Impact of Advanced Agriculture Technologies and Energy Consumption on Crop Yields in Modern Agriculture Using Deep Learning Techniques

Abstract

ABSTRACT

Growing concern over food security has drawn worldwide scholarly attention. Addressing food security issues highlights the vulnerability of agricultural yield to the complexity of agriculture inputs. Therefore, this study considers the intricacies of cultivation inputs and their effect on cereal production across 20 developing Asian countries from 1990 to 2022. First, advanced machine learning algorithms are employed to investigate the combined impact of the farming Product Complexity Index on agricultural yields. Second, the Granger causality test was used to uncover the causality direction between agricultural yield and exogenous variables. Both the causal inference neural network (CINN) and deep neural network (DNN) models show a rapid initial decrease in loss during the early epochs, followed by a more gradual decline, indicating effective learning and convergence. Notably, the CINN model consistently starts with a lower loss compared to the DNN model, suggesting superior performance in minimizing the training loss. These machine learning techniques have successfully predicted the synergistic relationships, leading to significant improvements in cereal yield forecasting. The Granger causality results revealed feedback causality between the agricultural Product Complexity Index and crop yields and the use of fertilizer and agricultural yields on different lags. These results emphasize the potential for targeted guidelines that harness the interactions between complexities in agriculture and the application of fertilizer to improve cereal yields.

Introduction

Food security is a paramount worldwide problem, ensuring that everyone consistently has access to enough wholesome, safe, and sufficient food for a healthy and active existence. According to the World Food Programme (World Bank 2024), 343 million people are acutely food insecure across 74 countries, marking a 10% increase from 2023 and nearly 200 million more than pre-pandemic levels. It is essential to the health, economic growth, and social unity of societies, as well as their stability and success. By assuring the availability and access to sufficient and nutritious food, we can mitigate the detrimental effects of malnutrition and hunger, which pose substantial barriers to human development and productivity. Moreover, it is crucial for attaining sustainable development objectives, as it fosters agricultural sustainability and aids in mitigating poverty. A key focus of this effort is to improve agricultural productivity, ensuring a consistent and abundant food provision. Improving agricultural methods (Nicholson et al. 2021), ensuring equitable access to food resources, and reducing the effects of climate change on food production are all part of addressing food security (Gwambene et al. 2023). Ultimately, ensuring food security is essential to creating a just and resilient world where everyone may prosper.

Food security is intricately connected to agricultural productivity, as enhanced farming operations' yield and efficacy are essential for generating a consistent and sufficient food supply to meet the growing global demand (Sekaran et al. 2021; Sridhar et al. 2023). Recognizing the importance of food security, researchers have focused on various factors, including agricultural water systems (Bi et al. 2024), land degradation (Jiang et al. 2023), environmental systems, and the carbon footprint in rain-fed conditions (Lin et al. 2024). Implementing ornamental agricultural practices increases crop productivity and reduces post-harvest losses, improving the accessibility and availability of nutritious food (Aly and Borik 2023; Kandegama et al. 2022). Hence, attaining enduring food security and promoting sustainable development hinge upon augmenting agricultural production. Developing nations are experiencing a substantial shortfall in meeting their objectives in agricultural production and ensuring enough food supply for their populations (Pawlak and Kołodziejczak 2020). Researchers should prioritize their attention and their endeavors toward this worldwide issue.

Food security is a pressing global issue shaped by the interplay of social, technological, and environmental factors. The growing complexity of agricultural technologies and their increasing energy demands underscore the urgent need for sustainable solutions. Over the decades, global challenges, particularly the sustained pressure on agricultural productivity within the context of climate change, have intensified the urgency of addressing these issues. Integrating advanced technologies with sustainable practices can pave the way for a resilient food system and ensure food security worldwide.

Agricultural production is significantly vulnerable to complexities in agricultural inputs (Zulfikri et al. 2024). To tackle this problem, the Product Complexity Index (PCI) (Baz et al. 2022) assesses the degree of this vulnerability and its effect on agricultural productivity (Figure 1). The PCI evaluates the variety and complexity of goods a region can produce, which is important for understanding the complexities related to agricultural contributions (Balland et al. 2022). PCI's high values show innovative technologies that enhance efficiency, productivity, and sustainability. Such complexity explains better inputs and technologies, reducing labor costs and improving agricultural yield, thus increasing overall agricultural performance. We can determine the influence of intricate and varied agricultural inputs on crop productivity by analyzing the PCI. This assessment is vital for identifying critical areas needing improvement to enhance agricultural production and contribute to global food security. Gaining awareness of and effectively dealing with the weaknesses the PCI exposes is crucial for developing effective strategies to enhance crop productivity and ensure sustainable agricultural practices.

