Urban Expansion-Driven Cropland NPP Change in the Beijing-Tianjin-Hebei Region, China (2001–2020): Spatiotemporal Patterns, Ecological Selectivity, and Spatially Varying Driver Effects

Highlights

What are the main findings?

• Urban land in the Beijing–Tianjin–Hebei (BTH) region expanded by 45.2% between 2001 and 2020.
• Cropland conversion dominates urban growth: 91.04% of newly urbanized land came from cropland, causing a time-weighted cumulative cropland net primary productivity (NPP) loss of 29.24 Tg C.

What is the implication of the main finding?

• The proposed Normalized Loss Efficiency (NLE) indicator quantifies the selectivity toward high-productivity cropland and peaks at 0.822 in the Southern Functional Expansion Zone (SFE).
• Geographically Weighted Regression (GWR) reveals spatially non-stationary driver effects across functional zones, supporting differentiated spatial controls rather than a one-size-fits-all land management approach.

Abstract

As the largest urban agglomeration and a critical grain production base in northern China, the Beijing–Tianjin–Hebei (BTH) region faces a sharp conflict between rapid urbanization and cropland conservation. Urban expansion inevitably leads to the loss of high-quality agricultural land, posing dual threats to food security and the terrestrial carbon cycle. To accurately assess the ecological costs of this process, this study integrates the CASA model with a time-weighted cumulative model to quantify the spatiotemporal impacts of urban expansion on cropland NPP in the BTH region from 2001 to 2020. Furthermore, a Geographically Weighted Regression (GWR) model was employed to examine the spatially varying effects of key driving factors on cropland NPP loss. The results indicate that urban land in the BTH region expanded by 45.2% over the past two decades, with 91.04% originating from cropland. Despite an overall upward trend in regional cropland NPP driven by climate change and agricultural intensification, the time-weighted cumulative cropland NPP loss attributable to urban encroachment over 2001–2020 reached 29.24 Tg C, which is equivalent to 0.751× the annual total cropland NPP in 2020 (used as a reference benchmark). Crucially, this expansion exhibits distinct ecological selectivity toward high-quality cropland, meaning that urban development has disproportionately encroached upon highly productive land with productivity levels exceeding the regional average. This selective occupation has led to a structural decline in the region’s potential agricultural production capacity. Additionally, GWR results reveal significant spatial non-stationarity in the relationships between cropland NPP loss and its drivers, revealing differentiated response patterns between plains and mountainous areas in terms of socio-economic drivers and physical constraints. These findings expose the hidden threats of urban expansion to food security, providing a crucial scientific basis for formulating differentiated land management policies and coordinating regional urbanization with cropland protection.

1. Introduction

Rapid urbanization stands as the most significant driver reshaping Earth’s land cover in the 21st century. Projections suggest that global urbanization will continue to accelerate over the coming decades, inevitably intensifying the expansion of urban land cover worldwide [1]. This land-use transformation primarily occurs at the expense of encroaching on surrounding agricultural land, as cities often originate and develop in fertile, highly productive plain agricultural regions [2]. In this context, net primary productivity (NPP) is a core indicator characterizing both ecosystem carbon sequestration capacity and crop production potential. The conversion of highly productive cropland into urban impervious surfaces leads to a direct decline in vegetation carbon uptake and a loss of potential agricultural productivity, thereby increasing carbon emission risks and weakening regional food security [3,4]. Therefore, against the backdrop of rapid urban expansion, quantifying cropland NPP change and its spatiotemporal patterns is essential for balancing urban development with cropland conservation and safeguarding food and ecological security [5].

Urban expansion-induced losses of agricultural NPP and their carbon implications have long been a critical topic in global environmental change research. Existing studies indicate that the expansion of urban impervious surfaces causes significant NPP losses through direct vegetation displacement [3,6,7,8]. Studies have employed multi-source remote sensing data (e.g., MODIS) and ecosystem process models (e.g., CASA) to conduct quantitative assessments across various spatiotemporal scales [9,10,11,12,13]. Especially in China, since the reform and opening-up, rapid economic growth and accelerated industrialization have triggered a sustained wave of rural–urban migration, accompanied by the rapid expansion of urban land use [14,15,16,17]. As a populous developing nation with scarce per capita arable land resources [18,19,20,21], China faces a particularly acute conflict between rapid urbanization and cropland conservation. Studies have consistently shown that urban expansion in China is spatially selective—often encroaching on cropland with relatively favorable natural conditions—while the magnitude and spatial patterns of cropland loss vary markedly across regions and city types under different development trajectories, land-use policies, and local constraints [22,23,24,25,26,27]. Despite strict policy requirements to maintain the red line for cropland area, the “replacing high-quality land with low-quality land” pattern under the cultivated land balance program often leads to markedly lower production potential for newly reclaimed cropland than that of the displaced land [28]. This structural decline in cropland quality triggered by urban expansion not only offsets part of the production gains from agricultural intensification [5,29], but also makes it increasingly urgent to precisely quantify the cumulative ecological costs and driver effects of this process within rapidly urbanizing megaregions. However, studies on NPP change in cropland due to urban expansion critically depend on obtaining NPP time series consistent with cropland masks, land-use change processes, and statistical grid resolutions to support change calculations and zoned attribution. Therefore, this study employs the Carnegie–Ames–Stanford Approach (CASA) model to estimate arable land NPP, utilizing the maximum photosynthetic efficiency parameter calibrated for Chinese ecosystems by Zhu et al. (2006) [30] to enhance the model’s accuracy and adaptability to China’s arable land systems.

The Beijing–Tianjin–Hebei region (BTH), as the largest urban cluster in northern China, serves not only as the nation’s political and economic core but also encompasses the North China Plain—one of China’s key winter-wheat cultivation regions. The major winter-wheat production regions (including the NCP and adjacent areas) account for about 84% of the sowing area and about 90% of winter-wheat production in China [31]. Despite relevant policy controls, the BTH region continues to face intense land resource constraints and development–conservation trade-offs under rapid urbanization. Rapid urbanization has led to substantial losses of fertile and productive cropland, exerting growing pressure on regional agricultural production potential and associated ecosystem functions [32,33]. Complicating matters further, the BTH region exhibits strong geographic and socioeconomic contrasts: from the ecologically sensitive mountainous areas in the northwest to the primary plain agricultural production zones in the southeast, different functional zones face distinctly divergent pressures from urban expansion and ecological constraints. Although existing studies have assessed NPP dynamics and related ecological impacts in parts of this region, revealing local-scale patterns of ecological costs [34,35], a systematic evaluation that comprehensively considers the cumulative temporal effects and spatial non-stationarity of driving factors for the entire Beijing–Tianjin–Hebei urban cluster remains lacking. In particular, prior work has often been constrained by (1) inconsistent accounting of multi-stage cropland-to-urban conversion and its long-term cumulative impacts, (2) limited quantification of ecological selectivity that links baseline cropland productivity to loss intensity in a comparable manner, and (3) the use of global models or administrative summaries that may obscure spatial non-stationarity in the associations between NPP loss and its potential socioeconomic and locational drivers. Addressing these limitations is essential for producing comparable loss estimates and for informing differentiated territorial spatial management strategies under heterogeneous development–conservation trade-offs.

To address the aforementioned issues, the specific objectives of this study are as follows: (1) to clarify the spatiotemporal dynamics of urban expansion over the past two decades and quantify the cumulative carbon consequences resulting from cultivated land displacement; (2) to reveal the ecological selectivity of urban expansion toward high-quality cropland and assess its potential threat to regional food production potential; (3) to investigate the spatially varying effects of key driving factors on cropland NPP change across different functional zones using the GWR framework. This study provides scientific evidence for optimizing regional land-use policies and balancing trade-offs between urbanization, food security, and ecosystem carbon functions.

2. Materials and Methods

2.1. Study Area

The Beijing-Tianjin-Hebei region (BTH) is situated in the northern part of China’s North China Plain (36°01′N–42°37′N, 113°04′E–119°53′E), encompassing Beijing Municipality, Tianjin Municipality, and Hebei Province, with a total area of approximately 218,000 km² (Figure 1a). The region features higher elevations in the northwest and lower elevations in the southeast, with diverse landforms encompassing the Yanshan Mountains, Taihang Mountains, foothill plains, and coastal lowlands (Figure 1c). As one of China’s three core economic growth poles, BTH also represents the largest urban cluster in northern China and a typical area of high-intensity human activity. Considering the spatial heterogeneity of natural geographical conditions and socioeconomic development levels, this study adopts a four-zone functional framework that is consistent with the policy discourse and planning practice of BTH coordinated development to stratify the research area (Figure 1b). The four functional zones include the Central Core Functional Zone (CCF), the Southern Functional Expansion Zone (SFE), the Eastern Coastal Development Zone (ECD), and the Northern Ecological Conservation Zone (NEC). As no unified official vector layer of these functional zones is publicly available at the regional scale, we implemented the zoning at the prefecture level using administrative boundary polygons.

