rapid ascent (Figure 6d). This indicates that increasing the forest patch density (PD_forest) to 40 patches/ha could more effectively enhance the crop yield, as it offers habitats for organisms within the cropland, potentially boosting cropland biodiversity. Under medium urbanization, the landscape is dominated by urban infrastructure, leaving croplands as isolated patches within an impervious surface area matrix. Here, the diversity of the cropland and the configuration of the ISA become critical for maintaining productivity (Figure 5c). Consistent with previous studies [30,31,72], our results suggest that landscape diversity has a strong positive impact on crop yields.

Under medium urbanization, the proportion of impervious surfaces exceeds 50%. Moreover, suburban croplands are often concentrated, with high landscape diversity around them. The higher the landscape diversity, the higher the crop yield (R² = 0.40, p < 0.01). These areas feature multiple land cover types such as woodlands, grasslands, and water bodies. The complementary effects among these land types contribute to yield stability. This complementarity is also linked to biodiversity, thereby enhancing overall crop yield [30]. This complementary effect may be closely associated with biodiversity. Bian’s research found that landscape diversity, in relation to the abundance of natural enemies of the corn borer, is less favorable for the aggregation of these natural enemies within croplands when the proportion of non-cultivated habitats reaches 20–30% in a highly diverse landscape context [64]. Notably, an increase in forested areas within diversified landscapes has been shown to effectively enhance the abundance of natural enemies on farmlands, thereby mitigating pest pressure on corn and ultimately improving yields. Moreover, with heavy urbanization, the direct impact of impervious surface on yield diminishes, possibly because most agricultural landscapes in these areas are already highly fragmented, and further increases in impervious surface may have a relatively smaller marginal effect on yield. Instead, the configurations of cropland patch density (PD_cropland) (R² = 0.32, p < 0.01) and cropland edge density (ED_cropland) (R² = 0.33, p < 0.01) became stronger determinants of productivity (Figure 5d and Figure 6d) because it reflects the ability of the remaining agricultural patches to sustain essential ecological processes.

Urbanization intensity plays a crucial role in shaping the causal relationship between suburban landscape patterns and cropland yield. Although the total area of arable land in China is strictly protected through legislative measures, rapid urban expansion has led to the spatial redistribution and fragmentation of farmland, fundamentally altering its surrounding landscape context. As urbanization intensifies, the proportion of impervious surfaces increases, and the remaining cropland becomes more isolated and less ecologically connected. These changes often reduce habitat heterogeneity and disrupt ecological processes such as pollination, pest regulation, and nutrient cycling. Consequently, the positive effects of favorable landscape configurations on crop yield tend to diminish under high urbanization intensity. In contrast, in less-urbanized areas, more continuous and well-connected landscapes help sustain beneficial ecological interactions, thereby stabilizing and enhancing agricultural productivity. Therefore, urbanization intensity acts as a key moderator that reshapes both the direction and magnitude of the landscape–yield relationship. Understanding this moderating effect is essential for designing sustainable land use policies that maintain agricultural productivity while accommodating urban growth. These findings suggest that managing landscape patterns to optimize connectivity and diversity is a critical strategy for maintaining agricultural productivity in rapidly urbanizing regions.

