Dissecting the “Black Box” of Agricultural Green Total Factor Productivity: An Analysis Using a Network SBM Model Based on Eight Consecutive Years of Soil Data

Abstract

Agricultural green total factor productivity (AGTFP) is a key indicator for evaluating the level of green development in agriculture. However, conventional approaches to AGTFP measurement often treat intermediate agricultural production processes as a “black box”, overlooking internal system mechanisms and thus leading to biased identification of efficiency bottlenecks. To address this limitation, this study introduces an analysis using a network slack-based measure (NSBM) model to move beyond the traditional single “input–output” transmission framework. By integrating soil sample data from 2017 to 2024, the agricultural production process is decomposed into three sequential stages—material inputs, nutrient transformation, and crop production—to evaluate AGTFP in Baoding, China. The results reveal that AGTFP in Baoding remains at a relatively low level overall, although a steady upward trend is observed over time. Specifically, overall efficiency increased by 18.18%, while the efficiency of converting agricultural inputs into soil nutrients (the input subsystem) improved by 36.13%. The input subsystem serves as the primary driver of AGTFP improvement, with a marginal contribution coefficient of 0.61% to overall efficiency (p < 0.01). Although the efficiency of transforming soil nutrients into agricultural output (the output subsystem) remains relatively high, its growth potential is constrained by biological limits. It is highly sensitive to external factors such as topography and natural disasters. At present, the key bottleneck to enhancing regional AGTFP is the low efficiency with which external inputs are converted into soil nutrients. These findings suggest that policy priorities should shift from simple input reduction to process-oriented management, with an emphasis on improving the conversion efficiency of external inputs into effective soil nutrients, thereby facilitating agricultural green transformation while mitigating non-point source pollution. Based on existing research frameworks, this study supplements and refines the original analytical framework by incorporating soil data from the agricultural production process, providing new empirical evidence for uncovering the “black box” of agricultural production.

1. Introduction

Intensifying global climate change and increasingly stringent resource and environmental constraints have made the coordinated safeguarding of food security and agricultural ecosystems a major challenge facing human society [1]. At present, China’s 15th Five-Year Plan identifies the steady improvement of total factor productivity as a key objective for economic and social development during this period. However, the intensification of agricultural non-point source pollution and ecological degradation has become increasingly pronounced in China [2], placing the agricultural sector at a critical transition stage from a “high-output” model to one emphasizing “high-quality” development [3]. As a key indicator for characterizing the coordination between food production and ecological sustainability [4], agricultural green total factor productivity (AGTFP) systematically captures the integrated efficiency of resource inputs, environmental constraints, and economic outputs in agricultural production, and serves as an important tool for enhancing the quality of green agricultural development [5]. However, although AGTFP integrates resource inputs, environmental constraints, and economic outputs, the soil-mediated nutrient transformation processes that connect agricultural inputs with crop production remain insufficiently represented in existing AGTFP evaluation frameworks.

In the measurement of AGTFP, data envelopment analysis (DEA) and stochastic frontier analysis (SFA) are the two most adopted approaches [6]. Studies employing DEA models primarily rely on the CCR model [7] and the slacks-based measure (SBM) model [8], both of which have been extensively applied to efficiency evaluation across various economic sectors [9]. However, these models generally treat the agricultural production system as an integrated “black box”, capturing only the static relationship between input factors and final outputs, while overlooking the multi-sectoral and multi-stage material flows and transformation processes within agricultural production. As a result, they are limited in revealing heterogeneity in internal resource allocation efficiency and its underlying constraints. To overcome the limitations imposed by the “black box” assumption, Tone and Tsutsui [10] proposed the network slacks-based measure (NSBM) model, which allows a decision-making unit to be decomposed into multiple sub-sectors linked by intermediate outputs. This framework enables the simultaneous evaluation of overall efficiency and subsystem-specific efficiencies, substantially improving the accuracy and interpretability of efficiency measurement [6]. To date, the NSBM model has been applied in fields such as industrial ecological efficiency [10] and energy system assessment [11]. However, its application to micro-level analysis of agricultural production processes remains limited, particularly in the explicit representation of internal biochemical processes, where the micro-level refers to the soil–crop system, including soil nutrient transformation, biochemical processes, and crop nutrient uptake during agricultural production.

Existing studies have documented pronounced regional disparities in AGTFP across China [12,13,14]. A range of econometric and spatial analytical approaches have been employed to further investigate the determinants of these regional disparities, including Tobit panel regression models [15], spatial Durbin models [16], and geographic detector models [17]. These studies indicate that spatial misallocation of resources plays a critical role in shaping AGTFP. Alleviating spatial resource misallocation can effectively enhance the quality of green agricultural development [18]. Persistent misallocation tends to cause losses in total factor productivity and exacerbate ecological degradation [19]. Research focusing on farm size suggests that the expansion of agricultural operating scale contributes positively to AGTFP improvement [19,20]. Some panel regression-based studies have found that increasing fiscal support for agriculture does not necessarily promote agricultural gross output value [21]. Subsequent studies employing spatial Durbin models further reveal that the effects of fiscal support for agriculture on AGTFP exhibit significant regional heterogeneity [22]. The impacts of environmental regulation on AGTFP also display pronounced regional heterogeneity [23], and such regulations may positively influence AGTFP through mechanisms such as resource reallocation and technological innovation [5].

Based on analyses of constraining factors, previous studies have proposed several policy-relevant strategies for improving AGTFP, including leveraging spatial spillover effects [24], promoting multi-factor coordination [25], and adjusting agricultural input structures in a region-specific manner [26]. Collectively, these studies provide important insights for enhancing AGTFP and advancing high-quality green agricultural development. Soil is not merely a passive storage medium but an active regulator of agricultural productivity and environmental sustainability. Through nutrient retention, mineralization, microbial transformation, and nutrient supply, soil governs fertilizer use efficiency, crop nutrient uptake, and nutrient losses. However, its role has been largely overlooked in existing AGTFP studies. Nevertheless, owing to the substantial challenges associated with large-scale continuous soil sampling and the acquisition of long-term experimental data, the mediating role of soil—linking external agricultural inputs to crop outputs [27]—has not yet been adequately examined in existing studies.