Prefecture-level administrative boundaries were obtained from the National Geographic Information Public Service Platform (https://www.tianditu.gov.cn (accessed on 28 November 2025)). Each prefecture-level city in the BTH region was assigned to one functional zone following its dominant functional positioning and the city list reported in this study (Appendix A). Specifically, CCF includes Beijing, Tianjin, Baoding, and Langfang; SFE includes Shijiazhuang, Hengshui, Xingtai, and Handan; ECD includes Tangshan, Qinhuangdao, and Cangzhou; and NEC includes Chengde and Zhangjiakou. All subsequent zonal statistics and comparative analyses were aggregated at the prefecture level. The zoning list and processing workflow are documented in Appendix A (

Table A1. Prefecture-level units and functional-zone assignment in the Beijing-Tianjin-Hebei (BTH) region.

Prefecture-Level Unit	Province/Municipality	Functional Zone
Beijing	Beijing Municipality	CCF
Tianjin	Tianjin Municipality	CCF
Baoding	Hebei Province	CCF
Langfang	Hebei Province	CCF
Tangshan	Hebei Province	ECD
Qinhuangdao	Hebei Province	ECD
Cangzhou	Hebei Province	ECD
Chengde	Hebei Province	NEC
Zhangjiakou	Hebei Province	NEC
Shijiazhuang	Hebei Province	SFE
Hengshui	Hebei Province	SFE
Xingtai	Hebei Province	SFE
Handan	Hebei Province	SFE

Note: CCF = Central Core Functional Zone; ECD = Eastern Coastal Development Zone; NEC = Northern Ecological Conservation Zone; SFE = Southern Functional Expansion Zone. Each prefecture-level unit within the BTH region was assigned to one functional zone following the city-to-zone list shown in this table. Prefecture-level administrative boundary polygons were obtained from the National Geographic Information Public Service Platform (https://www.tianditu.gov.cn (accessed on 28 November 2025)). The zoning attribute was then joined to the prefecture-level boundary polygons in a GIS environment, and all zonal statistics and comparative analyses in this study were aggregated at the prefecture level.

).

2.2. Data Sources and Processing

The datasets used in this study include land cover, meteorological parameters, remote sensing vegetation indices, and socio-economic statistics. The land use and cover change (LUCC) data from 2001 to 2020 were derived from the 30 m annual China Land Cover Dataset (CLCD) [36]. According to Yang and Huang [36], CLCD achieves an overall accuracy of 79.31% based on visually interpreted validation samples, with year-specific overall accuracies ranging from 76.45% to 82.51% in their evaluation. Meteorological data required for the CASA model, including monthly mean temperature and total precipitation, were obtained from the 1 km resolution dataset provided by the National Tibetan Plateau Data Center (https://data.tpdc.ac.cn/ (accessed on 29 November 2025)) [37]; solar radiation data were sourced from the ERA5-Land monthly averaged reanalysis dataset with a spatial resolution of 0.1° [38]. For vegetation parameters, the Normalized Difference Vegetation Index (NDVI) was acquired from the 250 m monthly synthesis product of the National Tibetan Plateau Data Center (https://data.tpdc.ac.cn (accessed on 29 November 2025)) to calculate the fraction of photosynthetically active radiation (FPAR). Additionally, the MODIS NPP product (MOD17A3HGF, 500 m) was acquired from the Google Earth Engine platform (https://earthengine.google.com (accessed on 1 December 2025)) and used as an external benchmark to evaluate the consistency (agreement) of the CASA-derived NPP at the regional scale. To analyze the spatially varying effects of potential driving factors, socio-economic drivers were represented by the 1 km grid GDP dataset generated through spatial interpolation of statistical data based on night-time light and land use [39] and LandScan global population database (https://landscan.ornl.gov (accessed on 3 December 2025)), while topographic factors (elevation and slope) were extracted from the SRTM DEM (90 m) downloaded from the Resource and Environment Science and Data Center (https://www.resdc.cn (accessed on 2 December 2025)). To ensure spatial consistency, all raster datasets were reprojected to the WGS 1984 UTM Zone 50N coordinate system and aligned to a 250 m × 250 m reference grid in ArcGIS v.10.2. Categorical land-cover data (CLCD) were upscaled from 30 m to 250 m using the MAJORITY rule to preserve the dominant class within each aggregated cell. For coarser-resolution continuous drivers (e.g., meteorology at 1 km and solar radiation at 0.1°), we resampled them to 250 m using the ArcGIS default NEAREST method, which preserves original pixel values and avoids generating artificial intermediate values through interpolation. Temporally, meteorological and radiation inputs were prepared at the monthly scale to drive CASA; monthly NPP was estimated and then aggregated to annual NPP. Annual CLCD masks were used to extract year-matched cropland NPP, ensuring that land cover and NPP refer to the same year. To evaluate uncertainty in cropland-mask upscaling, we additionally constructed two fractional-threshold cropland masks (θ = 0.5 and θ = 0.7) based on the proportion of 30 m cropland pixels within each 250 m cell and summarized the resulting differences in cropland extent in Appendix A (

Table A2. Sensitivity of the annual 250 m cropland mask to alternative fractional thresholds (θ = 0.5 and 0.7) in the Beijing-Tianjin-Hebei (BTH) region.

Year	Cropland Area (θ = 0.5, km2)	Cropland Area (θ = 0.7, km2)	Area Difference (km2)	Relative Difference (%)	Disagreement Pixels (Count)	Disagreement (%)
2001	108,249.5106	96,577.3209	11,672.1896	10.78	196,455	5.43
2005	104,906.2219	93,143.1287	11,763.0932	11.21	197,985	5.47
2010	100,983.8837	89,176.8836	11,807.0002	11.69	198,724	5.49
2015	97,224.8748	84,992.7076	12,232.1672	12.58	205,880	5.69
2020	97,029.8184	84,448.7125	12,581.1059	12.97	211,753	5.85

Note: θ denotes the minimum fraction of 30 m cropland pixels within each 250 m cell for labeling the cell as cropland (θ = 0.5 is a looser criterion; θ = 0.7 is stricter). “Disagreement pixels” are cells where the two masks differ (defined as abs (mask0.5 − mask0.7) = 1). This sensitivity test was conducted for the cropland mask derived from CLCD, which is the direct impact controlling cropland NPP accounting in this study.

). Variables for the GWR analysis were further aggregated to a 5 km × 5 km grid scale to match the granularity of macro socioeconomic data. Figure 2 shows the methodological framework of this study.

All carbon units were standardized as follows: 1 Mg = 10⁶ g, 1 Tg = 10¹² g, and 1 km² = 10⁶ m². Accordingly, 1 Mg C·km⁻² = 1 g C·m⁻². Annual rates are reported in g C·m⁻²·a⁻¹, whereas cumulative intensities over the study period are reported in Mg C km⁻² (or the equivalent g C·m⁻²) using the above conversions.

2.3. Methods

2.3.1. Estimating NPP Using the CASA Model

The annual cropland NPP in the BTH region was estimated using the Carnegie–Ames–Stanford Approach (CASA) model, a widely recognized light-use efficiency model that calculates vegetation NPP based on remote sensing data, temperature, precipitation, solar radiation, and land cover types [40,41]. This model calculates NPP based on two primary factors: the Absorbed Photosynthetically Active Radiation (APAR) by vegetation and the actual light-use efficiency (ε) at which this energy is converted into biomass. The calculation is expressed as:

$$\mathrm{NPP}\left(\mathrm{x}, \mathrm{t}\right)=\mathrm{APAR}\left(\mathrm{x}, \mathrm{t}\right)\times \mathsf{\epsilon }\left(\mathrm{x}, \mathrm{t}\right)$$

(1)

where t represents time and x represents spatial position. The APAR(x,t) is determined by the total solar radiation and the fraction of radiation absorbed by the plant canopy (FPAR). This relationship and the components of ε are detailed in the following equations:

$$\mathrm{APAR}\left(\mathrm{x}, \mathrm{t}\right)=\mathrm{PAR}\left(\mathrm{x}, \mathrm{t}\right)\times \mathrm{FPAR}\left(\mathrm{x}, \mathrm{t}\right)$$

(2)

$$\mathrm{PAR}\left(\mathrm{x}, \mathrm{t}\right)=\mathrm{L}\left(\mathrm{x}, \mathrm{t}\right)\times 0.5$$

(3)

where L(x,t) is the total solar radiation (MJ·m⁻²) at pixel x during month t. The constant 0.5 represents the proportion of solar radiation that is photosynthetically active (0.4–0.7 µm). FPAR(x,t) is derived from NDVI and the Simple Ratio (SR), and the final FPAR is obtained by averaging the two estimates:

$$\mathrm{S}\mathrm{R}\left(\mathrm{x}, \mathrm{t}\right)={\displaystyle \frac{1+\mathrm{NDVI}\left(\mathrm{x}, \mathrm{t}\right)}{1-\mathrm{NDVI}\left(\mathrm{x}, \mathrm{t}\right)}}$$

