Time: Mar 2026 – Jun 2026
Authors: Junxiong Lai (First author), Peiyuan Xu, Yixian Huang*
Venue: 2026 4th International Conference on Remote Sensing, Mapping and Geographic Information Systems (RSMG 2026) · Accepted; proceedings to be published by SPIE (indexed by EI Compendex, Scopus)
Summary: Taking the Zijinshan Gold-Copper Mine in Shanghang County, Fujian Province as the case study, this paper constructs a dynamic monitoring framework for ecological disturbance in mining areas that integrates the Remote Sensing Ecological Index (RSEI), GIS spatial-constraint features, and a machine-learning classifier (GIS-LightGBM). Ecological disturbance is classified into five levels using multi-temporal Sentinel-2 and Landsat observations (2018–2024) together with terrain, distance, land-use and neighborhood texture features. Under spatial block validation, the GIS-LightGBM model achieves the best performance, with Accuracy of 0.934, Macro-F1 of 0.923, Kappa of 0.901 and Balanced Accuracy of 0.927, providing a technical reference for disturbance identification, restoration-priority delineation and remote-sensing dynamic monitoring in similar metal mining areas.
📄 Acceptance Letter (PDF)

Time: Jan 2026 – Present
Authors: Peiyuan Xu, Junxiong Lai, Lijie Liu*, Xunyu Hu, Wenyuan Liu
Venue: Journal of Geochemical Exploration (JCR Q1, IF=4.6, manuscript submitted)
Summary: This study applies seven machine-learning models (SVM, MLP, TabPFN, HyperFast, FT-Transformer, TabR and CatBoost) to 2,269 pyrite trace-element datasets collected from 17 representative HS–IS–LS epithermal gold deposits worldwide to discriminate deposit subtypes. All models achieve accuracy above 93%, with TabPFN, CatBoost and FT-Transformer exceeding 95%. SHAP analysis identifies Cu, Au, Mo, Sb, Te and Ni as the core discriminant elements.

Time: Aug 2025 – Present
Authors: Le Huang†, Zekun Su†, Junxiong Lai†, Ziyi Shi*
Venue: Scientific Reports (JCR Q1, IF=4.9, manuscript submitted)
Summary: Using panel data for 31 provincial-level regions of mainland China (2014–2023) and CPC Y02 patent applications as the measure of carbon-neutral technology innovation (CNTI), this study integrates spatial Markov chains, the spatial Durbin model and geographically and temporally weighted regression. We found that provincial CNTI exhibits significant positive spatial correlation and strong path dependence, and that medium-innovation provinces are more likely to transition upward in higher-innovation neighborhoods. R&D investment is positively associated with local innovation, while positive cross-regional spillovers are more pronounced than the direct local effect, and these effects remain robust under alternative innovation measures; local associations are generally positive, with non-stationarity concentrated mainly over time.