To address the aforementioned research gaps, this study utilizes continuous soil nutrient data from 2017 to 2024 in the Baoding district and, based on previous studies, constructs a novel AGTFP evaluation framework incorporating intermediate agricultural production processes, namely “material inputs–nutrient transformation–crop production,” using the NSBM model. Specifically, the transformation of material inputs into soil nutrients is defined as the input subsystem, while the conversion of soil nutrients into agricultural outputs is classified as the output subsystem. The study estimates the efficiency of the input subsystem, the efficiency of the output subsystem, and the overall system efficiency by jointly considering agricultural inputs, intermediate products, and final outputs. Furthermore, the spatiotemporal evolution and driving mechanisms of these three efficiency measures are systematically analyzed.

This study aims to (i) construct an AGTFP evaluation framework that reflects the efficiency of intermediate processes in agricultural production, thereby deconstructing the inherent “black box” of traditional AGTFP assessments; (ii) compare the dynamic evolution and underlying mechanisms of subsystem efficiencies and overall system efficiency, in order to identify the key processes constraining AGTFP improvement; and (iii) propose targeted pathways for high-quality green agricultural development based on spatiotemporal patterns and multidimensional driving factor analysis, thereby providing scientific evidence to support the coordinated achievement of food security and ecological sustainability.

2. Data and Methods

2.1. Study Area

The study area was the Baoding district, located in central Hebei Province, China (Figure 1). The northwestern part of the city is dominated by mountainous and hilly terrain, and the southeastern part consists mainly of plains. Overall, the topography slopes from northwest to southeast, exhibiting a diverse range of geomorphological features. Baoding is endowed with abundant agricultural resources. According to statistics from the Baoding Municipal Bureau of Agriculture and Rural Affairs, the total grain-sown area reached 675,333 ha in 2024, with a total grain output of 4.211 million tons, making the city a representative major grain-producing region in North China. Meanwhile, Baoding also serves as a core city within the Beijing–Tianjin–Hebei world-class urban agglomeration [28]. The dominant soil types in Baoding include cinnamon soils, fluvo-aquic soils, and sandy ginger black soils. Overall, soil fertility is moderate, characterized by relatively high phosphorus and potassium contents but comparatively low levels of organic matter and nitrogen. Owing to the establishment of the Xiongan New Area from Xiong County, Anxin County, and Rongcheng County, as well as the separate administrative management of Dingzhou City at the provincial level, consistent statistical data for these areas are unavailable. In addition, Jingxiu District and Lianchi District constitute the urban core of Baoding and contain very limited cropland. Therefore, this study selects the remaining 18 counties and districts of Baoding district as the research area.

2.2. Data Sources

2.2.1. Input–Intermediate–Output Indicator System

This study selects chemical fertilizers, pesticides, plastic mulch, and cultivated land area as input variables [29,30], while grain output, non-point source pollution, and carbon emissions are treated as output variables [28]. In accordance with the transmission and transformation mechanisms of input–output factors within the agricultural production process [31,32], soil nutrient indicators—including total nitrogen [33,34,35,36], available phosphorus [37], available potassium [38], and soil organic matter [39]—are selected as intermediate variables for AGTFP assessment. The complete indicator system is presented in

Table 1. Indicator system of input, intermediate, and output variables.

Variable Category		Variable	Unit	Data Source
Input variables		Chemical fertilizer	t	Hebei Rural Statistical Yearbook
		Pesticides	t
		Agricultural plastic film	t
		Agricultural diesel fuel	t
		Cultivated land area	ha
Output variables	Desirable output	Grain yield	t
	Undesirable outputs	Carbon emissions	t	Calculated in this study
		Non-point source pollution	t
Intermediate variables		Soil organic matter	g kg−1	Field sampling and laboratory analysis
		Total nitrogen	g kg−1
		Available phosphorus	mg kg−1
		Available potassium	mg kg−1

.

Data on input variables and grain yield were obtained from the Hebei Rural Statistical Yearbook. Carbon emissions were calculated using the emission factors listed in

Table 2. Carbon emission factors.

Carbon Source	Emission Factor	Reference
Fertilizer	0.8956	West and Marland [40]
Pesticide	4.9341	Oak Ridge National Laboratory [19]
Plastic Film	5.18	Institute of Agricultural Resources and Environmental Sciences, Nanjing Agricultural University [28]

. Non-point source pollution was estimated following previous studies [19] based on the application rates of chemical fertilizers, pesticides, and plastic mulch. Intermediate variables were derived from laboratory analyses of soil samples.

2.2.2. Soil Nutrient Data

Soil nutrient data were primarily obtained from the long-term fixed-site monitoring program conducted by the Baoding Soil and Fertilizer Station, Hebei Province, supplemented by datasets collected from our previous research projects. A total of 221 permanent soil sampling sites were included in this study, and their spatial distribution is shown in Figure 2. Soil samples were collected annually from 30 September to 15 October, following maize harvest and prior to winter wheat sowing and fertilization, during the period of 2017–2024. At each site, composite soil samples were collected from the 0–20 cm plough layer using an S-shaped sampling method. According to the World Reference Base for Soil Resources (WRB 2022), the dominant soil groups in the study area are Gleyic Luvisols, Calcaric Fluvisols, and Calcaric Cambisols. The physicochemical properties of the sampled soils are presented in

Table 3. Soil physicochemical properties.