(4)

$${\mathrm{FPAR}}_{\mathrm{NDVI}}\left(\mathrm{x}, \mathrm{t}\right)={\displaystyle \frac{\mathrm{NDVI}\left(\mathrm{x}, \mathrm{t}\right)-{\mathrm{NDVI}}_{\min }}{{\mathrm{NDVI}}_{\max }-{\mathrm{NDVI}}_{\min }}}({\mathrm{FPAR}}_{\max }-{ \mathrm{FPAR}}_{\min })+{\mathrm{FPAR}}_{\min }$$

(5)

$${\mathrm{FPAR}}_{\mathrm{S}\mathrm{R}}\left(\mathrm{x}, \mathrm{t}\right)={\displaystyle \frac{\mathrm{S}\mathrm{R}(\mathrm{x},\mathrm{t})-{\mathrm{S}\mathrm{R}}_{\mathrm{m}\mathrm{i}\mathrm{n}}}{{\mathrm{S}\mathrm{R}}_{\max }-{\mathrm{S}\mathrm{R}}_{\mathrm{m}\mathrm{i}\mathrm{n}}}}({\mathrm{FPAR}}_{\max }-{ \mathrm{FPAR}}_{\min })+{\mathrm{FPAR}}_{\min }$$

(6)

$$\mathrm{FPAR}\left(\mathrm{x}, \mathrm{t}\right)=\min ({\mathrm{FPAR}}_{\mathrm{m}\mathrm{a}\mathrm{x}},\max ({\mathrm{FPAR}}_{\min },{\displaystyle \frac{{\mathrm{FPAR}}_{\mathrm{NDVI}}\left(\mathrm{x}, \mathrm{t}\right)+{\mathrm{FPAR}}_{\mathrm{S}\mathrm{R}}\left(\mathrm{x}, \mathrm{t}\right)}{2}}))$$

(7)

where ${\mathrm{FPAR}}_{\min }$ = 0.001 and ${\mathrm{FPAR}}_{\mathrm{m}\mathrm{a}\mathrm{x}}$ = 0.95 (Zhu et al. [30]). ${\mathrm{NDVI}}_{\min }$ and ${\mathrm{NDVI}}_{\max }$ were determined from the cropland NDVI distribution using the 5th and 95th percentiles of the multi-year cropland NDVI distribution (2001–2020) within the study area, respectively (Zhu et al. [30]). ${\mathrm{S}\mathrm{R}}_{\mathrm{m}\mathrm{i}\mathrm{n}}$ and ${\mathrm{S}\mathrm{R}}_{\max }$ were computed from ${\mathrm{NDVI}}_{\min }$ and ${\mathrm{NDVI}}_{\max }$ using Equation (4).

$$\mathsf{\epsilon }\left(\mathrm{x}, \mathrm{t}\right)=\mathrm{T}\mathsf{\epsilon }1\left(\mathrm{x}, \mathrm{t}\right)\times \mathrm{T}\mathsf{\epsilon }2\left(\mathrm{x}, \mathrm{t}\right)\times \mathrm{W}\mathsf{\epsilon }\left(\mathrm{x}, \mathrm{t}\right)\times \mathsf{\epsilon }*$$

(8)

where Tε1 and Tε2 are the low- and high-temperature stress scalars, respectively, and Wε is the moisture stress scalar. In this study, the standard CASA stress functions were implemented (full equations, variable definitions, and parameter settings are provided in Appendix A,

Table A3. Explicit formulations of the CASA stress scalars used in this study.

Component	Equation/Rule	Parameter Settings Used in This Study	Reference
Temperature stress 1	$\mathrm{T}\mathsf{\epsilon }1(\mathrm{x},\mathrm{t})=0.8+0.02 {\mathrm{T}}_{\mathrm{opt}}\left(\mathrm{x}\right)-0.0005{\mathrm{T}}_{\mathrm{opt}}{\left(\mathrm{x}\right)}^2$	${\mathrm{T}}_{\mathrm{opt}}\left(\mathrm{x}\right)$: temperature at which NDVI reaches its maximum during the growing season	Potter et al. [40]
Temperature stress 2	$\mathrm{T}\mathsf{\epsilon }2(\mathrm{x},\mathrm{t})={\displaystyle \frac{1.1814}{1+\exp \left(0.2\right({\mathrm{T}}_{\mathrm{opt}}\left(\mathrm{x}\right)-10-\mathrm{T}(\mathrm{x},\mathrm{t})\left)\right)}}\times$ ${\displaystyle \frac{1}{1+\exp \left(0.3\right({-\mathrm{T}}_{\mathrm{opt}}\left(\mathrm{x}\right)-10+\mathrm{T}(\mathrm{x},\mathrm{t})\left)\right)}}$	$\mathrm{T}(\mathrm{x},\mathrm{t})$: monthly mean air temperature (°C) at pixel x
Moisture stress	$\mathrm{W}\mathsf{\epsilon }(\mathrm{x},\mathrm{t})=0.5+0.5\times {\displaystyle \frac{\mathrm{EET}(\mathrm{x},\mathrm{t})}{\mathrm{PET}(\mathrm{x},\mathrm{t})}}$	EET/PET capped to [0, 1] before applying the formula
EET and PET	$\mathrm{Input} \mathrm{for} \mathrm{W}\mathsf{\epsilon }$	PET and EET were computed using the CASA water-balance submodel (Potter et al. [40]) driven by monthly temperature and precipitation.

Note: In CASA, EET and PET are computed by the water-balance submodel (Potter et al. [40]); the above scalars down-regulate ε* under non-optimal temperature and moisture conditions.

), using monthly mean temperature, precipitation, and potential evapotranspiration as inputs. The term ε* represents the maximum potential light-use efficiency under ideal conditions. In this study, the biome-specific ε* value for cropland was adopted from the calibrated results of Zhu et al. for Chinese ecosystems [30]. Specifically, we used the calibrated value for cropland: ε* = 0.542g C MJ⁻¹. CASA was implemented at the pixel level, and cropland NPP was extracted using the annual cropland mask from CLCD for each year (2001–2020). All input datasets, including monthly climate variables and NDVI, were resampled to a consistent spatial resolution of 250 m to ensure spatial consistency for pixel-wise CASA computation.

2.3.2. Quantifying the Urban Expansion Dynamics

To characterize the magnitude and speed of urban growth in the BTH region, we employed two standard indices: the Urban Expansion Area (UEA, km²) and the Urban Expansion Rate (UER, %) [42]. The values of these two indicators are calculated as follows:

$$\mathrm{UEA}={\mathrm{S}}_{t_2}-{ \mathrm{S}}_{t_1} , t_2>t_1$$

(9)

$$\mathrm{UER}=\left({\left({\displaystyle \frac{{\mathrm{S}}_{t_2}}{{\mathrm{S}}_{t_1}}}\right)}^{1/{(t}_2-t_1)}-1\right)\times 100\% , t_2>t_1$$

(10)

where ${ \mathrm{S}}_{t_1}$ and ${\mathrm{S}}_{t_2}$ denote the urban land area (km²) at times ${ t}_1$ and ${ t}_2$, respectively. UEA represents the absolute change in urban land area during the period, and UER represents the urban land growth rate expressed as an average annual percentage over the period, where $t_2-t_1$ is measured in years.

2.3.3. Measuring the Impact of Urban Expansion on Cropland NPP

We quantified the cumulative cropland NPP loss caused by urban expansion by improving the method developed by Milesi et al. [3]. To quantify the long-term opportunity cost of cropland conversion, we extended the Milesi et al. framework by introducing a temporal weighting factor to account for the duration of cropland displacement after conversion. To mitigate uncertainties associated with inter-annual climate variability and the “global greening” trend, we used the multi-year mean cropland NPP intensity during the baseline period (2001–2005) as a stable proxy for cropland productivity. The time-weighted cumulative cropland NPP loss over the entire study period (2001–2020) is calculated as follows:

$${\mathrm{CNPP}}_{\mathrm{loss}}={\sum }_{\mathrm{k}=1}^{\mathrm{M}}\left({\mathrm{W}}_{\mathrm{k}}\times {\sum }_{\mathrm{i}=1}^{\mathrm{N}}\left(\overline{{\mathrm{NPP}}_{\mathrm{i}}^{\mathrm{T}}}\times {\mathrm{A}}_{\mathrm{pix}}\times \left({\mathrm{Urban}}_{\mathrm{i},\mathrm{k}}^{{\mathrm{t}}_2}-{\mathrm{Urban}}_{\mathrm{i},\mathrm{k}}^{{\mathrm{t}}_1}\right)\right)\right)$$

(11)

$${\mathrm{W}}_{\mathrm{k}}={\mathrm{Y}}_{\mathrm{end}}-{\displaystyle \frac{{\mathrm{Y}}_{\mathrm{k}}^{\mathrm{t}1}+{\mathrm{Y}}_{\mathrm{k}}^{\mathrm{t}2}}{2}},{\mathrm{Y}}_{\mathrm{end}}=2020$$

