GX & Decarbonization Research

Zhonghao He, Haihong Jiang, Jing He +7

This paper proposes a process-aware deep learning approach for low-cost greenhouse gas sensing. Using composting as a case study, it incorporates process information to improve sensing accuracy. The method is scalable to other anthropogenic…

Augustine Okeke, Ifeanyi Ugbebor

This study introduces the SDG-Purpose Alignment Index (SPAI), a computational construct quantifying how CEO letters align thematically, tonally, and stylistically with specific SDG targets. Analyzing 740 CEO letters from 148 listed firms ac…

Jiaxiang Wang, T X Chen

This study uses MGWR-SHAP analysis to evaluate the scale-dependent and non-linear effects of land-cover configuration on carbon performance in Tianjin, China. It reveals that different land cover types have varying impacts at different spat…

Олег Ивченко, Iryna Ivchenko

This paper proposes reporting standards for energy consumption and carbon footprint of open-source AI training. It presents specific metrics and methodologies to enhance transparency, contributing to decarbonization in the AI sector.

T Zhang, Ting Jin, Kang Wu +1

This paper proposes a Temporal Dendritic Neural Model (TDNM) with a Dendritic Adaptive Learning algorithm for multivariate carbon emission forecasting. Using panel data from 54 Chinese cities (1999-2023) with 13 driving factors, it outperfo…

Q Zhang, 胜生 余, Tongna Liu

This paper provides a reproducibility artifact (v2.0.0) for an environmental decision model evaluating carbon-water trade-offs in geo-distributed AI workload allocation. It includes portable Python code and test data, supporting only scenar…

Priyanka Aggarwal, Archana U Singh, Deepali Malhotra

This study analyzes sustainability reports of 269 NIFTY 500 firms from FY2015-2023 using text mining. It identifies three phases of ESG reporting evolution: initial establishment (2015-17), governance consolidation (2018-20), and balanced i…

Kyung Bae Jang, Tae Ho Woo

This study uses a system dynamics approach to analyze how generative AI optimizes nuclear energy's contribution and mitigates vulnerabilities of carbon-intensive sources. Test 1 (low initial value) outperformed others, with the highest sens…

Pieter de Jong, Inga Timmerman, Alona Bilokha

This study introduces the Claim-Reality Divergence (CRD) index, combining satellite methane data (Sentinel-5P TROPOMI) with NLP analysis of SEC filings and earnings calls for 233 US heavy emitters (2018-2023). A one standard deviation incre…

Nguyen V.

This paper proposes a method to learn domain-specific embeddings for ESG question answering. It uses a retrieval-augmented generation framework to adapt embeddings to the ESG domain, enabling accurate extraction of answers from ESG document…

‬‬‬Mohammad Khaled al-Bashiti, M.Z. Naser

This paper presents an autonomous ML approach using Bayesian optimization and NSGA-II/III to optimize green concrete mixtures. Analyzing over 2,300 real mixes, it achieves an average 40% reduction in CO2 emissions and 30% reduction in mater…

Trésor Ndengo Tietia, Rostin Mabela Matendo Makengo, Pierre Kasengedia Motumbe +4

This paper proposes a smart network resource management model using AI (DQN, Q-Learning, PSO, ACO) to optimize energy consumption of IT infrastructures in the Democratic Republic of Congo. Simulations show up to 32% energy reduction and 21.…

Saher Elsayed

This paper presents the first phase-resolved carbon footprint characterization of RLHF, finding that the PPO phase accounts for 59-64% of total training emissions. It introduces CarbonAware-RLHF, a scheduling framework that adaptively shift…

Q Zhang, 胜生 余, Tongna Liu

This reproducibility artifact provides portable Python code, derived inputs, frozen results, and tests for a carbon-water trade-off analysis in geo-distributed AI inference allocation under water-stress screening. It enables verification of…

Niko Firzi Anansyah, Ratu Mutiara Siregar, Andi Prayogi +1

This study develops an automated ESG reporting system for the palm oil industry using a dual-LLM architecture (Groq Llama 3 primary, Gemini backup). It introduces a programmatic constraint that caps the ESG score at 50 if Ganoderma infectio…

Krisztina Finta, Csaba Dezső Dér, Klaudia Gabriella Horváth +1

This study investigates the sustainability paradox in AI-driven short-form video marketing, revealing that video generation consumes ~30x more energy than image creation. It models that a mid-sized team producing 2,500 AI videos annually em…

Saheed A. Adeniyi, Adeyemi A. Adefisan-Williams, Rukayat Abisola Olawale, Daniel O. Olokpo and Elizabeth I. Igbodor

This study develops a digital twin coupling a physics-based drying-kinetics engine with an autoencoder-driven predictive maintenance engine for a hybrid solar-electric dehydrator in tropical Nigeria. It predicted moisture ratio with RMSE 0.…

Damla Yalçıner Çal, Ecir Uğur Küçüksille

This study uses the EDGAR dataset (1970-2023) to forecast GHG emissions with LSTM and Transformer models. While LSTM captures long-term trends, it underperforms a naive benchmark under regime changes. Transformer shows improvement for the l…

Jiachen Shou, Waner Li, Hui Li +5

This study integrates Random Forest and DEMATEL to identify key drivers of transportation carbon emissions across 30 Chinese provinces from 1997 to 2022. Findings show that R&D expenditure, registered scientific achievements, and energy con…

Shimei Li, Yuqing Li, Jinman Luo +1

This paper proposes a dynamic optimization strategy for carbon emissions in an electrocarbon coupling system using reinforcement learning. The approach enables real-time emission reduction, outperforming static methods. It contributes to sy…