Property	Unit	Mean	SD	Max	Min
Plough layer thickness	cm	17.03653846	2.009977884	23	12
Bulk density	g cm−3	1.395769231	0.092452611	1.59	1.13
Sampling depth	cm	20	0	20	20
pH	-	7.871153846	0.271984543	8.5	7.3
Soil organic matter (SOM)	g kg−1	18.8375	5.4926797	32.5	2.79
Total nitrogen (TN)	g kg−1	1.183846154	0.335972491	1.89	0.23
Alkali-hydrolyzable nitrogen (AN)	mg kg−1	94.55769231	35.79128249	189	13
Available phosphorus (AP)	mg kg−1	35.55384615	28.17294434	127.7	4.9
Available potassium (AK)	mg kg−1	183.5	84.51592931	458	57
Slowly available potassium (SAK)	mg kg−1	971.5	312.4777223	1986	407

. Laboratory analyses were conducted to determine soil organic matter (SOM), total nitrogen (TN), alkali-hydrolyzable nitrogen (AN), available phosphorus (AP), and available potassium (AK) following standard analytical procedures described in previous studies.

2.2.3. Driving Factor Indicator System

Considering the spatial heterogeneity of AGTFP, this study constructs a comprehensive indicator system encompassing four dimensions: socio-economic level, agricultural production conditions, natural geographic conditions, and transportation accessibility. A total of 11 indicators were selected, as presented in Table 3. The socio-economic dimension includes regional gross domestic product (GDP) [41], the proportion of primary industry [9], and rural per capita disposable income [42], representing the influence of regional economic development on green agricultural production. The agricultural production conditions dimension comprises total power of agricultural machinery [43], area under facility agriculture (X5), and public budget expenditures for agriculture, forestry, and water [34], measuring agricultural productivity and the level of fiscal support for agriculture. The natural geographic conditions dimension includes temperature [42], elevation [41], and disaster-related losses [43], representing the constraints of environmental conditions and natural hazards on agricultural production. Transportation accessibility is represented by road network density [44], reflecting regional transportation convenience and the efficiency of agricultural product circulation.

Table 4. Indicator system of driving factors.

Observation Dimension	Influencing Factor	Indicator (Unit)	Code
Socioeconomic level	Regional economic development	GDP (CNY108)	X1
	Agricultural industrial structure	Share of the primary industry (%)	X2
	Endogenous development capacity	Per capita disposable income of rural residents (CNY)	X3
Agricultural production conditions	Level of agricultural mechanization	Total power of agricultural machinery (kW)	X4
	Level of agricultural facilities	Area under facility-based agriculture (ha)	X5
	Fiscal support for agriculture	Expenditure on agriculture, forestry, and water conservancy from general public budget (CNY 108)	X6
Natural and geographical conditions	Climatic conditions	Mean temperature (°C)	X7
	Topography	Elevation (m)	X8
	Natural disasters	Disaster-induced losses (CNY108)	X9
Transportation accessibility	Transport accessibility	Road network density (km·km−2)	X10

summarizes the indicator system used to analyze the spatial heterogeneity of AGTFP.

2.3. Methods

2.3.1. Network Slack-Based Measure (NSBM) Model

This study employs the Network Slack-Based Measure (NSBM) model proposed by Tone and Tsutsui [10] to evaluate agricultural green total factor productivity (AGTFP). All NSBM calculations were implemented using MATLAB (v2023b).Unlike conventional DEA models, which treat agricultural production as a single “black box”, the NSBM model explicitly represents intermediate production processes by decomposing the production system into interconnected subunits. This enables the simultaneous evaluation of overall system efficiency and subsystem-specific efficiencies.

In this study, the agricultural production process is conceptualized as “material input—soil nutrient transformation—crop production”, with soil nutrients serving as intermediate products linking external agricultural inputs to crop outputs. Throughout this study, the material input–soil nutrient transformation stage is referred to as the input subsystem, whereas the soil nutrient transformation–crop production stage is referred to as the output subsystem. The production system is divided into two interconnected subunits. The input subsystem evaluates the efficiency with which agricultural inputs are converted into soil nutrients, reflecting the soil’s capacity for nutrient retention and transformation. The output subsystem measures the efficiency of converting soil nutrients into desired agricultural outputs while accounting for undesired environmental outputs, representing crop nutrient uptake and utilization efficiency. Based on the efficiencies of these two subunits, the overall system efficiency is subsequently calculated by jointly considering inputs, intermediate products, and final outputs.

Four soil properties, namely soil organic matter (SOM), total nitrogen (TN), available phosphorus (AP), and available potassium (AK), are selected as intermediate variables. These variables are generated through the transformation of agricultural inputs within the soil and subsequently support crop growth through nutrient uptake, thereby establishing the functional linkage between the two subunits. The conceptual framework of the NSBM model is illustrated in Figure 3.

The formulation of the non-desired output super-efficiency NSBM model is as follows:

$$\mathit{min}\delta ={\displaystyle \frac{{\displaystyle {\sum }_{d=1}^D}{W}^d\left[1+\frac{1}{m_d+r_{(d,h)}}\left({\displaystyle {\sum }_{i=1}^{m_d}}\frac{s_i^{\mathit{zd}}}{x_{\mathit{io}}^d}+{\displaystyle {\sum }_{p=1}^r}\frac{s_p^{z{c}^h}}{z_{\mathit{cpo}}^{(d,h)}}\right)\right]}{{\displaystyle {\sum }_{d=1}^D}{W}^d\left[1-\frac{1}{s_{1d}+s_{2d}+r_{(d,h)}}\left({\displaystyle {\sum }_{k=1}^{s_{1d}}}\frac{s_k^{\mathit{yd}}}{y_{\mathit{ko}}^d}+{\displaystyle {\sum }_{l=1}^{s_{2d}}}\frac{s_l^{\mathit{bd}}}{b_{\mathit{lo}}^d}+{\displaystyle {\sum }_{p=1}^r}\frac{s_p^{z{c}^d}}{z_{\mathit{cpo}}^{(d,h)}}\right)\right]}}$$

(1)

The fixed linkage formulas for intermediate variables in the NSBM model are as follows:

$$z_o^{(d,h)}={\displaystyle {\sum }_{j=1}^n}{\lambda }_j^d{z}_j^{(d,h)}\hspace{1em}(\forall h,d);$$

(2)

$$z_o^{(d,h)}={\displaystyle \sum _{j=1}^n}{\lambda }_j^h{z}_j^{(d,h)}\hspace{1em}(\forall h,d)$$