(12)

where ${\mathrm{CNPP}}_{\mathrm{loss}}$ denotes the total time-weighted cumulative loss of cropland NPP over 2001–2020 (Tg C); i is the pixel index (i = 1, …, N); $\overline{{\mathrm{NPP}}_{\mathrm{i}}^{\mathrm{T}}}$ refers to the multi-year average NPP of the i pixel (g C·m⁻²·a⁻¹); ${\mathrm{A}}_{\mathrm{pix}}$ is the area of a pixel (m²). k indexes conversion stages and M = 4 in this study: 2001–2005, 2005–2010, 2010–2015, and 2015–2020. ${\mathrm{Urban}}_{\mathrm{i},\mathrm{k}}^{{\mathrm{t}}_1}$ and ${\mathrm{Urban}}_{\mathrm{i},\mathrm{k}}^{{\mathrm{t}}_2}$ are binary urban status variables (0/1) of pixel i at the start and end of stage k, respectively; thus ${\mathrm{Urban}}_{\mathrm{i},\mathrm{k}}^{{\mathrm{t}}_2}-{\mathrm{Urban}}_{\mathrm{i},\mathrm{k}}^{{\mathrm{t}}_1}=1$ indicates a cropland-to-urban conversion during stage k, and 0 otherwise. ${\mathrm{W}}_{\mathrm{k}}$ is the stage-level temporal weight defined as the remaining years of production absence from the midpoint of stage k to ${\mathrm{Y}}_{\mathrm{end}}=2020$. After multiplying ${\mathrm{NPP}}_{\mathrm{i}}^{\mathrm{T}}$ by ${\mathrm{A}}_{\mathrm{pix}}$ and ${\mathrm{W}}_{\mathrm{k}},$ the result expressed is g C, which is then converted to Tg C using 1 Tg = 10¹² g.

2.3.4. Assessing Ecological Selectivity of Urban Expansion Across Functional Zones

To quantify functional-zone differences in the ecological cost and selectivity of urban expansion on cropland productivity, we developed the Normalized Loss Efficiency (NLE) indicator. Unlike conventional absolute loss metrics, the NLE is designed to characterize the relative selectivity of urban expansion with respect to cropland quality, specifically evaluating whether urbanization disproportionately encroaches upon high-quality cropland relative to the regional productivity baseline. The NLE for a given functional zone z is defined as:

$${\mathrm{NLE}}_{\mathrm{z}}={\displaystyle \frac{{\mathrm{LE}}_{\mathrm{z}}}{{\mathrm{Q}}_{\mathrm{z}}}}={\displaystyle \frac{\frac{{\mathrm{CNPP}}_{\mathrm{loss},\mathrm{z}}}{{\mathrm{UEA}}_{\mathrm{z}}}}{{\overline{\mathrm{NPP}}}_{\mathrm{top}30\%}}}$$

(13)

where ${\mathrm{CNPP}}_{\mathrm{loss},\mathrm{z}}$ denotes the cumulative loss of net primary productivity (NPP) resulting from cropland-to-urban conversion within zone z, and ${\mathrm{UEA}}_{\mathrm{z}}$ represents the total area of newly expanded urban land during the study period. Their ratio, ${\mathrm{LE}}_{\mathrm{z}}$, reflects the loss efficiency (i.e., the realized ecological cost per unit urban expansion). To avoid mixing temporal scales, ${\mathrm{LE}}_{\mathrm{z}}$ is reported as a mean annual loss efficiency, calculated by dividing the study period loss per area (${\mathrm{CNPP}}_{\mathrm{loss},\mathrm{z}}$/${\mathrm{UEA}}_{\mathrm{z}}$) by the length of the study period, so that ${\mathrm{LE}}_{\mathrm{z}}$ has the same unit as ${\mathrm{Q}}_{\mathrm{z}}$ and ${\mathrm{NLE}}_{\mathrm{z}}$ is dimensionless. The term ${\mathrm{Q}}_{\mathrm{z}}$ serves as the regional cropland quality benchmark and is defined as the mean NPP value of the most productive cropland subset within zone z during the baseline period (2001–2005). NPP derived from remote sensing has been widely used as a proxy for cropland productivity and spatial variability, and high NPP values are commonly associated with prime cropland characterized by superior biomass production potential [43,44]. In this study, the top 30% of cropland NPP values within each zone are selected to represent high-quality cropland. The 30% threshold provides a practical balance between capturing the upper tail of cropland productivity and retaining a sufficiently large sample for stable estimation. Sensitivity tests using alternative thresholds (top 20% and 40%) yield robust NLE trends and inter-zonal ranking (

Table A4. Sensitivity analysis of the NLE benchmark to alternative percentile thresholds (Top 20%, 30%, and 40%).

Functional Zone	Valid Baseline Cropland Cells (n)	Qz20	Qz30	Qz40	NLE20 (2001–2020)	NLE30 (2001–2020)
CCF	465,362	457.37	441.21	428.28	0.723	0.749
ECD	382,070	452.92	435.53	422.09	0.742	0.772
NEC	335,284	462.10	434.27	411.17	0.647	0.688
SFE	499,306	463.85	449.57	437.48	0.774	0.799

Note: Q_z denotes the zone-specific productivity benchmark, defined as the mean baseline (2001–2005) cropland NPP of the top X% most productive cropland grid cells within each functional zone (X = 20, 30, 40); baseline nonpositive NPP values (≤0) were excluded prior to percentile selection. NLE values were recalculated under each threshold, and the results show that the inter-zonal pattern remains unchanged (SFE consistently highest; NEC consistently lowest). The maximum relative deviation of NLE from the 30% case is small (see “Max deviation (%)”), supporting the robustness of ecological selectivity patterns to threshold choice.

). This percentile-based approach allows the benchmark to capture the dominant productivity level of prime cropland while reducing sensitivity to extreme values and localized anomalies, thereby providing a robust reference for evaluating the selective encroachment of urban expansion.

As both the numerator and denominator share identical units (g C·m⁻²·a⁻¹), the NLE is a dimensionless indicator, enabling direct comparison across functional zones. An NLE value approaching or exceeding unity suggests a highly selective urban expansion pattern, in which urban growth preferentially occupies the most productive cropland. In contrast, lower NLE values indicate a less selective expansion pattern, characterized by urban development directed toward relatively marginal lands. The NLE index is evaluated at the county level rather than at a micro-grid scale to ensure a statistically sufficient expansion area (denominator), thereby avoiding undefined values (division by zero) and extreme outliers inherent in small-sample grid calculations.

2.3.5. Analysis of Driving Factors Using Geographically Weighted Regression

To capture the spatially varying associations between cropland NPP loss and its potential drivers, we employed Geographically Weighted Regression (GWR) [45,46]. Unlike global regression models (e.g., OLS) that assume constant relationships across space, GWR estimates location-specific parameters for each spatial unit, thereby allowing the effects of predictors to vary across space [47,48]. The analysis was conducted at a 5 km × 5 km grid scale to balance computational efficiency with the preservation of spatial details. We acknowledge that GWR results can be sensitive to the spatial support due to the modifiable areal unit problem (MAUP). The 5 km grid was selected as a pragmatic compromise that aligns with the granularity of macro-scale socioeconomic drivers while retaining sufficient spatial detail to capture regional heterogeneity; accordingly, all GWR-based interpretations in this study are framed at the 5 km scale and emphasize robust regional/subregional patterns rather than pixel-level micro-variations. The model is expressed as:

$${\mathrm{Y}}_{\mathrm{i}}={\mathsf{\beta }}_0\left({\mathrm{u}}_{\mathrm{i}},{\mathrm{v}}_{\mathrm{i}}\right)+{\sum }_{\mathrm{k}=1}^{\mathrm{p}}{\mathsf{\beta }}_{\mathrm{k}}\left({\mathrm{u}}_{\mathrm{i}},{\mathrm{v}}_{\mathrm{i}}\right){\mathrm{X}}_{\mathrm{i},\mathrm{k}}+{\mathsf{\epsilon }}_{\mathrm{i}}$$

(14)

where Y_i represents the sum of cropland NPP loss within the 5 km × 5 km grid cell i. Specifically, Y_i was obtained by summing pixel-level cropland NPP loss for all pixels falling inside grid i (Mg C per grid). (u_i,v_i) denotes the spatial coordinates; β_k(u_i,v_i) is the local regression coefficient for the k-th independent variable at location i; and ε_i is the error term. Five predictors were included (slope, DEM, distance, GDP, and population). All predictors were z-standardized; thus, each local coefficient ${\mathsf{\beta }}_{\mathrm{k}}\left({\mathrm{u}}_{\mathrm{i}},{\mathrm{v}}_{\mathrm{i}}\right)$ represents the expected change in Y_i associated with a one standard deviation increase in predictor ${\mathrm{X}}_{\mathrm{i},\mathrm{k}}$, holding other predictors constant. Multicollinearity was assessed using the variance inflation factor (VIF; all VIF < 7.5).

An adaptive bi-square kernel was adopted, and the bandwidth (neighbor number) was determined through a sensitivity analysis across multiple candidate neighbor sizes (N = 30, 40, 50, 60, 80, and 100), jointly considering model fit (AICc and adjusted R²) and numerical stability (

Table A6. Sensitivity of GWR bandwidth (neighbor number) and numerical stability.

