Related Experiment Video
Updated: Jul 12, 2026
Mining Spatial Transcriptomics Datasets using DeepSpaceDB
Published on: September 5, 2025
Multi-indicator water-quality prediction in mining areas using a feature-tokenizer transformer with spatiotemporal
Zihan Liu1, Xiang Sui1, Xianzhou Lyu1
1College of Earth Science and Technology, Shandong University of Science and Technology, Qingdao, 266590, PR China; State Key Laboratory of Disaster Prevention and Ecology Protection in Open-pit Coal Mines, Shandong University of Science and Technology, Qingdao, 266590, PR China.
A new Feature-Tokenizer Transformer (FT-Transformer) model accurately predicts mining area water quality across multiple indicators, outperforming existing methods. This advancement aids in assessing and managing water quality in complex mining environments.
Area of Science:
- Environmental Science
- Data Science
- Water Resource Management
Background:
- Mining impacts water quality, creating challenges for prediction in large, heterogeneous areas.
- Existing models struggle with stable, multi-indicator water quality prediction across diverse mining regions.
Purpose of the Study:
- To develop a robust multi-indicator framework for nationwide mining-area water quality prediction.
- To integrate diverse data sources for improved predictive accuracy and model understanding.
Main Methods:
- Developed a Feature-Tokenizer Transformer (FT-Transformer) multi-task framework.
- Integrated 55,744 monitoring records with spatial, mining, temporal, and temperature data.
- Jointly predicted pH, dissolved oxygen (DO), ammonium nitrogen (NH4-N), and permanganate index (CODMn) using cross-validation against benchmark models.
Main Results:
- The FT-Transformer achieved a mean R² of 0.790, outperforming Elastic Net, Random Forest, and XGBoost for most indicators.
- Achieved high R² values for NH4-N (0.826) and CODMn (0.820), with strong performance for pH (0.756) and DO (0.756).
- Multi-task learning improved joint predictions, and analyses revealed indicator-specific and context-dependent model behaviors.
Conclusions:
- The FT-Transformer provides a powerful tool for multi-indicator water quality prediction in heterogeneous mining areas.
- The framework supports model applicability screening and monitoring priority assessment.
- Understanding indicator-specific responses enhances environmental management strategies in mining regions.