(3)

Here, $n$ denotes the number of decision-making units, $D$ represents the number of sectors, and $\mathrm{m},{\mathrm{s}}_{1\mathrm{d}},{\mathrm{s}}_{2\mathrm{d}}$ correspond to the numbers of inputs, desirable outputs, and undesirable outputs in sector $d$, respectively; $(d,h)$ represents the transmission relationship from sector $\mathrm{d}$ to sector $\mathrm{h}$. The input, desirable output, and undesirable output vectors are expressed as $X=\left(x_j^d\right) R^{m_d\times n}$, $Y=\left(y_j^d\right) R^{s_{1d}\times n}$, $B=\left(b_j^d\right) R^{s_{2d}\times n}$. The intermediate output vector linking sector $d$ and sector $h$ is denoted as $Z=\left(z_j^{(d,h)}\right) R^{\left\{\right(d,h)\times }n\}$, $(j=1,\dots ,n; d=1,\dots ,D; h=1,\dots ,D)$. Within vector $\mathrm{Z}$, a distinction is made between variables excluded from efficiency calculation, denoted $\left(z_j^{(d,h)}\right)$, and variables included in efficiency calculation, denoted $(z{c}_j^{(d,h)})$ with $r$ elements. Let $X>0, Y>0, B>0, Z>0$. Then, the production possibility set of the NSBM model, $P=\left\{\right(x^d,y^d,b^d,z^{(d,h)}\left)\right\}$, is defined as:

$$x^d\ge {\displaystyle {\sum }_{j=1}^n}{x}_j^d{\lambda }_j^d\hspace{1em}(d=1,\dots ,D);$$

(4)

$$y^d\le {\displaystyle {\sum }_{j=1}^n}{y}_j^d{\lambda }_j^d\hspace{1em}(d=1,\dots ,D);$$

(5)

$$b^d\ge {\displaystyle {\sum }_{j=1}^n}{b}_j^d{\lambda }_j^d\hspace{1em}(d=1,\dots ,D)$$

(6)

2.3.2. Parameter-Optimized Geographical Detector Model

This study implements a parameter-optimized geographical detector model on the R platform. Building on the conventional geographical detector model, an automated search algorithm is used to optimize both the discretization method and the number of strata, replacing subjective manual discretization. This optimization maximizes the q-statistic, thereby enhancing the robustness and scientific validity of the results. Using the parameter-optimized model, we systematically analyzed the driving effects of individual factors and their interactions on the spatial heterogeneity of AGTFP in Baoding District from 2017 to 2024. Furthermore, interaction heatmaps were developed to analyze the combined effects of multiple factors on green agricultural development.

$$\begin{array}{rr}q& =1-{\displaystyle \frac{{\displaystyle {\sum }_{h=1}^L}{N}_h{\sigma }_h^2}{N{\sigma }^2}}\end{array}$$

(7)

Here, $\mathrm{q}\in [0,1]$ represents the explanatory power of the factor, $L$ denotes the number of variable partitions (or layers), $h=1,2,\dots ,L$ is the partition identifier, $N_h$ is the sample size within layer $h$, $N$ is the total number of samples in the entire region, ${\sigma }_h^2$ is the variance of $Y$ within layer $h$, and ${\sigma }^2$ is the total variance of $Y$ across the entire region.

2.3.3. Panel Regression Model

To further investigate the mechanisms through which different production stages affect overall AGTFP, a panel regression model was established using county-level observations from 2017 to 2024. Overall AGTFP efficiency was specified as the dependent variable, while the efficiencies of the input and output subsystems were included as the key explanatory variables. Considering the panel structure of the dataset, the model accounts for both temporal variation and regional heterogeneity, enabling a robust assessment of the dynamic relationships among the three efficiency measures. All panel regression analyses were conducted using Stata17.The estimated coefficients quantify the marginal contribution of each subunit to overall AGTFP, thereby identifying the critical production stage constraining green productivity improvement and providing a scientific basis for targeted agricultural management and policy interventions.

3. Results

3.1. Temporal Evolution Characteristics of AGTFP

As shown in Figure 4, AGTFP and its input subsystem efficiency remained at relatively low levels throughout the study period, ranging from 0.16 to 0.29, whereas the output subsystem efficiency was consistently higher, indicating an evident imbalance between the two dimensions of agricultural green production. From 2017 to 2024, the overall efficiency increased steadily from 0.234 to 0.286, representing an improvement of 18.18%. Although the increase was gradual, it reflects a continuous enhancement in the green development performance of Baoding’s agricultural system over time. The input subsystem efficiency exhibited a more pronounced upward trend, rising from 0.155 to 0.211 (36.13%), with only a slight decline in 2021 before recovering to the 2020 level in the following year. Compared with the relatively moderate growth in overall efficiency, the faster improvement in input efficiency suggests that recent gains in AGTFP were primarily achieved through more efficient allocation and utilization of production resources rather than substantial changes in production outcomes.

By contrast, the output subsystem efficiency remained relatively stable, fluctuating between 0.401 and 0.445 during the study period, with only a temporary decline of 8.9% in 2023. Its limited variation indicates that output-side performance has entered a relatively mature stage, where further efficiency improvements are increasingly constrained under existing production conditions. Consequently, the temporal evolution of overall efficiency closely followed that of the input subsystem, whereas the output subsystem displayed a comparatively independent trend. This divergence implies that the current stage of agricultural green transformation in Baoding is predominantly characterized by continuous optimization of production inputs, while improvements in output efficiency have become increasingly difficult to achieve through conventional agricultural practices alone. As a result, sustaining future growth in AGTFP is likely to depend not only on further improving resource-use efficiency but also on enhancing output quality through technological innovation and cleaner production practices.

