4.1. Interpretation of Resource Footprint Results
Utilizing the derived
flow matrix, we computed the resource footprints of copper, nickel, molybdenum, zinc, and cobalt across 31 Chinese provinces for the years 2012, 2015, and 2017 via the EE-MRIO model. The results are displayed in
Figure 5.
The combined footprint of all assessed mineral resources exhibited a significant surge over the five-year period from 2012 to 2017. This increase underscores China’s economic development, industrial upgradation, and urbanization during this period, resulting in amplified demand for various clean energy minerals. However, this growth also insinuates that China faces heightened pressure and challenges concerning environmental conservation and resource sustainability. Notably, the copper resource footprint swelled from 46,545,520 tons in 2012 to 120,655,120 tons in 2017, marking a 159.47% increase; the nickel resource footprint climbed from 1,145,800 tons in 2012 to 2,424,900 tons in 2017, a 111.68% rise; the molybdenum resource footprint expanded from 32,684,500 tons to 103,253,100 tons, a substantial 215.86% increase; the zinc resource footprint ascended from 35,093,600 tons to 123,609,100 tons, a hefty increase of 252.08%; finally, the cobalt resource footprint grew from 31,700 tons to 79,100 tons, representing a 149.7% surge.
Of note is the fact that, despite the overarching upward trajectory of the resource footprints of the five mineral resources, significant variations persist amongst the different resources. Specifically, copper and molybdenum displayed minor declines during certain intervals. The resource footprint of copper exhibited a decrease of 20.10%, declining from 46,545,500 tons to 37,190,000 tons between 2012 and 2015. Concurrently, the resource footprint of molybdenum underwent a decrease of 14.90%, from 121,335,800 tons to 103,253,100 tons between 2015 and 2017. This trend potentially correlates with China’s industrial restructuring efforts during this period and the introduction of key policies, such as the Guidance on Accelerating Energy Conservation and Emission Reduction from 2015 issued by the Chinese State Council.
The resource footprints of nickel and cobalt experienced modest growth from 2012 to 2015, with increases of 23,200 tons for nickel, representing a 2.03% growth, and 10,400 tons for cobalt, marking a 32.83% rise. Yet, from 2015 to 2017, their growth rates surged noticeably. Nickel’s footprint expanded by 1,255,800 tons, a growth of 107.41%, while cobalt’s footprint grew by 37,000 tons, an increase of 88.01%. This pronounced acceleration can be linked to the evolution of China’s new energy automobile industry, advancements in high-end manufacturing, and other emerging sectors, coupled with a rising demand for batteries, stainless steel, and various other products [
58,
59].
Examining geographical distribution, we observe significant disparities across different regions in terms of distinct mineral resource footprints, and a discernible shift in related industries. Primarily, the resource footprints of copper, molybdenum, and zinc are progressively migrating towards the western region. For instance, between 2012 and 2017, Guizhou Province’s copper resource footprint soared from 98,500 tons to 12,162,300 tons, Hunan Province’s increased from 2,695,900 tons to 12,188,600 tons, and Yunnan’s escalated from 2,748,100 tons to 23,142,600 tons. This shift is possibly due to the industrialization and urbanization processes [
60] in the western region fueling the demand for these resources. Additionally, the region’s abundant resource reserves contribute to lower production and operational costs [
61]. Secondarily, the resource footprint of nickel and cobalt is transitioning towards economically advanced regions, such as Tianjin, which observed an increase in its nickel resource footprint from 4700 tons to 430,000 tons. This transition suggests that the ongoing upgrade of the industrial structure in the economically affluent eastern and coastal regions is catalyzing the demand for nickel, cobalt, and other metals in sectors such as new energy and high-end manufacturing.
Between 2012 and 2017, a discernible growth in the resource footprints across various sectors in China is evident. Notably, the construction sector (S54 in 2012 and S53 in 2017) displayed a remarkable escalation in its footprint, surpassing all other sectors. Given the multitude of sectors, this analysis prioritizes the downstream sectors’ footprints, as illustrated in
Figure 6. During this period, significant growth was also observed in sectors such as finance (S58), real estate (S59), manufacture of transport equipment (S44), public administration, social insurance, and social organizations (S68), as well as the manufacture of leather, fur, feather, and related products (S34).