Neighbors (N)	AICc	R2	Adj. R2	Sigma	ENP	Invalid Diagnostic Points (COND NoData)
30	5213.542	0.894	0.809	164.251	171.760	279 (≈4.571%)
40	7852.889	0.849	0.774	186.928	193.222	158 (≈2.589%)
50	10,565.915	0.792	0.713	211.920	210.152	13 (≈0.213%)
60	12,689.049	0.787	0.704	216.009	212.730	0
80	16,893.719	0.753	0.699	239.430	215.192	0
100	19,892.132	0.733	0.688	254.410	203.409	0

Note: Invalid diagnostic points (COND flagged as NoData) may occur under smaller bandwidths and were excluded only for coefficient mapping and zonal summaries. For the final setting (N = 50), the proportion of invalid points was 13/6103 = 0.213%. Global fit statistics (AICc and adjusted R²) were reported for the full sample.

). Although smaller bandwidths (e.g., N = 30, 40) yielded lower AICc, they also produced a higher proportion of invalid local diagnostics, indicating less stable local solutions. In contrast, N = 50 provided a robust balance between fit and stability, with only 13/6103 (0.213%) invalid diagnostic points; therefore, N = 50 was adopted for coefficient mapping and zonal summaries.

Multicollinearity was assessed using the variance inflation factor (VIF) based on the corresponding global OLS diagnostics. The VIF values ranged from 1.2864 to 2.1298 (VIF (min/median/max) = 1.2864/1.3959/2.1298), indicating low multicollinearity and a collinearity risk well below the commonly used threshold (Table A6). To evaluate whether spatial dependence was adequately accounted for, residual spatial autocorrelation of both OLS and GWR was assessed using Global Moran’s I under identical spatial weights (

Table A5. Diagnostic comparison between OLS and GWR and residual spatial autocorrelation.

Model	AICc	R2	Adj. R2	Residual Moran’s I	z-Score	p-Value
OLS	93150.396	0.257	0.256	0.3049	83.667	<0.001
GWR	10565.915	0.791	0.713	−0.0048	−1.271	0.204

Note: Residual Moran’s I tests whether model residuals remain spatially autocorrelated; a non-significant p-value (p > 0.05) indicates approximately spatially random residuals. Both tests used identical spatial weights (inverse-distance, Euclidean distance, row-standardized) with the ArcGIS default neighborhood search threshold of 18,029.5592 m (n = 6103).

). Finally, we cross-checked that the statements regarding areas with the highest marginal sensitivity in the Conclusions are consistent with the coefficient maps and the zonal summaries.

3. Results

3.1. Spatiotemporal Dynamics of Urban Expansion in the BTH Region

During the twenty-year period from 2001 to 2020, the Beijing-Tianjin-Hebei (BTH) region underwent a period of intense and rapid urbanization. Overall results indicate (

Table 1. Spatiotemporal dynamics of urban expansion and associated cropland NPP loss in different functional zones of the Beijing-Tianjin-Hebei region from 2001 to 2020.

Functional Zone	UEA (km2)			UER (%)			Cropland Contribution (%)	Total Cropland NPP Loss (Tg C)	Cumulative Loss Intensity (Mg C·km−2)
	2001–2010	2010–2020	2001–2020	2001–2010	2010–2020	2001–2020
CCF	2077.25	2050.94	4128.19	2.47	1.80	2.01	93.90	13.23	3444.64
ECD	1125.88	1272.38	2398.26	2.11	1.95	2.03	89.26	6.69	3299.16
NEC	426.00	489.19	915.19	3.63	2.99	3.30	62.98	1.63	2828.92
SFE	941.25	1511.00	2452.25	1.18	1.65	1.41	98.46	7.69	3186.65
BTH (Total)	4570.38	5323.50	9893.88	2.10	1.86	1.98	91.04	29.24	3301.22

Note: UEA, Urban Expansion Area; UER, Urban Expansion Rate. Cropland Contribution indicates the proportion of newly expanded urban land derived from cropland. Cumulative Loss Intensity represents the total accumulated NPP loss per unit of expanded urban area over the entire study period (2001–2020), calculated using a time-weighted cumulative model.

) that the area of urban building land in this region increased from 21,885.69 km² in 2001 to 31,779.56 km² in 2020, representing an overall increase of 45.2%. During this period, the cumulative urban expansion area index (UEA) reached 9893.88 km², with an urban expansion rate (UER) of 1.98%. From a temporal perspective (Figure 3), urban expansion exhibited an “acceleration followed by deceleration” trend: the expansion peak occurred during 2010–2015, with UEA reaching 3364.88 km² during this period, reflecting the strong demand for construction land driven by the region’s rapid economic growth. However, from 2015 to 2020, the expansion rate significantly decreased, with UEA falling to 1958.63 km². This deceleration is temporally consistent with the implementation of the Beijing-Tianjin-Hebei Coordinated Development Strategy, particularly the Outline of Coordinated Development of the Beijing-Tianjin-Hebei Region (2015) [49], which emphasizes coordinated spatial development and the orderly relief of Beijing’s non-capital functions.

From a spatial distribution perspective (Figure 4), driven by geographical location and development positioning, urban expansion across the four functional zones shows pronounced spatial disparity. The Central Core Functional Zone (CCF) serves as the focal point of urban expansion, boasting the highest UEA at 4128.19 km², accounting for 41.72% of the total expansion area in the BTH region. This indicates that Beijing and Tianjin, as large cities, maintain high-intensity urban expansion, with their powerful polarization effects continuing to drive the outward spread of urban boundaries. The Southern Functional Expansion Zone (SFE) and Eastern Coastal Development Zone (ECD) serve as secondary regional growth poles, with UEA of 2452.25 km² and 2398.25 km², respectively. The SFE exhibits dense plain-filling expansion anchored by transportation hub cities like Shijiazhuang and Handan. The ECD exhibits pronounced axial coastal extension driven by port-industrial zones like Caofeidian and Bohai New Area. Both zones demonstrate robust growth momentum. In contrast, the Northern Ecological Conservation Zone (NEC) faces the smallest expansion scale—only 915.19 km² UEA—due to complex mountainous terrain barriers like the Yanshan Mountains and the Taihang Mountains and constraints from ecological red line policies.

From the perspective of urban expansion patterns (Figure 4), the NEC region exhibits the smallest UEA, yet its urban expansion rate (UER) reaches 3.30%, the highest among the four major functional zones. This indicates that, despite constraints imposed by ecological red lines, land development intensity remains highly active around major towns within this region. The ECD and CCF regions follow with expansion rates of 2.03% and 2.01%, respectively, demonstrating that the coastal economic belt and the Beijing-Tianjin dual-core areas maintain steady growth momentum. Overall, these rate differences reflect distinct urban spatial expansion patterns across the four functional zones: CCF and SFE, situated in vast inland plains, exhibit “concentric” or “infill” expansion from existing built-up areas (Figure 3a), forming dense, contiguous urban patches. The ECD region, guided by coastlines and port transportation corridors, exhibits distinct “coastal axial” strip-like extensions. The NEC region, characterized by complex topography, faces constraints from natural geographical conditions, leading to expansion patterns along river valleys and foothills and resulting in a dispersed spatial structure.

3.2. Impact of Urbanization on Cropland NPP

3.2.1. Characteristics of Cropland Occupied by Urban Expansion

Urban expansion onto cropland is the primary direct driver of cropland NPP loss in the Beijing-Tianjin-Hebei region. From 2001 to 2020, 91.04% of newly developed urban land was converted from cropland (Table 1), and the proportion was higher in the Southern Functional Expansion Zone (SFE), reaching 98.46%. This indicates that urban growth in the SFE occurred predominantly through cropland conversion.

Figure 5b further illustrates this imbalance, with most cropland-to-urban transitions concentrated in the SFE and CCF. In addition, the NPP intensity of the occupied cropland shows an increasing trend over time. Specifically, the mean NPP intensity of converted cropland increased from 321.5 g C·m⁻²·a⁻¹ to 353.76 g C·m⁻²·a⁻¹ in 2015–2020 (Figure 5a). This pattern suggests that urban expansion increasingly occurred on relatively high-productivity cropland, particularly in plains with favorable site conditions, implying higher ecological costs per unit area of cropland conversion.

3.2.2. Temporal Dynamics of Cropland NPP Loss

Despite the continuous reduction in cultivated land area due to urban expansion, the productivity level of cultivated land ecosystems in the Beijing-Tianjin-Hebei (BTH) region shows a fluctuating but overall increasing trend. Figure 5a illustrates the trajectory of cropland NPP loss and cropland NPP. Between 2001 and 2020, the total cropland area across the BTH region decreased from 104,600 km² to 93,800 km², representing a 10.3% decline. However, total cropland NPP did not decrease with the reduction in area; instead, it increased from 33.29 Tg C to 38.93 Tg C. In parallel, the mean NPP intensity of cropland increased from 318.23 g C·m⁻²·a⁻¹ in 2001 to 415.24 g C·m⁻²·a⁻¹ in 2020. Under the CASA framework, this increase in NPP intensity reflects changes in modeled light-use efficiency driven by vegetation activity (NDVI/FPAR) and climate-related constraints (temperature and moisture stress). Meanwhile, improvements in agricultural management and technology (e.g., irrigation, fertilization, and crop varieties) may also contribute to enhanced productivity, although such management effects are not explicitly parameterized in the current model inputs. Therefore, the observed increase in productivity per unit area likely partially offsets the negative impacts of cropland area loss, helping maintain relatively stable regional cropland carbon sequestration capacity.

