Arctic near-surface air temperature_60° N_90° N_3h_5km

To address the limitations of remote-sensing temperature observations in the Arctic, including insufficient accuracy, poor temporal continuity, and incomplete spatial coverage, the dataset was generated using a physics-guided spatiotemporal reconstruction and correction framework (PGSRC Framework) that takes advantage of the high-frequency revisit capability of polar-orbiting satellites at high latitudes.

Comprehensive validation based on ground observations and independent sounding data showed that the final product achieved an R² of 0.83 and an RMSE of 2.34 K, significantly outperforming traditional interpolation results (R² = 0.57, RMSE = 4.76 K). In addition, the model exhibited stable performance in time-based, station-based, and sample-based cross-validation, as well as in monthly validation (R² > 0.87), and maintained strong generalization capability under conditions of heavy cloud cover and severe data gaps.