3.2. Overall Efficiency and Subsystem Efficiencies

To further clarify the contributions of the two subunits to the temporal evolution of AGTFP, a panel regression model was employed using region and year as two-way effects, with overall efficiency as the dependent variable and the input and output subsystem efficiencies as explanatory variables. Model diagnostics confirmed the suitability of both the fixed-effects (FE) and random-effects (RE) specifications, allowing the robustness of the estimated relationships to be evaluated across different model assumptions.

As shown in

Table 5. Panel Regression Model Diagnostics.

Test	Statistic	p	Conclusion
F test	144.051	0	Fixed effects model preferred
Breusch–Pagan Lagrange Multiplier test	257.022	0	Random effects model preferred
Hausman test	25.687	0	Fixed effects model preferred

, the regression results consistently indicate that input subsystem efficiency is the dominant determinant of overall efficiency. The estimated coefficients reached 0.590 and 0.606 in the FE and RE models, respectively, and were statistically significant, demonstrating that improvements in resource allocation and input utilization translate directly into higher levels of AGTFP. In contrast, the coefficients for output subsystem efficiency were only 0.018 and 0.030 and failed to reach statistical significance in either model. This finding suggests that variations in output efficiency contribute little additional explanatory power once regional heterogeneity and temporal effects are taken into account.

Combined with the regression results presented in

Table 6. Regression Results of FE and RE Models.

Model	Variable	Coefficient	Std. Error	t	p	R2	F
EF	const	0.141	0.02	6.971	0	within = 0.568	F = 52.533
	input subsystem	0.606	0.064	9.493	0	between = 0.024	p = 0.000
	output subsystem	0.03	0.029	1.051	0.296	overall = 0.04
RF	const	0.146	0.035	4.224	0	within = 0.567	F = 51.039
	input subsystem	0.59	0.072	8.174	0	between = 0.035	p = 0.000
	output subsystem	0.018	0.031	0.568	0.572	overall = 0.053

, the temporal evolution observed in Section 3.1. The strong association between overall efficiency and input efficiency indicates that the recent improvement in AGTFP has been largely driven by reductions in input redundancy and more efficient use of production factors, rather than by substantial gains in output performance. Meanwhile, the insignificant marginal effect of output efficiency implies that agricultural production in Baoding may already operate at a relatively stable output level, leaving limited scope for further efficiency gains through output expansion alone. Therefore, under the current stage of agricultural development, optimizing input allocation and improving resource-use efficiency remain the primary pathways for enhancing AGTFP, while future breakthroughs are likely to require technological innovation capable of simultaneously increasing desirable outputs and reducing environmental costs.

3.3. Spatial Evolution Characteristics of AGTFP

As shown in Figure 5, the spatial distribution of AGTFP and its subunit efficiencies exhibited distinct regional differentiation during 2018–2024. Overall efficiency remained at a medium-to-low level but gradually increased, showing a stable spatial pattern of “higher in the west and lower in the east”. The western mountainous regions generally exhibited higher overall and input efficiencies, whereas the eastern plains showed relatively lower overall efficiency but higher output efficiency. This opposite distribution between input and output efficiencies indicates a spatial division of agricultural production functions, with western regions characterized by more efficient resource utilization and eastern regions maintaining stronger production output capacity.

For further analysis, the changes in efficiency were calculated by subtracting the 2017 values from the 2024 values for both subunits and the overall system.

Figure 6 further illustrates the spatial variations in efficiency changes. Overall efficiency increased in 94.44% of the regions, with an average increase of 5.20%, while input subsystem efficiency increased in 72.22% of the regions, with an average increase of 5.74%. In comparison, output subsystem efficiency showed weaker improvement, increasing in only 55.56% of the regions, with an average change of −3.91%. The improvement of overall efficiency was mainly concentrated in western mountainous and central regions, including Laiyuan, Fuping, Shunping, Boye, and Wangdu, whereas southeastern plain areas such as Anxin and Xiong showed relatively limited improvement.

Overall, the spatial evolution of AGTFP demonstrates a heterogeneous transformation pattern characterized by stronger efficiency improvement in western regions and relatively slower progress in eastern intensive agricultural areas. The dominant contribution of input efficiency improvement further confirms that resource optimization remains the primary pathway for enhancing agricultural green productivity in Baoding.

3.4. Analysis of Driving Factors

3.4.1. Factor Detection

As presented in

Table 7. Results of Geographic Detector Factor Analysis

Code	Overall		Input Subsystem		Output Subsystem
	q Value	p Value	q Value	p Value	q Value	p Value
X1	0.5024	0.0064	0.6183	0.0077	0.4943	0.0343
X2	0.6213	0.0171	0.4969	0.1033	0.434	0.1291
X3	0.7332	0.0006	0.741	0.0004	0.5597	0.038
X4	0.5802	0.0605	0.5693	0.0675	0.471	0.0602
X5	0.3662	0.4459	0.5913	0.0409	0.5747	0.018
X6	0.4637	0.1284	0.3451	0.5169	0.603	0.01
X7	0.7258	0.0012	0.6073	0.0201	0.5065	0.0962
X8	0.4799	0.0755	0.5675	0.0093	0.8283	0.0001
X9	0.6017	0.0108	0.2792	0.4658	0.8309	0
X10	0.5782	0.0416	0.5625	0.0223	0.6314	0.0106

, the factor detection results indicate that the spatial heterogeneity of AGTFP and its subunit efficiencies is jointly influenced by socioeconomic development, agricultural production conditions, and natural environmental characteristics, although their explanatory power differs among efficiency measures.

Among the socioeconomic factors, GDP and the per capita disposable income of rural residents significantly affected overall efficiency and both subunit efficiencies (q = 0.43–0.74), suggesting that regional economic development provides favorable conditions for improving agricultural green productivity through enhanced resource allocation and technology adoption. In contrast, the proportion of the primary industry only significantly influenced overall efficiency, implying a greater effect on the integrated production system than on individual production stages.

Agricultural production conditions showed differentiated effects. Agricultural mechanization was not significantly associated with any efficiency indicator, whereas agricultural facility coverage significantly affected both input and output efficiencies (q = 0.59 and 0.57, respectively), indicating the greater contribution of production infrastructure to agricultural efficiency. Fiscal agricultural support only significantly influenced output efficiency (q = 0.60).