[IMAGE OMITTED. SEE PDF]

Thus, to understand the complexity of this vulnerability in solving the food security goals, this study employed two advanced machine learning methods for causal inference (Brand et al. 2023), namely deep neural network (DNN) (Bi et al. 2023) and causal inference neural network (CINN) (Zhang et al. 2024a). The selection of these models is based on their distinct advantages in meeting the research objectives. CINN is particularly suited for tasks requiring bidirectional mapping and uncertainty quantification, enabling interpretable and robust predictions. DNN, on the other hand, excels in modeling nonlinear relationships and processing large, complex datasets, making it highly effective for capturing intricate patterns in the data (Amiri et al. 2024). DNN also enhances pre-training models, improves training efficiency, and yields better predictive results. The prognostic power of CINN and DNN, combined with their ability to infer causal relationships, provides insights that go beyond simple correlations. These models are particularly effective for decision-making by analyzing the factual effects of interventions and improving robustness and generalizability, especially in critical domains such as economic development and healthcare. Moreover, this research used PCI, which is crucial for agricultural productivity and highlights the complex inputs and machinery of the agriculture sector. Examining PCI shows the innovation and technological advancement embedded in agricultural equipment and machinery.

The remaining parts of the paper are organized as follows: Section 2 reviews previous studies on factors influencing agricultural yields, with a focus on developing countries. Section 3 presents information regarding the impact of agricultural product complexity on cereal yield, including the data set, theoretical framework, and econometric model. Section 4 discusses the results and analysis. Finally, Section 5 concludes the paper and provides policy recommendations.

Literature Review

Food security and agricultural yield are critical components of development and sustainability, as proven by widespread literature emphasizing their interdependence (Ingram 2011). Targeted agricultural productivity, accomplished through improvements in technology, farming practices, and crop varieties, is important for meeting the food demands of an increasing population (Qaim 2020; Khan et al. 2021). Literature underscores that progress in yield can alleviate food insecurity by improving the availability and accessibility of food, especially in regions vulnerable to environmental change and resource insufficiency (Ehi et al. 2024; Rahman et al. 2024). Moreover, combined methods that enhance the yield with sustainable practices are important to ensure long-run foodstuff security without compromising ecological health. For example, integrating reflectance datasets into the WOFOST-PROSAIL model has enhanced regional winter wheat yield estimates (Huang et al. 2019), while deep learning approaches have provided precise predictions and uncertainty analyses for wheat yield at the county level in China (Wang et al. 2020). Moreover, innovations in artificial intelligence and wireless sensing technologies, such as self-powered systems for smart industries (Li et al. 2024) and flexible optical sensors for real-time food monitoring (Zhang et al. 2024b), are revolutionizing precision agriculture and food security. Scholars also emphasize the significance of infrastructure, investment in the agriculture sector, and policy support to sustain and enhance yields, thereby improving worldwide food security (Rafael 2023; Su et al. 2023).

Technologies for soil preparation, threshing, and harvesting have significantly enhanced agricultural yield by improving efficiency, reducing labor expenditures, and increasing crop yields (Hasan et al. 2020; Kumar et al. 2023). Advanced harvesting machinery like combine harvesters and threshers integrate cutting and cleaning, radically saving time and field labor and reducing post-harvest losses (Gautam et al. 2023). Modern mechanical threshers professionally remove grain from the chaff, minimizing losses and increasing quantity, unlike conventional procedures (Liang and Wada 2023). Modern soil preparation machinery, harrows, ploughs, and seed drills enhance soil structure and placement of seed, while conversion cultivation methods reduce corrosion and improve water retention.

