During June—July 2023, North China (NC) experienced a record-breaking Extreme High Temperature (EHT) event with temperatures exceeding 40℃ at many meteorological stations. This event consisted of three relatively distinct processes, occurring during 14—18 June (P1), 21—25 June (P2), and 30 June—3 July (P3), respectively. Using hindcast data from the European Centre for Medium-Range Weather Forecasts (ECMWF) and the China Meteorological Administration (CMA) within the Subseasonal-to-Seasonal (S2S) Prediction Project and diagnostic analysis of the local temperature budget, surface energy budget, and large-scale atmospheric circulation, this study evaluates the performance of ECMWF and CMA dynamic models in forecasting the spatial distribution and intensity of the three EHT processes, and further reveals the sources of forecast errors and predictability. The results indicate that: (1) ECMWF and CMA models can predict the spatial distribution of Surface Air Temperature (SAT) anomalies over NC for P1, P2 and P3 with lead time of 10—11, 12—14, and 3—6 d, respectively. However, both models underestimate the amplitude of SAT anomalies, especially during P3; (2) when the forecast lead time exceeds 15 d for P1 and P2, and 5 d for P3, both models fail to capture features of the mid- to high-latitude Rossby wave train over Eurasia, resulting in prediction biases in both the location and intensity of localized high-pressure anomalies over NC; (3) prediction biases of the localized high-pressure anomalies over NC further induce biases in the forecast of local physical processes and SAT anomalies. During P1 and P3, the models' underestimation of both shortwave and longwave radiation leads to local negative temperature tendency biases, with diabatic heating being the primary contributor. In contrast, during P2, the underestimation of adiabatic heating is primarily responsible for the cold bias in predicted SAT. This study highlights the mid- to high-latitude subseasonal Rossby wave train teleconnection as a critical source of predictability for subseasonal EHT in NC, suggesting that better representation of this wave pattern is a key to improve EHT prediction skills over NC.

How to improve the extended-range predictive skill is a hotspot and frontier research issue, which is crucial for bridging the gap in seamless prediction system. Based on the observations and reanalysis data during December 2005—August 2014, the Singular Value Decomposition analysis is used to reveal the highly coupled modes between the low-frequency precipitation over southern China and intraseasonal tropical convection/mid-latitude wave trains in boreal winter and summer, respectively. The BCC-CPS-S2Sv2 (hereafter referred to as BCC S2S) model provided by China Meteorological Administration is used to construct a set of dynamical-statistical models for subseasonal prediction of low-frequency precipitation anomalies over southern China using the statistical downscaling method. The BCC S2S model participates the Subseasonal-to-Seasonal Project and exhibits reasonable skills on the forecast of rainfall anomalies over most of southern China at 10—15 d forecast lead times during the independent prediction period of December 2014—August 2019. However, the dynamical-statistical model outperforms the BCC S2S model on precipitation forecast in terms of temporal variability over coastal region of South China (north of the Yangtze river) during winter (summer) and the spatial distribution and extreme events beyond 15—20 d forecast lead. The idea and method proposed by this study can be widely applied to extended-range prediction of other regional meteorological elements and extreme events.