Natural environmental factors exhibited relatively strong explanatory power. Climatic conditions significantly affected overall and input efficiencies, while topography and natural disasters had particularly strong effects on output efficiency (both q = 0.83), indicating that output performance is more sensitive to environmental constraints. In addition, transport accessibility significantly influenced all three efficiency measures (q = 0.56–0.63), highlighting the role of regional connectivity in agricultural green development.

Overall, the results suggest that AGTFP is jointly determined by multiple dimensions, with socioeconomic development providing the development foundation, agricultural infrastructure supporting production efficiency, and natural environmental conditions largely constraining regional differences. The distinct responses of the two subunits further indicate that input and output efficiencies are driven by different mechanisms.

3.4.2. Interaction Detection

As illustrated in Figure 7, the interaction detection results show that all factor pairs enhanced the spatial explanatory power of AGTFP, with most exhibiting nonlinear enhancement, indicating that the combined effects of two factors consistently exceeded their individual contributions. This finding suggests that the spatial heterogeneity of AGTFP is primarily governed by synergistic interactions rather than single factors.

Agricultural mechanization, climatic conditions, topography, and transport accessibility exhibited relatively strong interactions with other variables, producing consistently high interaction q-values. In contrast, interactions involving GDP, endogenous development capacity, fiscal agricultural support, natural disasters, and the combination of topography and natural disasters showed comparatively lower explanatory power, indicating relatively weak synergistic effects.

Despite differences in interaction strength, the interaction structures of overall efficiency and the two subunit efficiencies were broadly consistent, suggesting that similar combinations of factors shape different stages of agricultural production. Overall, the prevalence of nonlinear enhancement demonstrates that agricultural green productivity is driven by coordinated effects among socioeconomic, production, and environmental factors rather than by isolated influences.

4. Discussion

4.1. Relationship Between Overall Efficiency and Subunit Efficiencies

The decomposition of AGTFP into input and output subsystems revealed that the primary constraint on agricultural green productivity in Baoding was not crop production itself, but the efficiency of soil-mediated nutrient transformation within the agricultural production system. During the study period, both overall efficiency and input subsystem efficiency remained at relatively low levels, although they exhibited gradual improvements, whereas output subsystem efficiency was consistently higher and remained comparatively stable. These contrasting trends indicate that different ecological processes dominate different stages of agricultural production. Overall efficiency reflects the integrated performance of the agricultural production system by simultaneously considering external agricultural inputs, intermediate soil nutrient pools, and final agricultural outputs. In contrast, input subsystem efficiency characterizes the transformation of agricultural inputs into soil nutrient resources, whereas output subsystem efficiency represents the efficiency with which crops acquire, assimilate, and convert soil nutrients into biomass production.

Compared with previous studies, the temporal variation in AGTFP observed in this study was generally consistent with reported trends [20], whereas the absolute efficiency values were noticeably lower. This difference is primarily attributable to the incorporation of intermediate soil nutrient transformation processes into the evaluation framework. Conventional AGTFP assessments generally assume a direct conversion relationship between agricultural inputs and outputs, thereby treating the agricultural production system as a “black box” [45]. In reality, however, agricultural inputs must first undergo a series of physicochemical and biological transformations within the soil before becoming available for crop uptake. By introducing soil nutrients as intermediate products, the proposed “material input–soil nutrient transformation–crop production” framework explicitly represents these internal ecological processes and therefore provides a more realistic assessment of agricultural production efficiency. Consequently, the lower overall efficiency estimated in this study should not be interpreted as reduced agricultural performance, but rather as evidence that conventional input–output frameworks may overestimate production efficiency by neglecting soil ecological processes.

The relatively low input subsystem efficiency suggests that the conversion of external agricultural inputs into plant-available soil nutrients remains the principal bottleneck limiting AGTFP improvement. Soil functions not merely as a storage pool for nutrients but also as a dynamic regulator of nutrient retention, transformation, and release. Following fertilizer application, nutrients may be immobilized through adsorption, fixed by soil minerals, incorporated into soil organic matter, transformed by microorganisms, or lost through leaching, runoff, volatilization, and denitrification. Therefore, only a fraction of the applied nutrients ultimately becomes available for crop uptake. This finding indicates that improving fertilizer application alone is insufficient to enhance agricultural green productivity unless soil nutrient transformation efficiency is simultaneously improved.

Among the various soil properties, soil organic matter (SOM) plays a pivotal role in regulating nutrient cycling and maintaining soil fertility [46]. Increasing SOM enhances soil aggregation, improves cation exchange capacity, strengthens soil water-holding capacity, and provides carbon substrates for microbial communities, thereby promoting nutrient retention and buffering nutrient losses. Moreover, soil microorganisms regulate key biochemical processes, including nitrogen mineralization, phosphorus solubilization, and potassium release, which directly determine the proportion of fertilizer nutrients transformed into plant-available forms [47,48]. Consequently, microbial nutrient transformation represents a critical linkage between agricultural inputs and crop production. The relatively low input subsystem efficiency observed in this study therefore likely reflects limited soil ecological functioning, including insufficient nutrient retention, suboptimal microbial activity, and reduced fertilizer use efficiency, rather than merely excessive agricultural inputs.

In contrast, output subsystem efficiency remained comparatively high throughout the study period, indicating that crop nutrient acquisition and utilization efficiency has reached a relatively stable level under the existing production conditions. Crop nutrient uptake is jointly controlled by soil nutrient availability, root system architecture, crop physiological characteristics, and genetic traits. Previous studies have demonstrated that hybrid maize contributes substantially to overcoming yield constraints [49], while physiological characteristics [50] and genetic regulation [51] largely determine crop yield potential. Similarly, crop varieties exert significant influences on nitrogen and phosphorus use efficiency [52,53]. These biological constraints imply that, once nutrient supply exceeds crop demand, further increases in fertilizer inputs contribute little to improving crop nutrient uptake efficiency. Instead, excessive fertilizer application may reduce fertilizer use efficiency and increase environmental risks through nutrient accumulation and loss.