4.3. Direct and Indirect Effects of Landscape Pattern and Diversity on Crop Yield in the Context of Rapid Urbanization

Using an SEM, this study investigated the direct and indirect effects of landscape composition, configuration, and diversity on suburban crop yield under different urbanization intensities. The findings revealed distinct patterns of direct and indirect influences, offering critical insights into how urbanization reshapes agricultural productivity through landscape-level interactions. Our study found that the composition of impervious surface, forests, and croplands in suburban areas had a significant impact on crop yield, which could be regulated by landscape configuration and diversity (Figure 7a). Specifically, impervious surface proportion (PLAND_ISA) was negatively associated with crop yield, both directly and indirectly, by reducing the connectivity and diversity of the landscape. Overall, impervious surfaces that make direct contact with the soil can inhibit rainwater infiltration, exposing crops to drought stress and consequently impairing their photosynthesis and nutrient uptake. Shao’s research documented that impervious surface disrupts hydrological processes, such as infiltration and soil moisture retention, which are critical for crop growth [73]. Impervious surface can also physically cut through the landscape, causing it to shatter and preventing the natural flow of water, soil, nutrients and organisms. The research findings of Zhao and Feng demonstrated that the expansion of impervious surface reduced landscape connectivity and biodiversity, further exacerbating the decline in agricultural yield [9]. In contrast, cropland proportion (PLAND_cropland) consistently had a positive influence on yield, either directly or by enhancing the largest croplands patch index. These results suggest that cropland expansion supports higher yields by maintaining structural continuity and reducing fragmentation. Larger cropland patches, as indicated by higher LPI, are likely to reduce edge effects and provide more stable microclimatic conditions conducive to crop growth. Simultaneously, croplands with a high LPI are conducive to high biodiversity [74]. Furthermore, cohesive landscapes can mitigate the pressure of urban encroachment by maintaining functional ecosystems within agricultural zones [29]. These results highlight the importance of maintaining a balance between urban development and cropland conservation.

Our study found that landscape configuration served as a key mediator of composition–yield relationships, with its effects varying significantly across urbanization intensity. In areas of low urbanization, the direct impact of impervious surface proportion on yield is due to rural-to-suburban transitional landscapes, where even small increases in ISA significantly disrupt agricultural systems (Figure 7b). The heat island effect caused by impervious surface may also be an important factor affecting production. The impervious surface absorbs solar heat, causing the local temperature to rise rapidly [75], which may accelerate soil water evaporation, dehydrate plants, and reduce crop productivity [76]. These processes explain the strong direct negative influence of impervious surface in low-urbanization areas, where agricultural systems are highly sensitive to urban encroachment. The positive influence of PLAND_forest on crop yield, mediated through ED and PD, indicates that the shape of the forest has a positive effect on crop yield. Shelterbelts at the edge of the cropland were distributed in stripes. This shape resulted in increased edge density of the forest. Shelterbelts can conserve soil and water, prevent wind erosion, and protect agricultural yields [23]. Urban forests can mitigate the negative effects of urban expansion by enhancing the connectivity of biodiversity habitats in croplands and regulating the microclimate, thereby reducing pressure on agricultural land. Therefore, strict control of impervious surface expansion is critical for protecting cropland productivity. In areas with low urbanization, urban policymakers should consider the construction of more forest patches. In this context, crop yield may depend more on maintaining the balance and complementarity between cropland and forest landscapes than on the direct effects of impervious surface expansion. However, the effect of PLAND_ISA on crop yield diminishes in areas with medium urbanization because its importance is low; therefore, it was excluded from the SEM. (Figure 