Spatial consistency quality control of daily surface air temperature observations in Viet Nam using inverse distance weighting and random forest

Tóm tắt

This study developed and evaluated a spatial consistency quality control (QC) method for daily surface air temperature observations from 186 SYNOP stations across Viet Nam during 1961-2023. The variables examined included daily mean (Tavg), minimum (Tmin), and maximum (Tmax) temperatures. The proposed framework combines inverse distance weighting (IDW) with a machine learning (ML) model. IDW serves as the baseline reference, whereas ML is employed to better capture nonlinear relationships associated with spatial location, elevation, seasonality, and climatic region. Final QC flags were derived from the combined outputs of both models, allowing alerts to be stratified into method-specific detections, concurrent detections by IDW and ML, and high-suspicion cases. The ML model consistently outperformed IDW for all three variables. Its mean absolute error was 0.47°C for Tavg, 0.59°C for Tmin, and 0.73°C for Tmax, compared with 0.58°C, 0.74°C, and 1.00°C, respectively, for IDW. Under the IDW-ML framework, 98.87-99.10% of the evaluated observations were classified as good quality, while alert rates ranged from 0.90% to 1.13% and high-suspicion cases accounted for only 0.05-0.13%. Approximately one-third of all alerts were detected concurrently by both methods, thereby narrowing the subset of stations requiring priority manual review. These findings demonstrate that the combined IDW-ML approach provides an effective framework for spatial QC of daily temperature observations, particularly in networks characterized by complex terrain, uneven station distribution, and island stations, as in Viet Nam.