Previous studies have likewise shown that, under conditions of adequate soil nutrient availability [54] and suitable soil moisture [55], reducing excessive agricultural inputs while improving nutrient management can significantly enhance ecological efficiency. These findings suggest that sustainable improvements in AGTFP should focus on increasing the efficiency of nutrient cycling rather than maximizing nutrient inputs. Synchronizing fertilizer application with crop nutrient demand, enhancing nutrient retention within soils, and improving microbial nutrient transformation capacity are therefore more effective strategies for increasing agricultural productivity while minimizing environmental impacts [56].

The marked decline in output subsystem efficiency observed in 2023 was most likely associated with the severe flooding event in Baoding, which affected approximately 79,000 ha of cropland according to the Hebei Rural Statistical Yearbook. Excessive soil moisture reduces oxygen availability within the root zone, suppresses microbial activity, alters nitrogen transformation pathways, and restricts root nutrient uptake, thereby weakening the conversion of soil nutrients into agricultural biomass [57]. This observation is consistent with previous findings demonstrating that crop production efficiency is highly sensitive to extreme climatic disturbances [58]. In contrast, soil nutrient pools generally exhibit greater resilience because soil organic matter and microbial communities provide buffering capacity against short-term environmental fluctuations. Consequently, improving soil health—including maintaining soil structure, increasing SOM accumulation, and promoting microbial activity—not only enhances fertilizer use efficiency but also strengthens ecosystem resilience under increasing climate variability [59,60].

Overall, the results demonstrate that the principal limitation to AGTFP in Baoding lies in the efficiency of soil-mediated nutrient transformation rather than in crop production capacity itself. From an ecological perspective, agricultural sustainability depends on maintaining the integrity of soil functions, including nutrient retention, nutrient cycling, and microbial regulation, which collectively determine fertilizer use efficiency, crop nutrient uptake, and ecosystem functioning. Therefore, future improvements in AGTFP should prioritize enhancing soil health through optimized nutrient management, organic matter accumulation, and practices that strengthen soil biological activity, thereby promoting a more efficient and resilient agricultural production system.

4.2. Comparison of Overall Efficiency and Subsystem Efficiencies

Although overall efficiency and subsystem efficiencies exhibited relatively stable spatial patterns during the study period, their contrasting geographical distributions indicate that different ecological processes govern different stages of agricultural production. Overall AGTFP displayed an evident “east-high, west-low” pattern, whereas output subsystem efficiency exhibited the opposite trend. In contrast, input subsystem efficiency showed no significant spatial clustering. These differences suggest that the mechanisms controlling soil nutrient transformation are not necessarily consistent with those regulating crop nutrient utilization, highlighting the importance of distinguishing intermediate ecological processes when evaluating agricultural green productivity.

The spatial pattern of overall AGTFP remained generally consistent with previous studies conducted in northern China [20,21,61], indicating that agricultural green productivity is strongly constrained by long-term regional resource endowments. Agricultural production systems evolve under relatively stable combinations of soil properties, climatic conditions, topography, and management practices, resulting in persistent regional differences in production efficiency [14]. Consequently, the limited temporal variation observed in this study suggests that short-term improvements in agricultural management alone are unlikely to fundamentally alter the spatial distribution of AGTFP. Instead, substantial improvements in green productivity require gradual enhancement of the underlying ecological functions that regulate nutrient cycling and resource utilization.

One notable finding is that output subsystem efficiency was more strongly influenced by topographic conditions than overall efficiency. According to the factor detection analysis, elevation contributed substantially to the spatial heterogeneity of crop nutrient utilization, whereas its influence on overall AGTFP was comparatively weak. This observation is consistent with previous studies reporting that topography generally exerts limited direct effects on AGTFP [62]. However, topography indirectly regulates agricultural production by influencing soil moisture redistribution, erosion intensity, drainage conditions, and nutrient transport processes. In western Baoding, where elevations are generally higher, greater surface runoff and soil erosion may accelerate nutrient losses while simultaneously reducing soil waterlogging, thereby creating relatively favorable conditions for crop root development and nutrient acquisition. Conversely, lower-lying eastern areas generally possess higher soil moisture availability and stronger nutrient retention capacity, which may promote the accumulation of soil nutrient pools but does not necessarily translate into higher crop nutrient use efficiency. These contrasting responses help explain why overall AGTFP and output subsystem efficiency exhibit different spatial distributions.

The absence of significant spatial clustering in input subsystem efficiency further suggests that soil nutrient transformation processes are regulated primarily by local management practices rather than broad geographical gradients. Unlike natural environmental factors, agricultural inputs such as fertilizer application, irrigation management, and cultivation practices are relatively homogeneous across Baoding because of similar cropping systems and coordinated agricultural policies [6,22,23,43]. Consequently, differences in nutrient transformation efficiency are more likely to arise from variations in soil ecological functioning than from differences in management intensity alone. Soil organic matter content, microbial community composition, and nutrient buffering capacity can vary considerably even under similar fertilization regimes, leading to substantial differences in fertilizer use efficiency and nutrient retention at the field scale. This finding further supports the rationale for explicitly incorporating soil nutrient transformation into AGTFP assessment, as these ecological processes cannot be adequately represented using conventional input–output frameworks.

From the perspective of ecosystem functioning, the contrasting spatial distributions of subsystem efficiencies illustrate the decoupling between soil nutrient accumulation and crop nutrient utilization. High soil nutrient content does not necessarily indicate efficient agricultural production if nutrients remain unavailable for plant uptake or are vulnerable to environmental losses [63]. Conversely, high crop nutrient uptake efficiency cannot compensate for inefficient nutrient transformation if the soil fails to supply sufficient plant-available nutrients. Therefore, agricultural productivity depends not only on nutrient inputs but also on the efficiency of ecological processes governing nutrient retention, microbial transformation, nutrient mineralization, and plant acquisition. These interacting processes collectively determine soil health and ultimately regulate the sustainability of agricultural ecosystems.