The research investigates the factors that consistently describe the adaptation of advanced technology across varied contexts, employing meta-analysis of regression methods (Ruzzante et al. 2021). The study examined the fact that technological adaptation with extended planning horizons, such as corrosion regulator techniques, suggestively enhances agricultural yield. Similar results were noted for high-tech adoption that enhanced the long-run sustainability of grains and crop yields by adopting proper farm supervision systems (Takahashi et al. 2020). These techniques are crucial for developing countries, particularly sub-Saharan countries, to ensure sustainable agricultural yield and food security. Due to the growing concern for advanced technology (Sarfraz et al. 2023), developing innovations such as the Internet of Things, artificial intelligence, machine learning, and advanced farming machinery assisted accurate decision-making to enhance agricultural yield. Another study (Chandio et al. 2023) examined the influence of information and technology on crop yield in Thailand, the Philippines, Malaysia, and Indonesia from 1991–2018. The findings proposed that technological innovation via pesticides and fertilizer significantly enhances crop production in the long term. Moreover, research (Pandey and Pandey 2023) similarly investigated the role of geospatial technology in agricultural yield and food security goals. It shows that lessening chemicals or adopting conventional techniques on farms can protect soil-faunal variety, which is threatened by the misuse of herbicides, insecticides, and pesticides.

From an eco-friendly perspective, energy usage is very important at every phase of agricultural yield, from fertilizers manufacturing to fueling machinery for soil preparation, planting, harvesting, and threshing. A study (Shah et al. 2023) examined the effect of energy consumption on crop production in India, China, Russia, and South Africa from 1990 to 2019. The findings indicate that while more energy enhances crop production, carbon emission negatively affects agricultural yield. Moreover, a two-stage investigation using a panel dataset from 2002–2019 showed that energy efficiency enhanced the agricultural yield by 84% within the range of 79%–86% in South Asia (Khan et al. 2023). However, these findings suggest that there is still 16% potential to improve energy for the agriculture sector. Agricultural activities reduce environmental quality in the long run, as shown in the Chinese agriculture sector from 1990Q1 to 2019Q4 (Ramzan et al. 2024). Additionally, agricultural development had a statistically negative impact on the environment, signifying that it contributes to climate change.

Research Methodology

Theoretical Framework

Developing a conceptual framework for scholarly investigation into the effects of agricultural technology complexity, carbon emissions, energy consumption, and fertilizer use on agricultural yields requires comprehending fundamental ideas and their interconnectedness (Figure 2). Agricultural technology complexity pertains to the level of intricacy and incorporation of contemporary technologies in farming, encompassing automation, precision agriculture, biotechnology, and smart farming instruments. The hypothesis suggests that this complexity has a favorable effect on agricultural output by improving efficiency and precision in the utilization of resources. The production and use of machinery, livestock, and fertilizers are predicted to generate carbon emissions, which will have a detrimental effect on crop yields. These emissions contribute to climate change and environmental degradation, resulting in adverse weather conditions that might impede the growth of crops. Energy consumption, which refers to the amount of energy used in agricultural activities, is believed to have a beneficial impact on crop yields when effectively controlled. This is because it guarantees sufficient power for crucial farming tasks like irrigation and machinery operation. Applying fertilizers, whether chemical or organic, is crucial for providing necessary nutrients to plants and is believed to have a favorable effect on crop yields when administered in the most effective way in terms of type, amount, timing, and method.

[IMAGE OMITTED. SEE PDF]

