Rossby waves can cross the equator and connect the Northern Hemisphere (NH) and Southern Hemisphere (SH), or be blocked in the vicinity of the equator. This work explores the windows and barriers for the cross-equatorial waves (CEWs) by the wave ray ensemble method. The eastern Pacific and Atlantic regions are identified as common windows in both boreal winter and summer, while the Africa–Indian Ocean section exists as a window only in boreal summer. The western–central Pacific is found to be a barrier section. These results are consistent with correlation analysis of reanalysis data. Moreover, the dependence on the wavenumber of CEWs is investigated, revealing that they are restricted to long waves with zonal wavenumbers less than 6 and that their wavenumber vectors exhibit a northwest–southeast (southwest–northeast) tilt when they cross the equator from the NH to SH (from the SH to NH). This long-wave dominance of CEWs results from the spectral-selective filtering mechanism, which suggests that long waves have narrower equatorial barriers than short waves. Finally, the main wave duct associated with each window is obtained by the global passing CEW density distribution. The results indicate that the main CEW ducts roughly follow a great circle–like pathway, except for the Africa–Indian Ocean window in boreal summer, which may be modulated by the cross-equatorial monsoonal flow.

Stagnation weather affects atmospheric diffusion ability, and hence causes the occurrence of haze events, which have been happening frequently in northern China (NC). This work puts forward an air stagnation index (ASI_TS) to characterize the stagnation weather in NC, in which the processes of ventilation, vertical diffusion, and wet deposition potency are concerned. ASI_TS can be applied to analyze air stagnation conditions with daily to monthly time scale. It is shown that the ASI_TS and particulate matter smaller than 2.5 μm in diameter (PM_2.5) concentrations own similar lognormal probability distribution functions on both daily and monthly time scales. And the correlation analyses between the ASI_TS and PM_2.5 concentrations indicate that the ASI_TS can reflect the monthly and daily variations in PM_2.5 concentrations in NC. In addition, ASI_TS could be used as a leading predictor of haze events since correlation coefficients of ASI_TS leading PM_2.5 concentrations by 1 day were significant and were larger than simultaneous correlation coefficients in almost all areas in NC. The robust relationship between ASI_TS and PM_2.5 concentrations exists possibly because the index can reflect the activities of synoptic systems. ASI_TS could be a useful statistical indicator for variations in PM_2.5 concentrations and haze events, and a good tool in analyzing the relationship between climate change and long-term variations in haze in NC.

Atmospheric chemistry transport models have been extensively applied in aerosol forecasts over recent decades, whereas they are facing challenges from uncertainties in emission rates, meteorological data, and over-simplified chemical parameterizations. Here, we developed a spatial-temporal deep learning framework, named PPN (Pollution-Predicting Net for PM_2.5), to accurately and efficiently predict regional PM_2.5 concentrations. It has an encoder-decoder architecture and combines the preceding PM_2.5 observations and numerical weather prediction. Besides, the model proposes a weighted loss function to promote the forecasting performance in extreme events. We applied the proposed model to forecast 3-day PM_2.5 concentrations over the Beijing-Tianjin-Hebei region in China on a three-hour-by-three-hour basis. Overall, the model showed good performance with R² and RMSE values of 0.7 and 17.7 μg m⁻³, respectively. It could capture the high PM_2.5 concentration in the south and relatively low concentration in the north and exhibit better performance within the next 24 h. The use of the weighted loss function decreased the level of “high values underestimation, low values overestimation”, while incorporating the preceding PM_2.5 observations into the encoder phase improved the predictive accuracy within 24 h. We also compared the model result with that from a state-of-the-art numerical model (WRF-Chem with pollutant data assimilation). The temporal R² and RMSE from the WRF-Chem were 0.30−0.77 and 19−45 μg m⁻³ while those from the PPN model were 0.42−0.84 and 15−42 μg m⁻³. The proposed model shows powerful capacity in aerosol forecasts and provides an efficient and accurate tool for early warning and management of regional pollution events.

Several blending methods have been developed in dynamic downscaling and rapid cycled data assimilation. Blending the large‐scale part of the global model (GM) analysis or forecast has led to improvement in regional model (RM) simulations. However, in previous studies the blended waveband of the GM has generally been determined using a fixed, arbitrarily chosen cutoff wave number. Here we introduce a new dynamic blending (DB) scheme with a dynamic cutoff wave number computed according to the spectral characteristics of GM forecast quality and the spectral distribution of errors in the RM. The DB scheme is described and applied to eight‐day summertime and seven‐day wintertime cycled Weather Research and Forecasting Model forecasts over a regional domain in the continental United States. The scheme can determine a cutoff wave number with significant temporal variations. The temporal variation results from the error growth property of the RM and has a clear diurnal oscillation, suggesting that fewer (more) GM waves should be introduced into the RM at noon (night). The cutoff wave number difference between the two periods indicates seasonal variation of the cutoff wave number with larger day‐to‐day change in winter. Comparison among no blending experiment, two fixed wave number blending experiments, and two DB experiments with and without vertically varying cutoff wave number suggests that the DB scheme with vertically averaged but temporally varying cutoff wave number results in less model bias and less disturbance to the RM dynamic balance. By reducing the forecast background error, the DB scheme can potentially provide improved first guess for a rapid‐update‐cycle weather forecast system.