Despite sustained macro-level growth, the potential loss of NPP from cultivated land due to urbanization remains significant. Calculations reveal (Table 1) that between 2001 and 2020, the cumulative potential loss of NPP resulting from cultivated land conversion to construction land reached 29.24 Tg C—equivalent to 75% of the region’s actual total NPP from cultivated land in 2020. Hereafter, phase-wise “cumulative loss” refers to the time-weighted cumulative loss contribution defined in Section 2.3.3 (i.e., the annual loss rate multiplied by the temporal weight factor, W_k). An analysis by phase reveals distinct temporal variations in loss contributions (Figure 5a). The period 2005–2010 exhibited the highest loss contribution rate, with cumulative losses reaching 10.79 Tg C, accounting for 36.9% of total losses (Table 1). This resulted from the combined effects of higher cropland conversion rates and longer time accumulation weights during this phase. The 2001–2005 period contributed 30.37% (8.88 Tg C) of total losses. In contrast, although the 2010–2015 period experienced the most intense land conversion (3069.13 km² converted), its cumulative carbon loss contribution was lower (27.74%) due to its later occurrence.

With the significant slowdown in urban expansion rates, the loss of arable land NPP showed a declining trend between 2015 and 2020. The time-weighted cumulative loss contribution of this late-stage period amounted to only 1.46 Tg C (4.99%), while the average annual loss rate decreased to 0.584 Tg C·a⁻¹. This temporal dynamic indicates that the loss of arable land carbon sinks due to urban expansion in the BTH region was primarily concentrated in the initial rapid expansion phase of the first decade. This phase dominated the overall pattern of arable land NPP loss throughout the entire twenty-year period through its long-term cumulative effects.

3.2.3. Spatial Disparities of Cropland NPP Loss Across Functional Zones

In the spatial dimension, the time-weighted cumulative loss contribution of cropland NPP due to urban expansion showed significant differences among the four major functional zones (Figure 5c). The Central Core Functional Zone (CCF) was the primary contributor to regional carbon loss, with a cumulative loss of 13.23 Tg C, accounting for 45.25% of the total regional loss (Table 1). This pattern is associated with the large extent of cropland conversion in the region (3840.75 km²) and indicates continued cropland-to-urban transition driven by the outward expansion of core urban agglomerations.

In contrast, although the Southern Functional Expansion Zone (SFE) exhibited a lower total loss (7.69 Tg C, accounting for 26.30%) than the CCF, it demonstrated a higher loss intensity per unit area. Data indicate (Table 1 and Figure 5c) that the cumulative loss intensity in this region reached 3186.65 Mg C·km⁻² (where 1 Mg C = 10⁶ g C), second only to the CCF and significantly higher than the NEC. As shown in Figure 6a, the average NPP intensity of cultivated land occupied in the SFE region was 360.03 g C·m⁻²·a⁻¹, significantly higher than the regional average (342.56 g C·m⁻²·a⁻¹). This indicates that urban expansion in this area primarily occurred within high-productivity, premium cropland.

The Eastern Coastal Development Zone (ECD) experienced a net primary production (NPP) loss of 6.69 Tg C (22.88% of the total), with a loss intensity of 3299.16 Mg C·km⁻², falling between the Central Core Functional Zone (CCF) and Southern Functional Expansion Zone (SFE). The Northern Ecological Conservation Zone (NEC) exhibited the lowest loss level, with cumulative losses of only 1.63 Tg C (5.57% of the total). This is attributed to the region’s lower average NPP baseline (316.88 g C·m⁻²·a⁻¹) and smaller scale of cultivated land conversion (576.19 km²).

Overall, the spatial distribution of NPP loss in cropland across the Beijing-Tianjin-Hebei region exhibits a pattern characterized by “high total loss in the central area, high-intensity in the southern area, and low impact in the northern area.” Losses in the CCF are primarily driven by the scale of construction land expansion, while losses in the SFE are more strongly influenced by the high productivity attributes of the occupied cropland.

3.3. Spatiotemporal Evolution of Ecological Selectivity in Urban Expansion

To quantitatively assess the tendency of urban expansion to occupy high-quality cropland, this study employs the normalized loss efficiency index (NLE) to analyze the spatiotemporal evolution of ecological selectivity in urban expansion across four major functional zones. The NLE calculation benchmark (Q_z) is defined as the mean baseline (2001–2005) cropland NPP of the top 30% most productive cropland grid cells within each functional zone, representing the region’s highest cropland productivity level. An NLE value closer to 1.0 indicates that newly urbanized land tends to overlap with the most productive cropland, whereas an increasing NLE over time suggests intensified selectivity. The robustness of NLE trends to alternative percentile thresholds (top 20% and 40%) is reported in Table A4. Statistical results (

Table 2. Spatiotemporal evolution of the Normalized Loss Efficiency (NLE) index and related NPP benchmarks across functional zones.

Functional Zone	Quality Benchmark	Zone Cropland Mean NPP	Normalized Loss Efficiency (NLE)
	g C·m−2·a−1	g C·m−2·a−1	2001–2005	2005–2010	2010–2015	2015–2020
CCF	441.21	360.71	0.717	0.730	0.772	0.778
ECD	435.53	345.07	0.751	0.781	0.795	0.759
NEC	434.27	309.23	0.648	0.651	0.720	0.734
SFE	449.57	376.51	0.758	0.799	0.817	0.822

Note: Q_Z represents the quality benchmark, defined as the mean baseline (2001–2005) cropland NPP of the top 30% most productive cropland grid cells within each functional zone. NLE is dimensionless and is computed using stage-annualized loss efficiency, consistent with the stage columns in this table. An NLE value closer to 1.0 indicates a higher ecological selectivity of urban expansion towards high-quality cropland.

and Figure 7a) reveal substantial between-zone differences in the high-quality cropland benchmark (Q_z) across the four functional zones: the Southern Functional Expansion Zone (SFE) exhibits the highest Q_z value (449.57 g C·m⁻²·a⁻¹), confirming its status as the core grain production base in the Beijing-Tianjin-Hebei region; the Central Core Functional Zone (CCF) follows (441.21 g C·m⁻²·a⁻¹), while the Northern Ecological Conservation Zone (NEC) exhibits the lowest benchmark value (434.27 g C·m⁻²·a⁻¹). This benchmark disparity indicates that the ecological costs, measured as arable land loss per unit area, show pronounced functional-zone contrasts.

The spatiotemporal evolution of the NLE index (Figure 6b) reveals an overall upward trend in NLE values across all functional zones from 2001 to 2020, indicating an increasing tendency for urban expansion to occupy high-quality cropland over time. The SFE region exhibited the strongest ecological selectivity, with its NLE value steadily climbing from 0.758 at the outset to 0.822 by the end (Table 2). This implies that the mean annual NPP loss per unit of newly urbanized land in the SFE has reached 82.2% of the high-productivity benchmark defined for this zone. This sustained high-growth trend confirms that urbanization in the SFE region spatially overlaps significantly with premium cropland distribution, indicating persistent encroachment on high-yield fields. The CCF region also exhibited a steady rise in NLE, reaching 0.778 during 2015–2020, suggesting that expansion around the core area continues to favor the occupation of remaining high-yield agricultural land. In contrast, the NLE evolution in the ECD region exhibited a “rise-then-fall” pattern, peaking at 0.795 between 2010 and 2015 before declining to 0.759. This shift may be attributed to the later relocation of development focus toward low-yield coastal lands. The NEC region consistently maintained the lowest NLE level (<0.75), indicating that urban development in this area relatively avoided locally high-yield cropland due to topographical constraints. Overall, the NLE indicator reveals that the SFE region faces the most severe “spatial mismatch” risk, where the areas with the strongest urban development demand coincide precisely with those richest in high-quality arable land resources.

3.4. Determinants of Urban Expansion-Induced Cropland NPP Loss Patterns

To investigate the spatial pattern of cropland NPP loss and its association with potential drivers, we conducted spatial autocorrelation analyses at the 5 km × 5 km grid scale in ArcGIS. Global Moran’s I, computed using inverse-distance spatial weights with Euclidean distance, row standardization, and a fixed distance threshold of 15,000 m, indicates significant positive spatial autocorrelation in NPP loss (Moran’s I = 0.492, p < 0.001), suggesting pronounced spatial clustering. Local Indicators of Spatial Association (LISA; 999 permutations) further identify statistically significant local clusters at p < 0.05 (with p < 0.01 classified as ‘very significant’ in Figure 8a). The LISA cluster map (Figure 8b) shows High–High clusters (high-loss grids surrounded by high-loss neighbors) concentrated around the CCF and SFE plain urban agglomerations, whereas Low–Low clusters (low-loss grids surrounded by low-loss neighbors) are predominantly located in the northern mountainous areas of the NEC; High–Low/Low–High indicates local spatial outliers. Given that LISA involves multiple local tests, we interpret these patterns as indicative spatial structures rather than definitive boundaries. Notably, cropland NPP loss in this study is a direct consequence of cropland-to-urban conversion driven by urban expansion. Accordingly, the following ‘drivers’ are discussed as determinants of where and how urban expansion encroaches on cropland, thereby shaping the spatial heterogeneity of urban expansion-induced NPP loss, rather than as direct biophysical controls on NPP.