Data Analysis

This study investigates the complexities of agricultural inputs and their impact on grain production in 20 emerging Asian nations from 1990 to 2022 (Figure 3). The variables cereal yield, energy consumption, carbon emissions, technology complexity, and fertilizer application are key to assessing agricultural sustainability. Cereal yield reflects farming productivity and resource efficiency, critical for feeding the growing population. Energy consumption indicates farming's environmental footprint, as high energy use often leads to more carbon emissions. Carbon emissions from agriculture contribute to climate change, making their reduction crucial for sustainable farming. The complexity of technology impacts productivity and environmental costs, with advanced methods requiring more energy and maintenance. Lastly, fertilizer application boosts yields but can harm the environment through runoff and emissions. These variables together help evaluate the balance between productivity, efficiency, and sustainability in agriculture. The exclusion of irrigated area is due to the reliance on rainfall in regions without irrigation systems, where crop growth depends primarily on natural precipitation. In such areas, irrigation is not a significant input, making it less relevant for the study focused on other variables like fertilizer application and energy consumption. The fundamental metric for measuring agricultural productivity is cereal yield, expressed in tons per hectare in Table 1, with data sourced from “Our World in Data” (WorldData 2022a). Energy consumption, quantified as primary energy consumption per capita in kilowatt-hours per person, represents the amount of energy used by each individual in the agriculture sector, and this data is also obtained from “Our World in Data” (WorldData 2022b). Carbon emissions, measured by annual per capita CO₂ emissions, quantify the environmental impact of agricultural activities, with data sourced from the same provider. The complexity of agricultural technology is determined by the degree of sophistication and integration of current technologies (Figure 4), such as automation and precision agriculture, with this information obtained from the Observatory of Economic Complexity. Fertilizer application, quantified as the amount of fertilizer used per hectare, emphasizes the significance of nutrient management in agricultural productivity, with data again sourced from “Our World in Data” (Ritchie 2023). The study aims to analyze these variables to gain insights into the impact of various cultivation inputs on grain production in emerging Asian countries, ultimately aiding in the formulation of strategies to improve agricultural productivity and sustainability. The statistical summary reveals that the Yield variable has a mean of 3.727 with a standard deviation of 1.527, indicating moderate variability in crop yields across the dataset in Table 2. Additionally, Fertilizer usage shows a high mean of 141.529, with a significant spread (standard deviation of 104.762), suggesting a wide range of fertilizer application levels, while Emission and Energy exhibit considerable variability in their values as well.

[IMAGE OMITTED. SEE PDF]

[IMAGE OMITTED. SEE PDF]

TABLE 1 Variables description.

Variables	Indicator	Measurement	Source
Cereal yield	Yield	Yield tons per hectare	WDI
Energy consumption	Energy	Kilowatt-hours per person	WDI
Carbon emissions	Emission	Kilowatt-hours per capita CO2 emissions	WDI
Technology complexity	PCI	Technology complexity index	OEC
Fertilizer application	Fertilizer	Fertilizer used per hectare	WDI

TABLE 2 Descriptive statistical analysis.

Variables	Mean	Std.	Min	Max
Yield	3.727	1.527	0.363	6.826
Energy	9.123	1.361	5.319	11.463
Emission	4.234	4.479	0.053	20.870
PCI	0.886	0.124	0.700	1.158
Fertilizer	141.529	104.762	0.640	515.580

Deep Learning-Based Causal Inference

We utilized deep learning models, namely the DNN and CINN, to conduct a causal inference analysis (LeCun et al. 2015). These models are essential for precisely capturing and interpreting the complex interconnections among various variables and their impacts on yield production (Vasilescu 2022). By employing CINN and DNN, we can deduce the interconnected causal impacts of PCI supply, fertilizer utilization, carbon emissions, and energy consumption on agricultural yield. The advanced characteristics of these models in managing non-linear interactions and high-dimensional data offer a strong foundation for empirical investigation (Runge et al. 2023). This allows us to provide a precise empirical rationale for the positive benefits of these factors, providing vital insights into their combined influence on crop productivity. Therefore, utilizing these advanced models is crucial for developing efficient agricultural practices and policies to enhance crop production and guarantee food security.

The CINN is a neural network specifically created to acquire the ability to learn invertible mappings between input and output data. It is especially valuable when achieving objectivity (a one-to-one correspondence) between input and output data is needed, such as in dynamic modeling, density estimation, and uncertainty quantification. Commonly employed for various purposes, it includes obtaining characteristics, predicting values, and categorizing data. The statistical illustration of a CINN and DNN can be written as follows in Equation (1):1$$$ f:x\to y $$$where $$$ x $$$ is the independent variable and $$$ y $$$ is the dependent variable. This implies the existence of a corresponding inverse function in Equation (2):2$$$ {f}^{-1}:y\to x $$$

The training statistics include the input variable $$$ x $$$ and the output variable $$$ y $$$. The output of each layer in the DNN is produced by performing an affine transformation, which is subsequently followed by a non-linear activation function.

Statistically, this can be shown as in Equation (3):3$$$ y=\sigma \left({W}_L.\sigma \left({W}_{L-1}\dots \sigma \left({W}_1.x+{b}_1\right)+{b}_{L-1}\right)+{b}_L\right) $$$

The symbols $$$ {W}_i $$$ and $$$ {b}_i $$$ denote the weights and biases of the $$$ {i}^{th} $$$ layer, respectively. The symbol σ represents the activation function, which might be tanh, sigmoid, or ReLU. The symbol $$$ L $$$ represents the total number of layers in the neural network.