7c). This suggests that impervious surface proportion is no longer the dominant factor influencing agricultural productivity. Instead, the role of SHDI became more significant. This finding may be attributed to the development of urban green spaces with urban expansion, accompanied by significant efforts by city managers. These efforts include extensive urban forest planning, creation of urban parks, and large-scale afforestation [38]. These parks feature a variety of land types, including water bodies, grasslands, flowerbeds, and forests. This landscape diversity might be due to the water bodies around the farmland, which regulate the regional microclimate, increase the humidity of the farmland air, and reduce the evaporation of water from the farmland. Meanwhile, the dead branches and fallen leaves in the forest and grassland in the landscape can provide organic matter sources for farmland and promote the nutrient cycle in the soil, while wildflowers and shrubs in the landscape can attract pollinating insects and birds. This diversity not only provides environmental support for nearby farmlands but also offers valuable habitats for pollinators [77]. Simultaneously, cropland proportion (PLAND_cropland) directly affected the crop yield and had an indirect impact. However, PLAND_cropland negatively affected the yield, although the effect was not statistically significant (Figure 7c). The expansion of impervious surface may have contributed to the increased fragmentation of croplands, which, in turn, may hinder efficient operation of agricultural machinery. This fragmentation has likely led to delays caused by labor inefficiencies [78]. Research shows that, on average, for every 1% increase in farmland fragmentation, the capital productivity of maize farmers decreases by 9.1% and 7.0%, respectively [79]. These challenges are influenced by a complex set of factors, each of which significantly affects agricultural yields. Under heavy urbanization, the factors influencing crop yield are dominated by the composition and configuration of croplands. Cropland proportion demonstrated a significant positive direct effect on yield (Figure 7d) but largest cropland patch index (LPI_cropland) and cropland edge density (ED_cropland) demonstrated a negative indirect effect on yield. Under heavy urbanization, impervious surfaces exceed 80% coverage, while suburban croplands exhibit highly fragmented conditions. A higher largest patch index indicates greater cropland proportion within the landscape. Such croplands, deeply embedded within cities, are more susceptible to multiple pollution sources including industrial emissions and urban runoff [32]. Simultaneously, they may be affected by the urban heat island effect. Elevated temperatures accelerate water evaporation, impair photosynthetic efficiency, reduce pollination success rates, and diminish agricultural productivity [80,81], obscuring its contribution to crop yield even as the proportion of croplands increases. Surplus croplands may be concentrated in low-quality marginal areas, such as those with poor soil or plots close to roads, which could further undermine their productivity. However, croplands retained at heavy urbanization levels are typically situated on high-value land, which is often subjected to a higher intensity of irrigation or fertilization to offset the negative effects of urbanization. This study has confirmed that landscape structure profoundly affects crop yields by regulating ecosystem services. This finding establishes an intrinsic connection between agricultural output and the ecological consequences of landscape fragmentation, revealing the fundamental trade-offs that exist among different ecosystem services in the context of agricultural intensification. The urban landscape can directly or indirectly affect crop yields by influencing the diversity of the landscape, the composition and configuration of forest areas, and the configuration of farmland. Such effects vary across different levels of urbanization. These findings underscore the need for differentiated landscape management strategies for varying urbanization intensities. It should be noted that the results of this study may also be affected by factors such as sample size, sampling error, and measurement error, leading to a certain degree of uncertainty in the results. As a preliminary exploration, this study needs to be further studied in combination with more urban and natural environment, soil, and humanistic environment factors in the future.

