Multi-task deep learning framework for forecasting CO2 and temperature using Conv1D-LSTM with explainable AI

Author First name, Last name, Institution

• Nadia Dahmani, Zayed UniversityFollow
• Amril Nazir, Zayed University
• Syed M.Salman Bukhari, Capital University of Science & Technology
• Anood Ibrahim, Zayed University

Document Type

Article

Source of Publication

Neural Computing and Applications

Publication Date

3-1-2026

Abstract

Climate change is driven by rising atmospheric carbon dioxide (CO₂) levels, presents severe environmental and socioeconomic risks. To address these challenges, this study proposes the multi-task learning (MTL) framework that combines one-dimensional convolutional neural networks (Conv1D) and long short-term memory (LSTM) networks to simultaneously predict CO₂ emissions and temperature variations. The framework utilizes shared feature representations to model complex interdependencies between environmental factors, thereby enhancing predictive accuracy and generalization. The robustness of the framework is evaluated using time-series cross-validation (walk-forward) and a separate test set. On the test set, the model achieves exceptional predictive performance, with R² values of 0.993 for temperature and 0.990 for CO₂, and low Mean Absolute Error (MAE) scores of 0.0133 for CO₂ and 0.0155 for temperature. In addition, statistical analyses, including Welch’s t-test, confirm the reliability and consistent performance of the model across both tasks, with minimal variance between folds. A key feature is the integration of explainable AI (XAI) using Local Interpretable Model-Agnostic Explanations which provides transparent, actionable insights into the key drivers. This transforms the model from a simple predictive tool into one that can directly inform real-world policy and decision-making. Ultimately, this study shows that integrating deep learning with XAI generates accurate, transparent, and stakeholder-relevant tools for climate forecasting, supporting the development of sustainable, data-driven strategies to mitigate climate change and enhance climate resilience.

DOI Link

10.1007/s00521-026-12039-6

ISSN

0941-0643

Publisher

Springer Science and Business Media LLC

Volume

38

Issue

6

Disciplines

Computer Sciences

Keywords

Climate Forecasting Environmental Modeling CO2 Emissions, Climate Resilience, Explainable AI, LSTM, Multi-Task Learning Conv1D

Scopus ID

105033866255

Recommended Citation

Dahmani, Nadia; Nazir, Amril; Bukhari, Syed M.Salman; and Ibrahim, Anood, "Multi-task deep learning framework for forecasting CO2 and temperature using Conv1D-LSTM with explainable AI" (2026). All Works. 7950.
https://zuscholars.zu.ac.ae/works/7950

Indexed in Scopus

yes

Open Access

no