Figure 7 delineates the distribution of resource footprints for copper, nickel, molybdenum, zinc, and cobalt across various downstream sectors in 2012 and 2017. Beyond the construction sector, cobalt’s footprint exhibited pronounced growth in sectors like manufacture of transport equipment, processing of petroleum, coking, processing of nuclear fuel, and education. Nickel predominantly displayed an upward trend in the manufacture of transport equipment, education, and processing of petroleum, coking, processing of nuclear fuel sectors. The footprint of copper was amplified within the real estate, finance, and manufacture of transport equipment sectors. Zinc manifested considerable growth in the finance, real estate, and manufacture of leather, fur, feather, and related products sectors. Concurrently, molybdenum’s footprint saw conspicuous growth across transport, storage, and postal services, finance, and real estate sectors.
The rapid growth in the construction sector’s resource footprint underscores China’s accelerated urbanization and infrastructure development during this period. As urbanization deepened and populations converged, there was a surge in demand for apartments, roads, bridges, and other fundamental infrastructures, exerting profound environmental pressures. In the non-manufacturing domain, particularly in the finance (S58) and real estate (S59) sectors, there was a notable footprint expansion. The prosperity of the finance sector, active capital movements, and the associated demand for technology and electronic devices are potential drivers for this surge. Concurrently, the boom in real estate not only emanates from rising residential needs but more from commercial properties and large-scale infrastructure projects, intensifying mineral resource consumption and subsequent environmental strains. In manufacturing, the manufacture of transport equipment sector (S44) witnessed a significant footprint rise, aligned with China’s hefty investments in innovative technologies like electric vehicles and high-speed railways, which heavily rely on specific minerals. Additionally, the growth in the manufacture of leather, fur, feather, and related products (S34) could be tethered to evolving consumer demands, fashion trends, and lifestyle shifts. As for supply and service sectors like transport, storage, and postal services (S56), education (S65), and public administration, social insurance, and social organizations (S68), their footprints also expanded. While these sectors might not be directly tied to resource production or consumption, they indeed impart indirect environmental impacts by spurring economic activities in other sectors.
4.2. Uncertainty Analysis Using the Non-Parametric Monte Carlo
In this study, we employed a non-parametric Monte Carlo simulation approach to delve into the uncertainties of the resource footprint. Drawing on historical data of ore grades and concentrate grades of clean energy minerals from 2012 to 2021, we utilized the quantile matching technique for random sampling, thereby generating the required random variable inputs for our simulation. For each set of such randomly generated inputs, the EE-MRIO model was applied to evaluate the potential resource footprint of clean energy minerals across different provinces. After completing 10,000 iterative random samplings and simulations, we accumulated a range of simulation results. Based on these, we further derived the average values, standard deviations, and the associated confidence intervals. It is worth noting that due to the absence of historical data regarding the ore grades and concentrated grades of cobalt, our uncertainty analysis solely encompasses the resource footprints of copper, nickel, molybdenum, and zinc.
Figure 8 visualizes the Monte Carlo simulation results for copper, nickel, molybdenum, and zinc resource footprints for 2012, 2015, and 2017 across all Chinese provinces. In addition to extracting the mean and standard deviation from the Monte Carlo simulation, we also computed the coefficient of variances (CV), a measure of relative volatility, derived from the ratio of the standard deviation to the mean. A small CV indicates a concentrated set of results, while a larger CV denotes a more scattered outcome.
Considering the overall relative standard deviation results, the CV for the resource footprints of copper, nickel, molybdenum, and zinc is relatively low. This low variability suggests that the EE-MRIO-derived resource footprints of clean energy minerals are both stable and reasonable. For instance, the CV for the copper resource footprint across all Chinese provinces falls within the 6% to 7% range, suggesting that variations in copper mining and production processes, although present, are not overly dramatic.
Similarly, the coefficients of variation for the molybdenum and zinc footprints are 10–11% and 10–13%, respectively. These values are slightly higher than those of copper but are still relatively low. This indicates general stability in the molybdenum and zinc resource footprints, despite minor fluctuations. In contrast, the CV for the nickel resource footprint spans a wider range, from 14% to 17%, which exceeds those of copper, molybdenum, and zinc. This suggests a relatively higher volatility in the nickel resource footprint, possibly due to the multiplicity of influences impacting nickel production processes and the considerable range of variability for these factors.