Granger causality is a statistical analysis used to determine if one data series can accurately predict another time series (Tiwari et al. 2023). The theory is that if a variable X is a Granger-causal factor of Y, then previous values of X contain data that assists in forecasting Y beyond the previous values of Y alone. Let's consider a two-time series denoted as follows in Equation (4):4$$$ {Y}_t={\alpha}_0+\sum \limits_{i=1}^p{\alpha}_i{Y}_{t-i}\sum \limits_{j=1}^q{\beta}_j{X}_{t-j}+{\varepsilon}_t $$$$$$ Y $$$ and $$$ X $$$ are dependent and independent variables at time $$$ t $$$, respectively. The coefficients represented by $$$ {\alpha}_i $$$ and $$$ {\beta}_i $$$, at lagged $$$ i $$$ for $$$ Y $$$ and $$$ X $$$ values, respectively. The error term is denoted by $$$ {\varepsilon}_t $$$, while $$$ p $$$ and $$$ q $$$ are the lags included in the test.

Results and Discussion

The first graph Figure 5, featuring green histograms, displays the unscaled raw data for the same variables, highlighting their initial ranges and frequency distributions. The second graph Figure 6, with blue histograms, shows the distributions of scaled variables (energy, emissions, PCI, fertilizer, and yield), where the data has been normalized to have a mean of 0 and a standard deviation of 1. While the KDE curves estimate the smoothed probability distribution, the histograms represent the frequency of scaled values. This scaling simplifies comparisons between variables with different units and ranges. Together, these graphs illustrate how scaling affects the data, preserving the underlying distribution while normalizing the range for easier comparison.

[IMAGE OMITTED. SEE PDF]

[IMAGE OMITTED. SEE PDF]

The graph Figure 7 also includes boxplots comparing the “Original” and “Cleaned” data for five variables: energy, emissions, PCI, fertilizer, and yield. In the original data (top row), several variables exhibit outliers (e.g., emissions and fertilizer), represented as points outside the whiskers of the boxplots. After cleaning (bottom row), these outliers are reduced or removed, indicating data preprocessing steps to handle anomalies and improve the dataset's quality for analysis.

[IMAGE OMITTED. SEE PDF]

The Figure 8A is a SHAP summary plot that interprets the contributions of features in a machine learning model. The x-axis displays SHAP values, which indicate both the magnitude and direction of each feature's effect on the model's output. Positive SHAP values suggest that a feature increases the prediction, while negative values indicate a decrease. The y-axis lists the features—fertilizer, emissions, energy, and PCI—ranked by their average importance. Each dot represents an individual instance in the dataset, with a color gradient ranging from blue (indicating a low feature value) to red (indicating a high feature value) to illustrate the magnitude of the feature in that instance.

[IMAGE OMITTED. SEE PDF]

Key insights from the plot are as follows: Fertilizer displays the greatest variability in SHAP values, with high feature values (red) typically linked to positive impacts on predictions, while low values (blue) often lead to reduced predictions. Emissions exhibit a similar trend, where higher values generally increase predictions, although their range of influence is narrower compared to fertilizer. Energy, on the other hand, shows both positive and negative impacts, indicating a more complex and less consistent relationship with the model output. Lastly, the PCI exhibits tightly clustered SHAP values around zero, suggesting a relatively weak influence on the model's predictions. Overall, the plot emphasizes the relative importance of features, their directional impacts, and the variability of their contributions across instances.

Figure 8B presents a SHAP summary plot that illustrates the impact of four features—fertilizer, emissions, energy, and PCI—on the predictions of a machine learning model. The x-axis represents SHAP values, which quantify each feature's contribution to increasing or decreasing the model's output, while the y-axis ranks the features by their importance. Each dot on the plot corresponds to an instance, with a color gradient ranging from blue (indicating low feature value) to red (indicating high feature value) that reflects the magnitude of the feature for that specific instance. Fertilizer has a strong positive influence on the model's predictions; higher feature values (red) lead to increased SHAP values, while lower values (blue) result in reduced predictions. Emissions exhibit a similar trend, with higher values generally boosting predictions, though the effect is less pronounced. Energy displays mixed effects, contributing both positively and negatively, which indicates variability in its influence. In contrast, PCI shows tightly clustered SHAP values near zero, suggesting that it has minimal impact on the model's output. Overall, this plot effectively visualizes feature importance, the direction of influence, and instance-level variability within the model.

