Scholarlink

<p>The significance of forecasting carbon emission is critically important for region like South and Southeast Asia due to rapid economic growth, high population density, and diverse socio-political landscapes. In this study, we develop and evaluate the efficiency of deep learning models to predict national CO<sub>2</sub>-per-capita by integrating socioeconomic data and different climate change scenarios. To capture the non-linear relationships in the data we employed three model architectures including Artificial Neural Networks (ANN), Convolutional Neural Networks (CNN), and Long Short-Term Memory (LSTM). Our findings show exceptional prediction of performance achieving a R<sup>2</sup> score of up to 99.7% for CNN, and mean absolute percentage error of 2.31 for LSTM. The SHAP analysis identified GDP and population as key drivers of emissions suggesting policymakers to address demographic trends and economic structure for a successful transition to a low carbon economy. To evaluate the robustness of our model, we utilized K-fold cross-validation, where CNN demonstrated greater reliability with the highest stability compared to the other models. We further projected future CO<sub>2</sub> trajectories under SSP1 (Sustainability), SSP2 (Middle of the Road), and SSP5 (Fossil-fueled Development) scenarios through 2050. It reveals moderate (0.3-1.5%) annual emission increases for high-income nations (e.g., Thailand, Malaysia, and Singapore). On the contrary, Laos, Vietnam, and Brunei may face higher risks of elevated per capita CO<sub>2</sub> emissions, reaching 28.15 ± 16.26 metric tons. Combining deep learning with SSPs, this research offers a robust predictive framework and actionable insights for policymakers and stakeholders, providing a roadmap for steering resilient and climate-aligned development in South and Southeast Asia.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Modeling future CO2 trajectories in fast-growing Asian economies: synergizing deep learning and shared socioeconomic pathways

  • 期刊论文

  • 发表日期：

    2026年4月14日

查看全文
  • Bijoy Mitra,
  • Mohammed Sakib Uddin,
  • Farzana Jamal,
  • Khaled Mahmud,
  • Syed Masiur Rahman,
  • Mohammad Shahedur Rahman,
  • Muhammad Muhitur Rahman

摘要

The significance of forecasting carbon emission is critically important for region like South and Southeast Asia due to rapid economic growth, high population density, and diverse socio-political landscapes. In this study, we develop and evaluate the efficiency of deep learning models to predict national CO2-per-capita by integrating socioeconomic data and different climate change scenarios. To capture the non-linear relationships in the data we employed three model architectures including Artificial Neural Networks (ANN), Convolutional Neural Networks (CNN), and Long Short-Term Memory (LSTM). Our findings show exceptional prediction of performance achieving a R2 score of up to 99.7% for CNN, and mean absolute percentage error of 2.31 for LSTM. The SHAP analysis identified GDP and population as key drivers of emissions suggesting policymakers to address demographic trends and economic structure for a successful transition to a low carbon economy. To evaluate the robustness of our model, we utilized K-fold cross-validation, where CNN demonstrated greater reliability with the highest stability compared to the other models. We further projected future CO2 trajectories under SSP1 (Sustainability), SSP2 (Middle of the Road), and SSP5 (Fossil-fueled Development) scenarios through 2050. It reveals moderate (0.3-1.5%) annual emission increases for high-income nations (e.g., Thailand, Malaysia, and Singapore). On the contrary, Laos, Vietnam, and Brunei may face higher risks of elevated per capita CO2 emissions, reaching 28.15 ± 16.26 metric tons. Combining deep learning with SSPs, this research offers a robust predictive framework and actionable insights for policymakers and stakeholders, providing a roadmap for steering resilient and climate-aligned development in South and Southeast Asia.