Wang et al. (2025) Harnessing TabTransformer Model and Particle Swarm Optimization Algorithm for Remote Sensing-Based Heatwave Susceptibility Mapping in Central Asia

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Atmosphere
• Year: 2025
• Date: 2025-10-07
• Authors: Antao Wang, Linan Sun, Huicong Jia
• DOI: 10.3390/atmos16101166

Research Groups

Not explicitly stated in the provided text.

Short Summary

This study introduces a fully remote sensing-based framework for mapping heatwave susceptibility using a Particle Swarm Optimization (PSO)-optimized TabTransformer deep learning model. The framework successfully achieves accurate, scalable, and spatially detailed heatwave susceptibility mapping in data-scarce Central Asian regions, outperforming a baseline model and identifying key environmental predictors.

Objective

• To determine if a fully remote sensing-driven, PSO-optimized TabTransformer deep learning model can achieve accurate, scalable, and spatially detailed heatwave susceptibility mapping in data-scarce regions such as Central Asia.

Study Configuration

• Spatial Scale: Five Central Asian countries (Turkmenistan, Uzbekistan, southern Kazakhstan, and adjacent lowlands), producing high-resolution susceptibility maps.
• Temporal Scale: Not explicitly stated for the study period, but utilizes ERA5-derived heatwave evidence and predictors.

Methodology and Data

• Models used: TabTransformer deep learning model, Particle Swarm Optimization (PSO) for hyperparameter tuning.
• Data sources: ERA5-derived heatwave evidence and thirteen environmental and socio-economic predictors (implied to be remote sensing-based or derived from remote sensing for the "fully remote sensing-based framework").

Main Results

• The PSO-optimized TabTransformer model significantly outperformed the standalone TabTransformer.
  • Optimized model: RMSE = 0.123, MAE = 0.034, R2 = 0.938, AUROC = 0.940.
  • Standalone model: RMSE = 0.132, MAE = 0.038, R2 = 0.93, AUROC = 0.933.
• The PSO-tuned model demonstrated faster convergence, lower final loss, and more stable accuracy during training and validation.
• Spatial outputs revealed heightened susceptibility in southern and southwestern sectors, including Turkmenistan, Uzbekistan, southern Kazakhstan, and adjacent lowlands.
• Statistical tests (Chi-squared, Friedman, Wilcoxon) confirmed statistically significant improvements in spatial precision and class delineation (p-values < 0.0001).
• Feature importance analysis consistently identified maximum temperature, frequency of hot days, and rainfall as the dominant predictors of heatwave susceptibility.

Contributions

• Pioneers a fully remote sensing-based framework for heatwave susceptibility mapping.
• Integrates the TabTransformer deep learning model with Particle Swarm Optimization for robust hyperparameter tuning in environmental hazard assessment.
• Addresses the critical need for accurate and scalable heatwave susceptibility mapping in data-scarce regions, particularly Central Asia.
• Validates the potential of data-driven, deep learning approaches for reliable environmental hazard assessment, crucial for climate adaptation planning.

Funding

Not explicitly stated in the provided text.

Citation

@article{Wang2025Harnessing,
  author = {Wang, Antao and Sun, Linan and Jia, Huicong},
  title = {Harnessing TabTransformer Model and Particle Swarm Optimization Algorithm for Remote Sensing-Based Heatwave Susceptibility Mapping in Central Asia},
  journal = {Atmosphere},
  year = {2025},
  doi = {10.3390/atmos16101166},
  url = {https://doi.org/10.3390/atmos16101166}
}