A Novel Multimodel Pattern Projection Superensemble Method: Experiments on Subseasonal Forecasts of Surface Air Temperature

Abstract A novel hybrid multimodel ensemble forecast method, namely, the pattern projection superensemble (PPSUP), is proposed to enhance forecast accuracy by integrating spatial bias correction and multimodel aggregation in the study. Utilizing daily maximum (Tmax) and minimum (Tmin) temperature forecast datasets from the Subseasonal to Seasonal (S2S) project, we conduct experiments for subseasonal temperature forecasts over East Asia. The PPSUP reduces mean absolute errors by ∼1°C across all lead times and achieves hit rate improvements of ∼18% and ∼17% compared to the control forecasts for Tmax and Tmin, respectively, with statistical significance confirmed at the 95% confidence level. Substantial enhancements are demonstrated in challenging regions such as the Tibetan Plateau and eastern and northern East Asia, with effective lead times being significantly extended. Moreover, after PPSUP calibration, forecast biases become centered around zero, with positive and negative biases balanced in frequency and magnitude. This effect is particularly notable for originally larger biases, indicating its potential in handling extreme predictions. Besides, the consistently lower skills for Tmax than Tmin are basically attributed to the higher inherent variability in Tmax, though these postprocessing procedures show more significant improvements in Tmax forecasts. Significance Statement Accurately predicting temperature patterns weeks in advance is crucial for weather-sensitive sectors, such as agriculture, energy, and disaster preparedness. This study introduces the pattern projection superensemble (PPSUP), a novel hybrid forecasting approach that combines multiple models to improve subseasonal predictions. By reducing errors by about 1°C and significantly enhancing prediction accuracy, particularly in challenging regions like the Tibetan Plateau, PPSUP offers a promising tool for better anticipating weather extremes. Notably, the study also reveals the inherent uncertainties in forecasting daily maximum temperatures (Tmax), which are typically more challenging than minimum temperatures (Tmin). These advancements could lead to more informed decision-making and better resource management in response to weather-related risks, highlighting its potential to shape future subseasonal forecasting efforts.