In the context of standard deviation, the inherent uncertainty in the resource footprint of the four investigated minerals exhibits significant variation across mineral types and temporal dynamics. Regardless of the specific mineral, it is observed that provinces with extensive resource footprints consistently demonstrate larger standard deviations. This correlation suggests that an expansion in production scale could potentially induce an increase in the complexity of the associated production processes. Such complexity could originate from a multitude of factors, including but not limited to, the multiplicity of extraction sites, disparities in extraction and processing technologies, and an increased number of supply chain participants. These elements are likely to introduce greater variability into the production process, leading to an increase in the standard deviation.
Furthermore, we employed Monte Carlo simulations to ascertain the median and confidence intervals of our dataset, as depicted in
Table 3. At a national level, the mean values derived from the simulations closely approximate the median for the period 2012 to 2017, irrespective of the mineral species under consideration. Concurrently, the lower confidence interval for the copper resource footprint exhibits a reduction from the mean of −9.27% to −9.30%, while the upper limit escalates by 16.26% to 16.27%. Comparable trends are observed for the nickel, molybdenum, and zinc resource footprints. The extremes of the confidence interval consistently lie below 25% of the mean, indicative of a relatively confined confidence interval.
In addition, we undertook a differential analysis juxtaposing the resource footprint calculated by the EE-MRIO model, and the mean resource footprint derived from the Monte Carlo simulations, as illustrated in
Figure 9. It is evident that usually, the calculated results bear a strong resemblance to the simulated outcomes, suggesting that the EE-MRIO model is proficient at generating estimates that closely align with the results from Monte Carlo simulations. The divergence in the resource footprints of copper, nickel and molybdenum was relatively modest and within acceptable bounds. However, a substantial deviation was noted in the zinc resource footprint for 2012, accounting for 29.6% of the simulated results.
This significant discrepancy primarily stems from the differences in core methodologies and data sources between the two models. The EE-MRIO model relies on economic activity and trade data from specific years, thereby precisely reflecting the market dynamics of that year. In contrast, the Monte Carlo model leans heavily on historical data, aiming to forecast long-term trends and associated uncertainties. The close alignment between the simulated and calculated results of the Monte Carlo model for the years 2015 and 2017 suggests that it successfully captures the overall market trend. Given that ore grade and concentrate grade are key parameters, the data deviation in 2012 might imply that high-quality zinc ore was extensively mined that year. This further indicates that, with the rapid extraction of high-quality zinc ore, the mining industry is gradually shifting towards lower-grade sources. This shift might be associated with the depth, scale, and type of the ore deposit as well as its grade fluctuations, suggesting that the primary source of uncertainty in the Monte Carlo model is geological factors.
4.3. Random Forest Regression Results
4.3.1. Model Results
To ensure the robustness of our model and to address the inherent uncertainties in Bayesian optimization, we implemented a strategy of multiple hyperparameter searches. Specifically, we conducted 10 independent Bayesian optimization processes, each aimed at discovering the optimal combination of hyperparameters. In each optimization, we evaluated the objective function up to 100 times, allowing the algorithm to search for the best solution across a broad parameter space. The result of each optimization was a specific set of hyperparameters, which were then used to train and evaluate the random forest model.
Table 4 lists these optimal hyperparameter combinations.
To determine the best model, we compared the performance of models under different hyperparameter combinations using a series of key performance indicators, including root mean square error (RMSE), mean absolute error (MAE), mean squared error (MSE), and the coefficient of determination (R
2). Lower values of RMSE, MAE, and MSE typically indicate smaller prediction errors and higher predictive accuracy, while a R
2 value close to one suggests a strong capability of the model in explaining data variability.
Figure 10 provides a detailed revelation of the performance metrics and the relative importance of features of the selected best model. We thoroughly considered these indicators in the specific context of our data and research to ensure a deep and accurate understanding of the model’s accuracy. Further analysis of these results revealed that the footprints of different clean energy mineral resources are influenced by various factors. In particular, the total population significantly affects the resource footprints of copper, molybdenum, and zinc, while technology market turnover has a greater impact on the resource footprints of nickel and cobalt.