Given the significant spatial clustering of losses, this study further employs OLS as a global benchmark and incorporates GWR to explore spatially varying associations between drivers and losses. Model diagnostics reveal that GWR achieves significantly superior fit compared to OLS (GWR: Adjusted R² = 0.713; OLS: Adjusted R² = 0.257), with lower multicollinearity risk (VIF < 2.2 for all variables). Global Moran’s I tests on residuals under identical spatial weighting settings revealed significant spatial autocorrelation in OLS residuals, whereas GWR residuals showed no significant autocorrelation (Table A5). This indicates that GWR effectively reduces residual spatial dependence, making it more suitable for capturing spatial variations in driver effects within the study area. Balancing bandwidth sensitivity and numerical stability, an adaptive bandwidth of N = 50 was ultimately selected. A small number of invalid diagnostic points (13/6103 = 0.213%) were removed only during coefficient mapping and zoning statistics (Table A6). Thus, the GWR framework is used to explain the spatially varying associations between urban expansion determinants (pressure/suitability) and the observed pattern of cropland NPP loss.

The regression results of the GWR model indicate (Figure 9) that socioeconomic factors are the dominant determinants of the spatial heterogeneity of urban expansion-induced cropland NPP loss, reflecting stronger urban expansion pressure and a higher likelihood of cropland-encroachment in rapidly urbanizing areas. In over 74% of cropland units across the region, NPP loss exhibits a positive correlation with population growth and GDP growth, quantitatively confirming the trade-off between socioeconomic development and cropland conservation under rapid urbanization, consistent with socioeconomic forces driving differences in urban expansion intensity and location. To mitigate the impact of extreme outliers, this study employs the median to characterize differences in driving intensity across functional zones. Statistical results (

Table 3. Median values of local regression coefficients (GWR) for driving factors across different functional zones.

Functional Zone	Socio-Economic Drivers		Physical and Locational Constraints
	GDP	Population	Distance	Elevation	Slope
CCF	682.9	134.3	−92.1	−83.3	−136.2
ECD	2352.3	108.3	−58.3	−413.3	−183.2
NEC	2625.3	117.1	5.3	−99.3	−41.3
SFE	4283.6	137.4	−144.3	−260.1	−27.5
BTH (Total)	2486	124.3	−72.4	−214	−97.1

Note: Dependent variable Yi is the sum of pixel-level cropland NPP loss within each 5 km × 5 km grid (unit: Mg C per grid; 1 Mg = 10⁶ g). Predictors are z-standardized; thus, coefficients indicate the change in Yi (Mg C per grid) associated with a 1 SD increase in the predictor (holding other variables constant).

) reveal that the SFE exhibits the highest ecological sensitivity to economic activity, with a median GDP regression coefficient of 4284—significantly higher than that of the CCF (682.9). This indicates that in the Southern Plains, equivalent units of economic growth lead to significantly greater losses in cropland productivity compared to the Core Area, reflecting high marginal ecological costs. In contrast, despite having the largest economic scale, the Core Area exhibits the lowest GDP coefficient across all zones, suggesting that its economic growth may rely less on expanding new construction land.

Natural geographical conditions and locational constraints shape the pathways and forms of urban expansion, thereby indirectly influencing where cropland is converted and where urban expansion-induced NPP loss concentrates. The median slope coefficient across the entire region is negative (−97.1), indicating terrain’s blocking effect on construction activities (Table 3). This effect is most pronounced in the ECD zone (Slope = −183; Elevation = −413), suggesting that coastal lowland development faces stronger constraints from micro-topography. In contrast, the slope coefficient for the SFE exhibits an anomalous pattern (−28), approaching zero with locally positive values (Figure 9e). This result reflects that, under the micro-topography of the North China Plain, relatively elevated areas (such as river embankments and uplands) may be prioritized for development, leading to increased cropland occupation in these zones. Regarding location factors, the SFE exhibits the strongest negative distance coefficient (−144) across the entire region (Figure 9c), indicating that southern plain cities expand closer to existing urban centers in a concentric pattern. The positive distance coefficient (5) in the NEC region suggests that development in this area is more dispersed due to terrain fragmentation and relies less on central cities.

4. Discussion

4.1. Evaluation of NPP Simulation Results

Simulating ecosystem productivity in highly heterogeneous urbanized areas often requires model results to undergo plausibility checks using external references. Given the difficulty in obtaining long-term, continuous ground-based observation data for large-scale regions, this study employs a consistency-based approach. It compares pixel-level data between CASA-derived cropland NPP and the internationally recognized MODIS NPP product (MOD17A3HGF) to assess their consistency in temporal variation and spatial distribution statistics. The MOD17A3 series has been widely applied in various global and regional ecological studies, often serving as a reference product for regional-scale NPP assessments [50,51]. It is well acknowledged that the MODIS product is also a model-derived estimate and shares similar meteorological drivers and land cover inputs with the CASA model. Therefore, this cross-validation is not fully independent, and MODIS serves primarily as a reference for assessing spatiotemporal consistency rather than an absolute “true” value. Intercomparison results demonstrate high temporal consistency between CASA simulations and the MODIS MOD17A3HGF product. Regression analysis based on 2000 randomly sampled points indicates significant positive correlations across all five evaluation years (2001, 2005, 2010, 2015, and 2020). The overall R² for the entire period reached 0.735, with annual R² values consistently above 0.72. This statistical outcome confirms good consistency between the two products in interannual variability and spatial statistical characteristics. To further objectively assess systematic biases, comprehensive error statistics were evaluated. The overall Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) were 94.89 and 84.72 g C·m⁻²·a⁻¹, respectively, with an overall linear regression equation of y = 0.479x + 234.89. A systematic positive bias of 82.55 g C·m⁻²·a⁻¹ was observed. Furthermore, the consistency between CASA and MODIS estimations remains robust across different topographical features, including both the plain agricultural zones and the mountainous areas.

Notably, Figure 10 shows that CASA model estimates are generally slightly higher than MODIS product values. This systematic difference primarily stems from scale effects related to spatial resolution. Constrained by its 500 m resolution, the MODIS product is more prone to detecting mixed pixels of impervious surfaces, bare land, and vegetation in urban fringe transition zones and fragmented cropland areas, thereby lowering overall NPP values. In contrast, the 250 m resolution CASA model employed in this study enables finer identification of high-yield cropland patches embedded within the urban matrix, effectively mitigating this underestimation bias. Indeed, prior evaluations have suggested that MODIS NPP products may be slightly lower than ground-based observations while remaining closely correlated [52]. In contrast, the 250 m CASA implementation, combined with the annual cropland mask better represents cropland productivity patterns required by this study, supporting subsequent estimation of cropland NPP losses. Finally, given the extreme scarcity of long-term continuous ground-based observation data for large-scale regions, we also compared our simulated CASA NPP results with existing literature calculations for the BTH region (

Table A7. Cross-comparison of the simulated mean cropland net primary productivity (NPP) with existing literature in the Beijing-Tianjin-Hebei (BTH) region and adjacent areas.

Neighbors (N)	Study Area	Period	Method	Mean Cropland NPP (g C m−2 a−1)
This study	BTH Region	2001–2020	CASA	375.15
Zou et al. (2022) [66]	BTH Region	2001–2020	MODIS	386.57
Chen et al. (2024) [34]	Beijing	2000–2020	CASA	342.14
Long et al. (2024) [67]	North China Plain	2013–2017	DNDC	510.00

Note: Due to the severe scarcity of continuous, long-term field-measured NPP data specifically covering croplands across the entire study area, direct large-scale site validation remains highly challenging. This cross-comparison with independent modeling and field-based assessments from the existing literature is provided to further validate the magnitude and reliability of our CASA model estimations. The NPP values represent the multi-year average for croplands.

). The mean cropland NPP estimated in this study aligns well with previous independent assessments, providing further validation of our magnitude estimations.