The relatively stable spatial patterns observed throughout the study period also suggest that improvements in AGTFP should not rely solely on increasing agricultural inputs or adjusting production intensity. Instead, management strategies should focus on enhancing the ecological resilience of agricultural soils. Practices such as increasing organic matter inputs, improving residue return, adopting conservation tillage, optimizing fertilization schedules, and promoting beneficial microbial activity can strengthen nutrient cycling and improve soil health, thereby increasing nutrient retention and fertilizer use efficiency over the long term [64,65]. Compared with conventional input-oriented management, these soil-centered approaches simultaneously enhance agricultural productivity and ecosystem functions while reducing nutrient losses and environmental risks.

Overall, the spatial heterogeneity identified in this study demonstrates that the ecological mechanisms governing soil nutrient transformation and crop nutrient utilization differ substantially across the agricultural production process. By separating these two processes within the NSBM framework, this study reveals spatial characteristics that are obscured in traditional AGTFP assessments and provides a more mechanistic understanding of the interactions among soil processes, crop production, and ecosystem functioning. These findings highlight that improving agricultural green productivity requires coordinated enhancement of both soil ecological functions and crop nutrient utilization, rather than simply increasing resource inputs or maximizing crop yields.

4.3. Enhancement Strategies

The findings of this study indicate that sustainable improvements in AGTFP should focus on enhancing the ecological efficiency of the agricultural production system rather than simply increasing agricultural inputs. By explicitly incorporating soil nutrient transformation into the evaluation framework, this study demonstrates that the primary constraints on AGTFP arise from two interconnected processes: the conversion of external agricultural inputs into plant-available soil nutrients and the subsequent uptake and utilization of these nutrients by crops.

Improving fertilizer use efficiency should therefore become a key management objective. Excessive fertilizer application often exceeds the nutrient retention capacity of soils, resulting in nutrient fixation, leaching, runoff, and gaseous losses, which reduce fertilizer use efficiency and increase environmental risks. Rather than increasing fertilizer inputs, nutrient management should emphasize synchronizing nutrient supply with crop demand through balanced fertilization and site-specific management.

Enhancing soil health is fundamental to improving the efficiency of nutrient transformation. Soil organic matter (SOM) improves soil structure, nutrient retention, and water-holding capacity while providing substrates for microbial communities. Soil microorganisms regulate nutrient cycling through processes such as nitrogen mineralization and phosphorus mobilization, thereby increasing the proportion of nutrients available for crop uptake. Consequently, practices that increase SOM and stimulate microbial activity can improve fertilizer use efficiency, strengthen nutrient cycling, and enhance ecosystem resilience.

At the crop production stage, improving nutrient uptake efficiency is equally important. Crop nutrient acquisition depends not only on soil nutrient availability but also on root architecture, physiological characteristics, and field management. Optimizing cropping systems, selecting nutrient-efficient cultivars, and improving irrigation and fertilization management can better synchronize soil nutrient supply with crop demand, thereby enhancing resource use efficiency.

Overall, this study highlights that soil functions as the ecological intermediary linking agricultural inputs with crop production. Improving soil health, nutrient retention, microbial nutrient transformation, and crop nutrient uptake simultaneously can enhance fertilizer use efficiency, strengthen ecosystem functioning, and ultimately promote sustainable improvements in AGTFP. These findings provide a mechanistic basis for developing integrated soil–crop management strategies that support both food production and ecological sustainability. In practice, agricultural management should prioritize precision fertilization, integrated nutrient management, and the incorporation of organic amendments to improve soil nutrient transformation efficiency. At the policy level, promoting soil health-oriented management practices and site-specific nutrient management can improve fertilizer use efficiency while reducing agricultural non-point source pollution, thereby supporting long-term agricultural sustainability.

4.4. Limitations and Future Directions

Although this study employs the NSBM model to measure AGTFP and quantifies the influence of intermediate processes in agricultural production on green agricultural development, several limitations remain. First, due to the workload associated with soil sampling, this study focuses solely on the Baoding district. The relatively small spatial scope may limit the generalizability of the findings. Although the study refines the agricultural production process, it only considers the relationship between agricultural production and the agro-ecosystem, neglecting the primary actors of production—the farming communities—and the economic benefits they gain, particularly regarding rural development.

Future research could expand the study area from Baoding to larger-scale agricultural regions and extend the temporal sequence to enhance the generality and timeliness of conclusions, thereby identifying efficiency evolution patterns across different regions and stages. Additionally, integrating indicators such as agricultural economic returns, household income, and ecological compensation into AGTFP measurement would enable the construction of an integrated economic-ecological evaluation system, quantifying the comprehensive benefits of green production and revealing the coordination mechanisms between ecological and economic efficiency.

5. Conclusions

Based on soil nutrient data from farmlands in the Baoding district during 2017–2024, this study employs the NSBM model to construct a novel AGTFP evaluation framework capable of capturing the efficiency of intermediate processes in agricultural production, and systematically investigates the key processes constraining AGTFP improvement.

The main findings are as follows: Compared with conventional AGTFP measurement approaches, the framework based on soil nutrient data can effectively reflect the efficiency of individual stages within the agricultural production process. This allows for precise identification of internal factors limiting AGTFP growth and enables targeted improvement measures, thereby enhancing the accuracy and effectiveness of policy recommendations. The overall growth of agricultural green total factor productivity is primarily driven by improvements in the efficiency of the input subsystem, whereas the efficiency of the output subsystem remains largely unchanged. This stability is attributed to the biophysical characteristics of crops and their root environments, as the efficiency of nutrient uptake and conversion remains relatively constant under sufficient soil nutrient conditions. Furthermore, the determinants of overall system efficiency largely align with those of the input subsystem, while the responses of the input and output subsystems to these factors differ significantly. Consequently, measuring only overall efficiency without accounting for subsystem efficiencies may hinder effective improvement of internal production processes, thereby constraining the sustainable development of agriculture.