özden (2025) Data-Driven Decision Support in Environmental Management: Hybrid GNN-PINN Modeling of Subsurface Soil Temperature

Identification

• Journal: Mendeley Data
• Year: 2025
• Date: 2025-12-05
• Authors: özden, cevher
• DOI: 10.17632/kwj724c62f.1

Research Groups

• Cukurova Universitesi

Short Summary

This paper presents a comprehensive, preprocessed dataset of daily meteorological and subsurface soil temperature records from 15 diverse stations across Turkey, spanning five years (2020-2024), specifically designed to support research in environmental decision support systems and advanced hybrid GNN-PINN modeling.

Objective

• To provide a rigorously preprocessed dataset of daily meteorological and subsurface soil temperature records to enable the development and benchmarking of hybrid Physics-Informed Neural Networks (PINNs) and Graph Neural Networks (GNNs) for applications in soil temperature forecasting, thermal diffusion modeling, agricultural optimization, and climate generalization analysis.

Study Configuration

• Spatial Scale: 15 distinct meteorological stations across diverse climatic and geographic zones in Turkey (including arid central, coastal Mediterranean, continental highlands, and humid Black Sea regions).
• Temporal Scale: Daily resolution over a five-year period, from 1 January 2020 to 31 December 2024.

Methodology and Data

• Models used: The dataset is specifically designed to support the training and benchmarking of hybrid Physics-Informed Neural Networks (PINNs) and Graph Neural Networks (GNNs).
• Data sources: Daily meteorological records obtained from the Turkish State Meteorological Service (MGM) via the MEVBIS platform.
• Data components:
  • Subsurface Data: Daily mean, minimum, and maximum soil temperatures at five depths: 0.05 m, 0.10 m, 0.20 m, 0.50 m, and 1.00 m.
  • Surface Data: Daily mean 2-meter air temperature (°C).
  • Geospatial Metadata: Station-specific latitude, longitude, and altitude (meters).
• Data Processing:
  • Temporal Filtering: Removal of corrupted entries and sensor errors; exclusion of sequences with consecutive missing values exceeding three days.
  • Normalization: Min-max scaling of temperature values to the range [0, 1].
  • Structuring: Data structured to support depth encoding as a continuous input variable.

Main Results

• The paper successfully compiles and describes a comprehensive, multi-depth, multi-station dataset of daily soil and air temperatures for Turkey.
• The dataset includes daily mean, minimum, and maximum soil temperatures at five distinct depths (0.05 m, 0.10 m, 0.20 m, 0.50 m, 1.00 m) and daily mean 2-meter air temperature, along with geospatial metadata for 15 stations.
• Rigorous preprocessing steps, including temporal filtering and min-max normalization, have been applied to ensure the data's suitability for advanced computational modeling, particularly for hybrid GNN-PINN architectures.

Contributions

• The primary contribution is the creation and public release of a unique, rigorously preprocessed, and spatially diverse dataset specifically tailored for the development and benchmarking of hybrid GNN-PINN models in subsurface soil temperature dynamics.
• It provides a valuable five-year observational record from diverse climatic zones in Turkey, addressing a gap in publicly available datasets suitable for advanced machine learning applications in environmental decision support and precision agriculture.
• The dataset's structure and preprocessing facilitate research into soil temperature forecasting, thermal diffusion modeling, agricultural optimization, and climate generalization under extreme seasonal conditions.

Funding

Not specified in the provided text.

Citation

@article{özden2025DataDriven,
  author = {özden, cevher},
  title = {Data-Driven Decision Support in Environmental Management: Hybrid GNN-PINN Modeling of Subsurface Soil Temperature},
  journal = {Mendeley Data},
  year = {2025},
  doi = {10.17632/kwj724c62f.1},
  url = {https://doi.org/10.17632/kwj724c62f.1}
}