The resource footprint of copper is notably shaped by total population and energy consumption, which carry characteristic importance values of 0.8472 and 0.7076, respectively. This could be attributable to the prevalent use of copper in electrical and electronic products, where copper demand is escalating in line with population growth and rising energy needs. Study [
49] highlights the potential interplay of population migration on copper’s resource dynamics, further affirming our observations. The model returned an R
2 value of 0.7311, a MSE of 0.5702, a MAE of 0.5739, and a RMSE of 0.7551. These metrics indicate that the model effectively encapsulates the relationship between these variables and the copper resource footprint. For sustainable resource management, due consideration should be given to the implications of population growth and energy policies.
When considering the resource footprint of nickel, the primary influencing factors appear to be technology market turnover and total population, with respective characteristic importance values of 0.9043 and 0.5198. This is likely due to nickel’s widespread application in high-tech industries, such as battery manufacturing. The burgeoning technology markets, coupled with an increasing population utilizing these products, may have spurred nickel demand. However, the model’s R2 value of 0.6426, an MSE of 1.5173, MAE of 1.0249, and RMSE of 1.2318 suggest that additional, unexplored factors might also be influential, necessitating further research.
The resource footprint of molybdenum is primarily driven by the total population, followed by total carbon emissions, exhibiting characteristic importance values of 0.9155 and 0.4565, respectively. With its prevalent use in the steel and chemical industries, the demand for molybdenum is poised to escalate as population growth spurs industrial activity. Nonetheless, the model’s R2 value of 0.5861, an MSE of 0.8347, MAE of 0.7711, and RMSE of 0.9136, suggest that the resource footprint of molybdenum may be shaped by an intricate interplay of several factors.
For zinc, mirroring copper, the pivotal factors are the total population and energy consumption, with respective characteristic importance values of 0.7934 and 0.5544. Given Zinc’s extensive utilization in construction and infrastructure, it exhibits a strong correlation with population growth and energy demand. The model reports an R2 value of 0.7307, an MSE of 0.7083, MAE of 0.6783, and RMSE of 0.8416, which, like copper, signify that the model adequately captures the essence of these relationships.
Lastly, the resource footprint of cobalt is largely dictated by technology market turnover, with an importance of 1.0568, likely tied to cobalt’s role in advanced technologies like lithium-ion batteries. The study [
62] reinforces this viewpoint, suggesting that technological and industrial advancements are reshaping the demand for nickel and cobalt. GDP also emerges as a significant determinant, with an importance of 0.4928, potentially reflecting the influence of economic activities on cobalt demand. The model yields an R
2 value of 0.6945, an MSE of 1.8703, MAE of 1.1748, and RMSE of 1.3676, indicating that, while the model has some degree of predictive power concerning the cobalt resource footprint, there remains room for improvement.
4.3.2. Robustness Testing
To further evaluate the generalization ability and stability of our model, we adopted a random seed-based data partitioning method. This technique effectively mirrored various real-world data scenarios, ensuring consistent performance of our model across diverse datasets. In our robustness testing, we utilized the ten sets of hyperparameters previously identified through Bayesian optimization, conducting 50 iterative training and evaluation cycles. Each cycle employed a distinct random seed to generate new splits of the data, ensuring the representativeness and diversity of each training and testing scenario. This approach differs from data shuffling; in every iteration, it produces an entirely new combination of data, thereby ensuring the representativeness and diversity of the training and testing sets.
Table 5 presents the average results for key performance metrics such as R
2, MAE, MSE, and RMSE under various hyperparameter combinations, affirming the model’s adaptability and generalization across different data partitioning scenarios.
Significantly, the performance parameters for the five resource footprints exhibited relatively marginal changes across the ten modeling iterations. Specifically, these performance parameters remained largely stable, save for the R2 value of the cobalt resource footprint, which depreciated from 0.6945 to 0.5964, and its corresponding MSE, which escalated from 1.8703 to 2.7038. Despite these shifts, the average R2 value for the cobalt resource footprint hovers near 0.6, suggesting that the model retains a certain degree of explanatory power. Even though hyperparameter modifications may influence model performance, the overarching trend and the significance of key features maintain their relative constancy across different hyperparameter combinations. Total population persists as the most salient factor influencing the resource footprints of copper, molybdenum, and zinc, while the technology market turnover chiefly impacts the resource footprints of nickel and cobalt. This consistency indicates that our chosen optimal model is reliable, with its findings demonstrating robustness. While individual models may be swayed by the selection of specific hyperparameters, the general trends and conclusions remain steadfast across a spectrum of configurations.