4.2. Ecological Selectivity of Urban Expansion and Food Security Impact

The impact of urban expansion on cropland extends beyond mere area reduction to encompass the preferential conversion of high-productivity cropland, as indicated by the NLE results. Across the entire BTH region, urbanization processes consistently exhibit a pronounced preference for premium cropland, with new construction land tending to concentrate in productive plains with higher baseline productivity. However, the strength and direction of this selectivity can vary substantially across functional zones, as suggested by prior evidence that the impacts and determinants of urban expansion differ with geographic settings, urban size, economic development levels, and policy contexts [27,53]. Unlike the ECD region, where the NLE index declined due to coastal saline–alkali wasteland development, and the NEC region, constrained by topography, maintaining low levels, the SFE region exhibits the strongest combination of high conversion pressure and high ecological selectivity. This area not only bears the overwhelming majority of the region’s arable land conversion pressure but also features occupied cropland with productivity benchmarks far exceeding the regional average. This high-intensity targeted selection implies that urbanization in the SFE involves a persistent and difficult-to-offset conversion of high-productivity cropland. From a land use efficiency perspective, this expansion model incurs substantial ecological opportunity costs. Permanently sealed soil resources often possess irreplaceable agricultural attributes, and their loss cannot be effectively compensated by merely increasing the cultivation area of marginal lands [29,54,55]. Moreover, urban land expansion is closely linked to multi-level socioeconomic and institutional drivers, implying that these opportunity costs are also shaped by broader development transitions and governance contexts [56]. This selectivity pattern is robust to alternative percentile thresholds used to define high-productivity cropland (Table A4).

This high ecological opportunity cost directly translates into potential risks for regional food security systems. Food production is intrinsically linked to national well-being [57,58]. Although advances in modern agricultural technology have temporarily boosted crop yields per unit area, partially masking the food production gap caused by arable land loss, such technological compensation cannot fundamentally reverse the long-term ecological fragility stemming from declining cultivated land quality (potential land productivity) [28,59,60]. Our findings reveal that the BTH region has not only lost vast quantities of cropland to urbanization, but that much of this lost land consists of high-yield fields with superior soil and water conditions. The loss of these premium agricultural resources fundamentally weakens the ecological resilience of the regional agricultural ecosystem. Compared to marginal lands more susceptible to climate fluctuations, these high-yield fields serve as the “ballast” safeguarding national food security. Once permanently converted into impervious urban surfaces, the marginal benefit of future yield increases through technological inputs will sharply diminish. This renders regional ecosystems more vulnerable to extreme climate events and complicates the achievement of long-term food security [61,62,63]. These findings indicate that the current cropland compensation policy, centered on area balance, has limitations in maintaining regional production stability. There is an urgent need to shift toward integrated management that prioritizes quality and ecological functions.

4.3. Differentiated Policy Implications for Functional Zones

The GWR analysis reveals distinct spatial non-stationarity in how socioeconomic, physical, and locational factors shape urban expansion pressure and its cropland-encroachment pattern, thereby explaining the spatial heterogeneity of urban expansion-induced cropland NPP loss across functional zones. This indicates that uniform land management policies struggle to accommodate the complex landscape of the Beijing-Tianjin-Hebei region, necessitating differentiated spatial control strategies tailored to functional zoning units [64,65].

For the Southern Functional Expansion Zone (SFE), which exhibits high economic sensitivity and a preference for occupying micro-topography within the plains, policy priorities should focus on curbing the disorderly sprawl of construction land and strictly safeguarding high-quality cropland. It is recommended to strictly delineate urban growth boundaries and permanent basic cropland protection red lines in this area, implementing stricter avoidance and control measures for high-potential upland fields and cropland near river embankments. This will guide incremental development toward optimizing existing stock and low-ecological-cost spaces. Although the Northern Ecological Conservation Zone (NEC) has a smaller urban scale, it exhibits greater marginal responsiveness to economic and population growth. Therefore, regulatory focus should prioritize strengthening access controls and development intensity constraints for construction land to prevent fragmented, scattered urban land development from compromising the integrity of high-quality cropland resources. For the Eastern Coastal Development Zone (ECD), leveraging its abundant coastal saline–alkali land resources can guide new construction land toward low-yield fields or non-agricultural areas, thereby alleviating pressure on high-quality cropland. Overall, the Central Core Functional Zone (CCF) exhibits a trend of relative decoupling between economic growth and land expansion, offering a reference for the entire Beijing-Tianjin-Hebei region. Future urbanization in the BTH region should reduce excessive reliance on new land resources and instead prioritize optimizing the use of existing construction land. Revitalizing inefficient industrial land and renovating old urban districts can increase economic output per unit of land area, thereby reducing the occupation of surrounding cropland. Simultaneously, planning for newly developed cropland should avoid converting high-quality ecological land into agricultural use whenever possible, thereby mitigating the long-term ecological risks associated with “replacing high-quality land with low-quality land.” Achieving sustainable development in the Beijing-Tianjin-Hebei region hinges on balancing the trade-off between high-intensity economic growth and cropland protection through innovative regional governance mechanisms, ultimately realizing the goal of coordinated regional development.

4.4. Limitations and Future Perspectives

This study employs a multidimensional indicator approach to analyze the spatiotemporal characteristics of urban expansion and cropland NPP loss in the Beijing-Tianjin-Hebei region. Several limitations should be noted. First, we primarily quantify direct NPP losses caused by land-cover conversion, while indirect ecological effects associated with urbanization are not explicitly considered. For example, the urban heat island may alter vegetation phenology, and changes in urban rainfall patterns could indirectly influence cropland productivity [34]. Second, the spatially varying effects of driving factors were analyzed at a 5 km grid scale; although this choice balances computational efficiency and spatial detail, it may obscure micro-scale features. We also note that local coefficient magnitudes in GWR can be scale-dependent; therefore, our conclusions regarding spatially varying associations are explicitly interpreted at the 5 km grid support. Third, while cropland NPP is a key indicator of productivity potential, it does not directly translate into grain yield because harvest indices vary across crop types (e.g., wheat and maize). In addition, uncertainty arising from resolution harmonization should be acknowledged. Key CASA drivers have coarser native resolutions (e.g., radiation at 0.1° and meteorology at 1 km) than the 250 m analysis resolution. Resampling these drivers to 250 m using the NEAREST method preserves original pixel values but yields a block-wise representation and may under-represent sub-grid energy/moisture gradients within a coarse pixel. Uncertainty also arises from the land-cover input (CLCD) used to delineate cropland and urban land. CLCD reports an overall accuracy of 79.31%, with year-specific overall accuracies ranging from 76.45% to 82.51% (Yang and Huang [36]), implying that cropland–built-up confusion may locally over- or under-estimate cropland-to-urban conversion, particularly along fragmented urban–rural boundaries. Similarly, upscaling CLCD land cover from 30 m to 250 m using the MAJORITY rule may generalize heterogeneity near cropland–noncropland boundaries. The cropland-mask sensitivity diagnostics (Appendix A Table A2) suggest that annual 250 m cropland extent varies within a limited range under alternative fractional thresholds, indicating that most discrepancies are concentrated in boundary areas. Therefore, we emphasize robust regional/subregional patterns and stage-wise comparisons, while treating fine-scale pixel variability and absolute magnitudes as subject to scale-related uncertainty.

Future research will therefore focus on two directions. First, we will incorporate ecosystem process-based simulations to help disentangle the “land-use effect” from the “environmental effect” of urban expansion. We will also conduct multi-scale analyses of spatially varying driver effects to explore potential response thresholds for cropland protection policies across spatial resolutions. Second, by integrating high-resolution crop distribution maps with agricultural statistics, we will develop conversion models linking NPP to grain yield, enabling more intuitive and policy-relevant assessments of urban expansion impacts on regional food security.

5. Conclusions

This study employs the CASA model and high-resolution land-use data to construct a comprehensive assessment framework. It systematically reveals the complex impacts of urban expansion in the Beijing-Tianjin-Hebei (BTH) urban cluster on arable land NPP from 2001 to 2020, examining three dimensions: spatiotemporal evolution, ecological selectivity, and the interplay of driving factors. Results indicate that over the past two decades, the Beijing-Tianjin-Hebei region underwent rapid urbanization, with large-scale expansion of construction land primarily achieved at the expense of cropland. Although the regional mean NPP of cropland fluctuated upward over time, this trend failed to mask the substantial cumulative ecological costs incurred by urban expansion. Calculations reveal a cumulative loss of 29.24 Tg C in arable land NPP, exhibiting a pronounced spatial divergence between total loss and intensity: the Central Core Functional Zone (CCF) dominated total loss, while the Southern Functional Expansion Zone (SFE) showed the highest loss intensity per unit area. Crucially, this expansion demonstrated significant ecological selectivity. The NLE index in the SFE region steadily climbed to 0.822, revealing the urbanization process’s prioritization of high-productivity, premium cropland. This intense “reverse selection” led to a structural decline in regional agricultural production potential, exposing the southern plains—the core grain-producing region of North China—to the most severe ecological opportunity costs and food security risks. Furthermore, GWR analysis indicates spatial non-stationarity in the relationship between drivers and arable land NPP loss: socioeconomic factors exhibit the highest marginal sensitivity in northern ecologically fragile zones, while natural location factors show anomalous micro-topography occupation preferences in SFE regions.

In summary, this study not only quantifies the cumulative negative effects of urbanization on arable land carbon sinks but also reveals the latent threats to regional food security from the perspectives of arable land quality and spatially varying driver effects. By proposing differentiated management strategies tailored to the spatial characteristics of distinct functional zones, this research provides crucial scientific evidence and decision-making references for optimizing territorial spatial patterns and achieving coordinated development of food security and urban clusters within the context of Beijing-Tianjin-Hebei coordinated development.