Figure 9 displays two scatter plots that compare the projected yields to the actual yields for two distinct prediction models: a DNN and a CINN. The left-hand plot, labeled “CINN Predictions vs. Actual,” shows a group of blue points, each demonstrating a pair of actual and predicted yield values for the CINN model. The dashed red line, which denotes the equation y = x, visually portrays the predictions' accuracy. The proximity of the points to this line indicates the model's accuracy level. The graphic on the right, labeled “DNN Predictions vs. Actual,” has a similar format, but green dots represent the DNN model's predictions. Both figures exhibit a robust positive correlation between the actual and expected yields, suggesting that both models operate admirably. The clustering of data points around the red dotted line indicates that both models exhibit precise predictive capabilities. By analyzing the density and distribution of data points around this line, one can deduce the relative accuracy of the models. Outliers, data points far from the red line, highlight instances where projections differ greatly from the actual values. The CINN and DNN models exhibit commendable performance, with subtle variations in point distribution that may indicate disparities in prediction errors or variability.

[IMAGE OMITTED. SEE PDF]

Figure 10 depicts the residuals for two methods, a DNN (right) and a CINN (left), which compared the assessment error (two different magnitudes of residual error). The residuals of the CINN model (difference between the predicted yield and actual yield) are plotted as blue dots in the left-hand plot, while the residuals of the DNN model are plotted as green dots in the right plot. Both graphs display residuals within the interval of −0.5 to 0.5. The symmetrical arrangement of dots down the horizontal axis (y = 0) suggests that there is no discernible pattern of bias in the predictions of either model. The clustering of residuals within the designated range indicates that both models exhibit relatively minor and comparably scattered prediction errors, with no notable disparities in their performances.

[IMAGE OMITTED. SEE PDF]

Figures 11 and 12 show the training and validation losses for DNN and CINN models. Both models show convergent trajectories in their loss curves, indicating reliable and steady training procedures. The DNN model was thus employed for comparison to validate the prediction accuracy of the CINN model. A good prediction accuracy of the CINN model is shown by a high R² (0.79) and narrow error margins. By comparison, the DNN model's R² value of 0.77 was somewhat lower than the CINN model's, indicating that although it performed well, it was slightly less accurate. In the following diagram (Figure 13), the training loss curves of the two models (CINN and DNN) are shown for 1000 epochs: Both models show an initial decrease in loss during the first epochs, followed by a more gradual reduction, which suggests effective learning and convergence. Compared to the orange (DNN model), the CINN model, represented by the blue line, also starts training with a slightly lower loss almost always. This indicates better performance in terms of the minimization of training loss. The CINN and DNN models consist of three dense layers as shown in Table 3: the first dense layer has 64 units and 320 parameters, the second dense layer has 32 units and 2080 parameters, and the output layer has 1 unit with 33 parameters. The “none” in the output shape represents the flexible batch size, which can vary during training or inference. In total, the model has 2433 trainable parameters and no non-trainable parameters.

[IMAGE OMITTED. SEE PDF]

[IMAGE OMITTED. SEE PDF]

[IMAGE OMITTED. SEE PDF]

TABLE 3 Architecture and parameter summary of CINN and DNN models.

Layer type	Output shape	Parameters
Input_1 (input layer)	(none, 4)	0
Dense (dense)	(none, 64)	320
Dense_1 (dense)	(none, 32)	2080
Dense_2 (dense)	(none, 1)	33
Total parameters	2433
Trainable parameters	2433
Non-trainable parameters	0

After employing advanced machine learning methods like CINN and DNN to examine the influence of independent factors on agricultural productivity, it is essential to determine the causal relationship among these variables. To achieve this, we utilized a Granger causality test. The results of the p-values are displayed in Figure 14, with yield being considered as the dependent variable and yield being utilized as the independent variable. The results indicate no causal relationship between carbon emission and yield. Unidirectional causality is noted running from energy to agriculture yield. Nonetheless, a bidirectional nexus was acknowledged between agricultural complexity and yield, and fertilizers and yield, suggesting that agricultural output is affected by both agricultural complexity and fertilizer application. These data suggest that the degree of agricultural complexity and the use of fertilizers significantly impact agricultural productivity. Agricultural inputs are crucial in enhancing agricultural output and ensuring food security.

