Estimating the Material Footprint at the National Level from 1993 to 2022 Based on Multi-Feature CNN-BiLSTM

Lizhi Miao, Yannan Wang, Kaiwen Wu, Lei Huang, Mei-Po KwanISPRS Int. J. Geo Inf 2025 · DOI

TL;DR

This paper introduces a novel CNN-BiLSTM deep learning model to accurately estimate and predict national material footprints from 1993-2022. The model outperforms traditional methods, offering crucial insights into global resource consumption trends and informing sustainable development strategies.

What Problem Does This Paper Address?

Global environmental issues are escalating, making accurate measurement of environmental pressure and sustainable resource utilization critical. The "material footprint" is a key indicator of this pressure, representing the total amount of material extraction required to meet a nation's consumption. However, traditional methods for calculating material footprints (like Material Flow Analysis and input-output tables) are often hampered by data collection challenges, high computational complexity, and an inability to predict future trends. This paper aims to overcome these limitations by developing a more robust and predictive framework using deep learning.

Methods

The core of this research is the development of a Multi-Feature CNN-BiLSTM model. This hybrid approach combines:

  • Convolutional Neural Networks (CNNs): Used for feature extraction from the input data, helping to reduce dimensionality and mitigate overfitting. CNNs excel at identifying local patterns within grid-like data.
  • Bidirectional Long Short-Term Memory (BiLSTM) networks: These are advanced recurrent neural networks designed to handle sequential data. BiLSTMs process information in both forward and backward directions, allowing them to capture long-term dependencies and context from both past and future states, which is crucial for time-series prediction.

The model preprocesses data by normalizing it and then feeding it into a four-dimensional structure required by the CNN layers. The entire process involves data collection, grey relational analysis (GRA) to identify key influencing factors, multicollinearity testing, normalization, dataset partitioning, model training, and performance evaluation using metrics like R² and NRMSE.

Datasets

The study utilizes data from 77 countries or regions spanning the period from 1993 to 2022. The primary data sources include:

  • Material Footprint: From the Global Material Flows Database.
  • GDP: From the World Bank.
  • Oil Consumption, Natural Gas Consumption, Coal Consumption: From the Energy Institute's Statistical Review.
  • Population and Urbanization Rate: Data for China from its National Bureau of Statistics, and for other economies from the World Bank.

The selection of these 77 countries is justified by their representation of approximately 78% of the world's total population, ensuring a degree of global representativeness.

Key Insights and Ideas

  • Deep Learning for Material Footprints: The paper demonstrates that deep learning, specifically the CNN-BiLSTM architecture, can significantly improve the accuracy and predictive power of material footprint estimations compared to traditional methods.
  • Interconnectedness of Factors: Grey relational analysis confirmed strong positive correlations between material footprint and influencing factors like urbanization, oil, natural gas, and coal consumption.
  • Global Disparities: The study highlights significant differences in material footprints, with China, the EU, and the US consistently showing the highest total footprints, while Iceland has the lowest. Qatar leads in per capita material footprint, with Pakistan at the bottom.
  • G20 Dominance: A substantial portion of countries with high per capita material footprints are G20 members, indicating their significant environmental pressure.
  • Dynamic Trends: The analysis reveals both increasing and decreasing trends in material footprints across different regions over the study period, with notable regional variations.

Key Results

  • The CNN-BiLSTM model achieved excellent predictive performance, with an R² of 0.861 and an NRMSE of 0.063. This represents a significant improvement over other comparative models (RNN, LSTM, BiLSTM).
  • China's mainland reported the highest total material footprint from 2013 to 2022, while Iceland had the least.
  • Qatar had the highest per capita material footprint, and Pakistan had the lowest.
  • Among the top 50% of countries by average annual per capita material footprint (2013-2022), 12 were G20 members, including all G7 nations except Italy.
  • The study identified absolute decoupling between economic and material footprints for six countries (Argentina, Australia, Germany, Canada, South Africa, Saudi Arabia), suggesting progress in sustainable development.

Limitations and Future Work

  • Data Quality and Availability: The accuracy of the model is dependent on the quality and completeness of input data, which can vary across economies, especially those with less developed data collection systems.
  • Assumption of Past Trends: The model assumes that past trends will continue, which might not hold true in rapidly changing economies or regions undergoing significant economic or technological transitions.
  • Complexity of Material Footprint: The material footprint is influenced by numerous interconnected factors (policy, technology, global events), and while the model captures many, it may not fully account for all unforeseen influences.
  • Limited Time Window: The study's focus on 1993-2022 might limit its ability to capture very long-term structural changes or emerging trends beyond this period.
  • Standalone Metric: The material footprint is analyzed as a standalone metric. Future research could integrate it with other environmental indicators (e.g., carbon emissions, water usage) for a more holistic view.
  • Global Coverage: The dataset does not encompass all global economies.

Future work should aim to expand data sources, introduce higher-resolution data, integrate real-time policy changes, and enhance the operational analysis of prediction results for policy-making.

Summary updated

BibTeX
@article{miao2025estimating,
  author = {Lizhi Miao and Yannan Wang and Kaiwen Wu and Lei Huang and Mei-Po Kwan},
  title = {{Estimating the Material Footprint at the National Level from 1993 to 2022 Based on Multi-Feature CNN-BiLSTM}},
  journal = {ISPRS Int. J. Geo Inf.},
  year = {2025},
  doi = {10.3390/ijgi14020086},
}
Figure