Stratiform and convective rain types are associated with different cloud physical processes, vertical structures, thermodynamic influences and precipitation types. Distinguishing convective and stratiform systems is beneficial to meteorology research and weather forecasting. However, there is no clear boundary between stratiform and convective precipitation. In this study, a machine learning algorithm, K-nearest neighbor (KNN), is used to classify precipitation types. Six Doppler radar (WSR-98D/SA) data sets from Jiangsu, Guangzhou and Anhui Provinces in China were used as training and classification samples, and the 2A23 product of the Tropical Precipitation Measurement Mission (TRMM) was used to obtain the training labels and evaluate the classification performance. Classifying precipitation types using KNN requires three steps. First, features are selected from the radar data by comparing the range of each variable for different precipitation types. Second, the same unclassified samples are classified with different k values to choose the best-performing k. Finally, the unclassified samples are put into the KNN algorithm with the best k to classify precipitation types, and the classification performance is evaluated. Three types of cases, squall line, embedded convective and stratiform cases, are classified by KNN. The KNN method can accurately classify the location and area of stratiform and convective systems. For stratiform classifications, KNN has a 95% probability of detection, 8% false alarm rate, and 87% cumulative success index; for convective classifications, KNN yields a 78% probability of detection, a 13% false alarm rate, and a 69% cumulative success index. These results imply that KNN can correctly classify almost all stratiform precipitation and most convective precipitation types. This result suggests that KNN has great potential in classifying precipitation types.

This study aims to assess the performance of different dynamical downscaling methods using updated land surface information. Particular attention is given to obtaining high-resolution climate information over China by the combination of an appropriate dynamical downscaling method and updated land surface information. Two group experiments using two land surface datasets are performed, including default Weather Research and Forecasting (WRF) land surface data (OLD) and accurate dynamically accordant MODIS data (NEW). Each group consists of three types of experiments for the summer of 2014, including traditional continuous integration (CT), spectral nudging (SN), and re-initialization (Re) experiments. The Weather Research and Forecasting (WRF) model is used to dynamically downscale ERA-Interim (reanalysis of the European Centre for Medium-Range Weather Forecast, ECMWF) data with a grid spacing of 30 km over China. The simulations are evaluated via comparison with observed conventional meteorological variables, showing that the CT method, which notably overestimates 2 m temperature and underestimates 2 m relative humidity across China, performs the worst; the SN and Re runs outperform the CT method, and the Re shows the smallest RMSE (root means square error). A comparison of observed and simulated precipitation shows that the SN simulation is closest to the observed data, while the CT and Re simulations overestimate precipitation south of the Yangtze River. Compared with the OLD group, the RMSE values of temperature and relative humidity are significantly improved in CT and SN, and there is smaller improved in Re. However, obvious improvements in precipitation are not evident.

A moderate precipitation event occurring in northern Xinjiang, a region with a continental climate with little rainfall, and in leeward slope areas influenced by topography is important but rarely studied. In this study, the performance of lightning data assimilation is evaluated in the short-term forecasting of a moderate precipitation event along the western margin of the Junggar Basin and eastern Jayer Mountain. Pseudo-water vapor observations driven by lightning data are assimilated in both single and cycling analysis experiments of the Weather Research and Forecast (WRF) three-dimensional variational (3DVAR) system. Lightning data assimilation yields a larger increment in the relative humidity in the analysis field at the observed lightning locations, and the largest increment is obtained in the cycling analysis experiment. Due to the increase in water vapor content in the analysis field, more suitable thermal and dynamic conditions for moderate precipitation are obtained on the leeward slope, and the ice-phase and raindrop particle contents increase in the forecast field. Lightning data assimilation significantly improves the short-term leeward slope moderate precipitation prediction along the western margin of the Junggar Basin and provides the best forecast skill in cycling analysis experiments.

A dual-resolution, hybrid, three-dimensional ensemble-variational (3DEnVAR) data assimilation method combining static and ensemble background error covariances is used to assimilate radar data, and pseudo-water vapor observations to improve short-term severe weather forecasts with the Weather Research and Forecast (WRF) model. The higher-resolution deterministic forecast and the lower-resolution ensemble members have 3 and 9 km horizontal resolution, respectively. The water vapor pseudo-observations are derived from the combined use of total lightning data and cloud top height from the Fengyun-4A(FY-4A) geostationary satellite. First, a set of single-analysis experiments are conducted to provide a preliminary performance evaluation of the effectiveness of the hybrid method for assimilating multisource observations; second, a set of cycling analysis experiments are used to evaluate the forecast performance in convective-scale high-frequency analysis; finally, different hybrid coefficients are tested in both the single and cycling experiments. The single-analysis results show that the combined assimilation of radar data and water vapor pseudo-observations derived from the lightning data is able to generate reasonable vertical velocity, water vapor and hydrometeor adjustments, which help to trigger convection earlier in the forecast/analysis and reduce the spin-up time. The dual-resolution hybrid 3DEnVAR method is able to adjust the wind fields and hydrometeor variables with the assimilation of lightning data, which helps maintain the triggered convection longer and partially suppress spurious cells in the forecast compared with the three-dimensional variational (3DVAR) method. A cycling analysis that introduced a large number of observations with more frequent small adjustments is able to better resolve the observed convective events than a single-analysis approach. Different hybrid coefficients can affect the forecast results, either in the single deterministic or cycling analysis experiments. Overall, we found that a static coefficient of 0.4 and an ensemble coefficient of 0.6 yields the best forecast skill for this event.