[IMAGE OMITTED. SEE PDF]

The remarkable performance of both models demonstrates the interconnected and comprehensive cause-and-effect links among energy use, carbon emissions, fertilizer use, and agricultural complexity concerning yield output. Essentially, the production of yield functions as a dependable measure of a portfolio focused on food security. This includes factors such as access to agricultural inputs (agricultural complexity), energy consumption, environmental quality (carbon emissions), and fertilizer use. In detail, well-organized energy consumption in agriculture increases production through modernization, transportation, and irrigation but also raises apprehension about sustainability because it influences carbon emissions (Zhang et al. 2023; Yang 2023). The agricultural sector, an important foundation of greenhouse gases, requires exercises that decrease emissions, such as the accuracy of agriculture and the transition to clean energy. The complexity in agriculture, concerning various exercises and access to advanced machinery, increases the agriculture sector's resilience, productivity, and sustainability (Javaid et al. 2023; Javaid et al. 2022). The application of fertilizer is important for crop yield, and soil fertility should be accomplished to stop environmental degradation and soil-health decay (Pahalvi et al. 2021; Baweja et al. 2020). High and stable production, reflecting the relationship of these factors, is necessary for meeting foodstuff demand, guaranteeing economic constancy, and reaching food security. These findings provide a significant policy framework that supports the utilization of this synergy to attain food security. These factors can influence economic growth: the presence of agricultural resources, the availability of renewable energy, the exchange of advanced and complex technological goods, and the improvement of environmental conditions.

However, effectively utilizing this collaboration requires careful and detailed policy development, specifically addressing the environmental consequences related to the availability of renewable energy and the diffusion of agricultural technologies, particularly in developing nations. Low-income families frequently fall short of reaching the average level of crop production and contribute significantly to food security, which is crucial for sustaining sustainable economic growth. Combining traditional agricultural methods and the absence of accessible fertilizers could have significant implications for food security. The shift from using fossil fuels to utilizing clean energy resources is a crucial element in agricultural production, as the dependence on fossil fuels considerably contributes to carbon emissions and hastens the process of climate change. The consequences would be completely retrogressive without a transition to renewable energy and environmentally conscious economic growth.

Conclusion and Policy Implication

This study examines the interdependent and causal nexus between fertilizer use, energy consumption, agriculture yields, carbon emission, and various aspects of product complexity in the agriculture sector. From a food security perspective, aspects comprise inorganics, organometallics, soil preparation, threshing, harvesting, and planter germination products. To this end, an advanced cross-methodological method was employed, integrating machine learning causal inference approaches like DNN, CINN, and the conventional Granger causality technique. The combination ensured vigorous and precise findings, providing valuable visions for policy endorsements. Machine learning is the best approach to predict the causal effect between exogenous and endogenous variables. To the best of research, this comprehensive model accounted for overall effects to ensure complete causal inferences.

The outcomes propose combining fertilizer application and energy consumption, ensuring a proper supply chain of complex innovated agriculture machinery, and advanced research and development capabilities that collectively enhance cereal yields while controlling carbon emission levels. The targeted goal can be achieved by strengthening region-wise ability and knowledge in manufacturing state-of-the-art, complex, resource-based agriculture products and machinery and defusing its various and sophisticated range of technological trade. Eventually, this will enhance the overall agricultural yields and ensure food security, especially in developing regions.

These results emphasize the significance of leveraging this interaction to enhance various aspects of agriculture products complexity, such as organic and inorganic products, soil preparation, harvesting, threshing, and planter germination, etc. this approach can foster green agriculture growth through advanced agriculture machinery, high-tech exports, and research innovation in the agriculture sector. Policymakers must consider the following suggestions to harness this nexus and manage associated risks efficiently.

