Wang et al. (2026) Anomaly Detection and Correction for High-Spatiotemporal-Resolution Land Surface Temperature Data: Integrating Spatiotemporal Physical Constraints and Consistency Verification

Identification

• Journal: Remote Sensing
• Year: 2026
• Date: 2026-03-24
• Authors: Yun Wang, Mengyang Chai, Xiao Zhang, Huairong Kang, Xuanbin Liu, Siwei Zhao, Cancan Cui, Yinnian Liu
• DOI: 10.3390/rs18070972

Research Groups

• Shanghai Institute of Technical Physics, Chinese Academy of Sciences, Shanghai, China
• University of Chinese Academy of Sciences, Beijing, China

Short Summary

This paper proposes a two-stage anomaly detection and correction framework for high-spatiotemporal-resolution Land Surface Temperature (LST) data, integrating temporal physical constraints and spatial consistency verification. The method significantly enhances LST data quality by effectively distinguishing physically plausible weather changes from data errors, outperforming conventional statistical methods with substantial improvements in accuracy and correlation.

Objective

• To develop a robust and physically interpretable two-stage anomaly detection framework for high-spatiotemporal-resolution Land Surface Temperature (LST) data that prioritizes temporal physical pre-screening followed by spatial statistical verification, thereby reducing the risk of mis-correcting physically reasonable temperature variations and enhancing data reliability.

Study Configuration

• Spatial Scale: East Asia, with a spatial resolution of 0.02° (approximately 2 km). Validation conducted at 7 ground observation sites in the Heihe River Basin and 1 independent site (Huailai Station) in the Haihe River Basin. Spatial consistency tests used a 7 × 7 pixel window.
• Temporal Scale: Hourly LST data from 2016 to 2021, covering a 6-year period with a 1-hour temporal resolution.

Methodology and Data

• Models used:
  • Anomaly Detection: Two-stage framework:
    1. Temporal physical pre-screening: Piecewise empirical model (exponential decay and sinusoidal function) based on typical diurnal LST variation.
    2. Spatial consistency test: Robust Z-score calculated using Median Absolute Deviation (MAD) within a 7 × 7 pixel neighborhood.
  • Anomaly Correction: Inverse Distance Weighting (IDW) interpolation using a 7 × 7 pixel window.
  • Comparative Methods: Seasonal-Trend decomposition using Loess (STL), double standardization method, robust Holt–Winters method.
• Data sources:
  • Hourly LST data: 0.02° seamless hourly land surface temperature dataset over East Asia (2016–2021), derived from Himawari-8/Advanced Himawari Imager (AHI) thermal infrared observations, iTES retrieval algorithm, bias correction techniques, and Multiresolution Kalman Filter (MKF).
  • Ground observation data: Land surface temperature (T_g) from 7 long-term ground observation stations in the Heihe River Basin and Huailai Station in the Haihe River Basin (2016–2021), calculated from upwelling and downwelling longwave radiation.
  • Broadband emissivity: Global Land Surface Satellite (GLASS) Broadband Emissivity product (1 km spatial resolution, 8-day temporal resolution).

Main Results

• The proposed method detected a total of 1175 spatiotemporal anomaly points from hourly LST sequences across all sites in the Heihe River Basin between 2016 and 2021.
• For over 87% of the detected anomalies, the method demonstrated positive improvement rates across all evaluation metrics.
• The overall average improvement rates after correction were:
  • Root Mean Square Error (RMSE): 23.61% reduction.
  • Mean Absolute Error (MAE): 18.79% reduction.
  • Pearson Correlation Coefficient (R): 16.46% increase.
  • Coefficient of Determination (R^2): 61.33% increase.
• The method exhibited stable correction effectiveness across various land surface types (alpine meadow, farmland/artificial vegetation, desert shrubland, wetland) and different seasons.
• Cross-site validation at Huailai Station showed comparable performance, with overall average improvement rates of 21.75% for RMSE, 18.23% for MAE, 23.01% for R, and 56.85% for R^2.
• The proposed method significantly outperformed comparative methods (STL, double standardization, robust Holt–Winters) in terms of improvement proportions (e.g., >91% for RMSE/MAE vs. ~60-64%) and overall average improvement rates (e.g., 23.61% RMSE vs. 5.41% for STL, -3.67% for double standardization, 1.69% for robust Holt–Winters).

Contributions

• Introduced a novel two-stage anomaly detection framework that sequentially applies temporal physical pre-screening and spatial statistical verification, effectively distinguishing physically plausible weather fluctuations from data errors in high-frequency LST data.
• Prioritized the reduction of "mis-correction" risk, ensuring higher physical consistency and reliability of LST data for sensitive applications like energy balance analysis and trend studies.
• Demonstrated superior performance in anomaly detection accuracy, correction magnitude, and stability compared to existing single-stage statistical and robust filtering methods (STL, double standardization, robust Holt–Winters).
• Provided a robust and physically interpretable solution for quality control of high-temporal-frequency remote sensing LST data, enhancing its applicability in complex land surface thermal infrared scenarios.

Funding

• Major Program of the National Natural Science Foundation of China (Grant No. 42192582)
• National Key Research and Development Program of China (Grant No. 2022YFB3902000)
• Strategic Priority Research Program of the Chinese Academy of Sciences (Grant No. XDB0580000)
• Youth Innovation Promotion Association of the Chinese Academy of Sciences (Grant No. 2023246)
• Shanghai Institute of Technical Physics (SITP) Innovation Project (Grant No. CX-477)

Citation

@article{Wang2026Anomaly,
  author = {Wang, Yun and Chai, Mengyang and Zhang, Xiao and Kang, Huairong and Liu, Xuanbin and Zhao, Siwei and Cui, Cancan and Liu, Yinnian},
  title = {Anomaly Detection and Correction for High-Spatiotemporal-Resolution Land Surface Temperature Data: Integrating Spatiotemporal Physical Constraints and Consistency Verification},
  journal = {Remote Sensing},
  year = {2026},
  doi = {10.3390/rs18070972},
  url = {https://doi.org/10.3390/rs18070972}
}