4.4. Limitations, Implications for Other Regions, and Directions for Future Research

This study focuses on the suburban areas of Changchun, Jilin Province, and caution is thus warranted when extrapolating the conclusions to other regions. Changchun lies within the black soil belt of Northeast China, characterized by a temperate continental sub-humid climate and a typical urbanization pattern dominated by corn–soybean cropping systems. These conditions provide a suitable context for revealing the interaction mechanisms between urban landscapes and crop yield. However, this does not imply that the mechanisms identified here are universally applicable across all climate–crop–governance settings. Furthermore, our landscape yield model is affected by temporal mismatch, which may limit the model’s stability under climate change conditions. Future research will collect multi-year synchronous landscape and yield data to enhance the model’s spatiotemporal applicability and prediction accuracy. It is important to note that this study employed bivariate analysis and did not control for potential confounding variables. Future research should integrate factors such as soil quality, irrigation, and topography using multivariate or spatial regression models to establish causal inference.

Our key findings can be summarized as follows. The negative impacts of increased impervious surface area and reduced forest cover on yield are well-documented [1,2] and may occur through multiple pathways, including microclimatic alterations, accelerated soil degradation [82], and the weakening of biodiversity and ecosystem services [83,84]. The nonlinear yield response to the edge density of impervious surface area (ISA)—with a threshold of approximately 200 m/ha in this study—suggests that fragmentation can trigger functional shifts. At this threshold, the loss of connectivity may precipitate a sudden collapse of ecosystem services, aligning with the theory of critical transitions in fragmented landscapes [85]. Furthermore, we observed that the impact of urbanization on production exhibited a hierarchical transformation, passing through three stages: from composition, to the Shannon Diversity Index (SHDI), and finally to the internal cropland structure. Importantly, these mechanisms are regulated by local context. In irrigated paddy fields, warm and arid regions, or under high-input intensive management, the influence of landscape patterns may weaken or shift, altering both thresholds and sensitivities. Accordingly, this study should be viewed as providing a transferable mechanistic framework rather than precise, universally applicable predictions. Its applicability is bounded by climate, soil, crop types, management intensity, and spatial scale (here, mainly a 1 km buffer zone).

To test the generality of these patterns and quantify the context dependence of thresholds, we plan to integrate multi-source datasets (e.g., Sentinel-1 SAR and PlanetScope) to conduct multi-scale, cross-seasonal comparative studies in Beijing (a large temperate-zone city), Hangzhou (a subtropical rice-growing region), and Haikou (a tropical city with tourism-driven land use). Meanwhile, we aim to determine how the landscape indirectly affects production through its impact on biodiversity.

These studies will address the following questions:

(1)

Are the observed transformations in dominant drivers robust across contexts?

(2)

How do thresholds vary with climate, crops, and irrigation regimes?

(3)

How strong is the regulatory role of landscape diversity under different biodiversity and governance conditions?

(4)

Can urban impervious surfaces, cropland protection belts, and urban greenways alleviate urbanization-related stress and boost agricultural sustainability?

5. Conclusions

Using field surveys and remote sensing satellite images of suburban croplands in Changchun, we explored the relationship between landscape patterns, landscape diversity, and suburban crop yield. The main conclusions are as follows:

(1)

We found that crop yield declined continuously with increasing urbanization intensity. The crop yield under low urbanization was 46.5% higher than that under heavy urbanization. Within heterogeneous agricultural landscapes, impervious surface-cropland landscape dominated by cropland landscapes (CI) landscapes have the highest crop yield at 6.25 t/ha, whereas impervious surface-cropland landscapes dominated by impervious surface landscapes (IC) have the lowest crop yield at 4.06 t/ha, and that of impervious surface-cropland landscapes with cropland-impervious surface equilibrium landscapes (CIE) is 5.95 t/ha.

(2)

The main factors of landscape patterns and diversity that influence crop yield differ with different urbanization intensities. Overall, edge density ED, proportion, and largest patch index of the impervious surface were the main factors influencing crop yield, where the threshold for ED_ISA = 200 m/ha. In low-urbanization areas, forest patch density, forest proportion, and impervious surface proportion were important drivers, and the threshold for PD_forest = 40 patches/ha. Under medium urbanization, Shannon’s diversity indices, impervious surface edge density, and largest impervious surface patch index were important factors influencing crop yield. During heavy urbanization, the patch density, edge density, and largest cropland patch index of cropland are important factors. This indicates that the control landscape indicators of croplands, forests, and impervious surface can effectively influence cropland productivity.

(3)

We found that crop yield was mainly affected by the landscape composition and configuration of croplands and impervious surface. The crop yield effect mechanisms of the driving factors at different urbanization intensities are complex. In low-urbanization areas, forest proportion indirectly enhanced crop yield through forest edge density and forest patch density. Under moderate urbanization, cropland proportion positively influenced crop yield by enhancing largest cropland patch index and Shannon’s diversity indices. Finally, under heavy urbanization, cropland proportion was directly positively correlated with crop yield and indirectly affected crop yield through largest cropland patch index and cropland edge density. Limiting impervious surface and their irregular shapes as well as increasing vegetation cover to alleviate cropland fragmentation can help sustain crop yields and mitigate the negative impacts of urbanization.

Our results suggest how agricultural productivity can be enhanced in suburban urban areas by optimizing landscape diversity in the context of rapid urbanization, thereby providing a scientific basis for the sustainable development of regional agriculture.