First, investing in advanced technologies for organic and inorganic fertilizers, high-quality seeds, soil preparation tools, and efficient harvesting equipment can significantly enhance agricultural productivity in emerging regions. By focusing on sustainable practices, such as bio-based inputs and precision farming, these innovations will drive growth, improve food security, and empower smallholder farmers of developing countries. Because these developing countries often unmet their agriculture production targets, large-scale agriculturalists still rely on outdated cultivation approaches. To this end, subsidizing advanced technology costs, accommodating technology-sharing plans, and increasing training programs to educate farmers on advanced technology use and maintenance is essential. Second, fertilizer application should be enhanced by funding and sponsoring first-class fertilizer, applying widespread soil-testing facilities, and conducting workshops to train agriculturalists on the best use of fertilizer. Third, an integrated policy structure is important to address the synergic issues of technological access, energy consumption, and fertilizer application. This comprises multi-sector cooperation, promoting agricultural technologies through research and development incentives, and better infrastructure, such as storage and road facilities.

The study primarily focuses on developing agricultural production countries based on publicly available data, emphasizing their limitations. Consequently, the results and discussion may not universally apply to different contexts. Therefore, it is crucial to consider this constraint when extending findings to nations with more complex agricultural products that drive their technological and economic advancement.

Acknowledgments

This study received funding from Zhejiang Social Science Fund, grant number 23QNYC13ZD and National Social Science Fund in China, grant number 22&ZD083.

Disclosure

Declaration of Generative AI and AI-Assisted Technologies in the Writing Process: During the preparation of this work, the authors (being non-native English speakers) used unpaid ChatGPT v.3.5 in order to only remove grammatical errors and enhance clarity. After using this tool, the authors keenly reviewed and edited the content as needed and took full responsibility for the content of the publication.

Conflicts of Interest

The authors declare no conflicts of interest.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.

References

Aly, A. A., and Z. M. Borik. 2023. “Food Crops, Growth and Productivity as an Important Focus for Sustainable Agriculture.” In Plant Small RNA in Food Crops, 3–23. Elsevier.

Amiri, Z., A. Heidari, N. J. Navimipour, M. Unal, and A. Mousavi. 2024. “Adventures in Data Analysis: A Systematic Review of Deep Learning Techniques for Pattern Recognition in Cyber‐Physical‐Social Systems.” Multimedia Tools and Applications 83, no. 8: 22909–22973.

Balland, P.‐A., T. Broekel, D. Diodato, et al. 2022. “The New Paradigm of Economic Complexity.” Research Policy 51, no. 3: [eLocator: 104450].

Baweja, P., S. Kumar, and G. Kumar. 2020. “Fertilizers and Pesticides: Their Impact on Soil Health and Environment.” Soil Health 59: 265–285.

Baz, K., D. Xu, J. Cheng, et al. 2022. “Effect of Mineral Resource Complexity and Fossil Fuel Consumption on Economic Growth: A New Study Based on the Product Complexity Index From Emerging Asian Economies.” Energy 261: [eLocator: 125179]. [DOI: https://dx.doi.org/10.1016/j.energy.2022.125179].

Yang, F. 2023. “Impact of Agricultural Modernization on Agricultural Carbon Emissions in China: A Study Based on the Spatial Spillover Effect.” Environmental Science and Pollution Research 30, no. 39: 91300–91314.

Zhang, R., M. Wang, T. Zhu, Z. Wan, X. Chen, and X. Xiao. 2024b. “Wireless Charging Flexible In‐Situ Optical Sensing for Food Monitoring.” Chemical Engineering Journal 488: [eLocator: 150808].

Zhang, X., X. Long, Y. Liu, K. Zhou, and J. Li. 2024a. “A Generic Causality‐Informed Neural Network (CINN) Methodology for Quantitative Risk Analytics and Decision Support.” Risk Analysis 44: 2677–2695.

Zhang, Y., M. Ge, Q. Zhang, S. Xue, F. Wei, and H. Sun. 2023. “What Did Irrigation Modernization in China Bring to the Evolution of Water‐Energy‐Greenhouse Gas Emissions?” Agricultural Water Management 282: [eLocator: 108283].

Zulfikri, A., E. M. N. Ningsih, I. Harsono, and H. Susanto. 2024. “Agricultural Adaptation Strategies to Weather Fluctuations for Improved Agribusiness Sustainability in West Java.” West Science Nature and Technology 2, no. 1: 17–23.