Convective initiation (CI) nowcasting is crucial for reducing loss of human life and property caused by severe convective weather. A novel deep learning method based on the U-Net model (named as CIUnet) was developed for forecasting CI during the warm season with eight interest fields of Himawari-8 Advanced Himawari Imager (AHI) and terrain height. The results showed that the CIUnet model produced probability forecasts of CI occurrence location and time with probability of detection (POD) at 93.3% +/- 0.3% and false alarm ratio (FAR) at 18.3% +/- 0.4% at a lead time of 30 min. Sensitivity and permutation importance experiments on the input fields of the CIUnet model revealed that the dif-ferences in brightness temperature for spectral channels were more critical for CI nowcasts than the original infrared chan-nel brightness temperatures. The brightness temperature difference between band 10 (7.3 mm) and band 13 (10.4 mm), which represents the cloud-top height relative to the lower troposphere, is identified as the most important input fields for CI nowcasting. The tri-spectral brightness temperature difference (TTD), which represents cloud-top glaciation, is ranked the second and it significantly reduced the FAR of the CI forecast. Using terrain heights as an extra input feature improved the POD, but slightly overestimated CI over complex terrain. In addition, a layer-wise relevance propagation (LRP) analyses was performed, and confirmed that the CIUnet model can effectively identify the crucial regions and features of the input fields for accurate CI prediction. Therefore, both permutation importance experiments and LPR analyses are useful for improving the CIUnet model and advancing the understanding of CI mechanisms.

Planetary boundary layer (PBL) is formed by the interaction of the atmosphere and the surface and surface conditions have an important influence on the PBL state. This study developed a method to retrieve temperatures in the lower troposphere using temperature reanalysis at 2 m (T2) with a U-net based deep learning model (briefly, lower troposphere retrieval with U-net, LTRUnet). High-resolution four-dimensional data assimilation (FDDA) reanalysis data over Southern China was selected to demonstrate the ability of the deep-learning network. The normalization method and the hyper-parameters of LTRUnet were determined through experiments. The results show that LTRUnet can project the temperatures in the layers below about 1.3 km with good accuracy, with a Mean Absolute Error (MAE) <0.5 K on the test set compared to the reanalysis. The performance of LTRUnet deteriorates with height and has a strong diurnal variation. By utilizing auxiliary physical information of the terrain height and the first-guess temperature (Tb) as additional channels of the LTRUnet input, the error (MAE) of the projected temperature was further reduced by 19.5%. Using the first guess background fields also allows LTRUnet to achieve the same prediction performance with much less training data. Sensitivity tests of the input variables indicate that the topographic information helps the model to better predict the topography-induced temperature distribution, and Tb plays a more significant role in higher levels than T2. Finally, by training LTRUnet independently for different phases of PBL with a physics-guided data grouping strategy, the model results are further improved.

A summer convective precipitation case, occurring in eastern China on 16-17 July 2020, is selected to investi-gate the impact of joint assimilation of ground-based profiling platforms and weather radars on forecasting convective storms using observational system simulation experiments (OSSEs). The simulated profiling platforms include the Doppler wind lidar (DWL), a wind profiler (WP), and a microwave radiometer (MWR). Results show that joint assimilation of WP and radar data produces a better analysis of convective dynamical structure than joint assimilation of DWL and radar data, since WP detects deeper layer winds. Joint assimilation of MWR and radar data enables rapid adjustment of temperature and humidity and thus, avoids the potential errors introduced by the latent heat term of the radar diabatic initialization in the early stage. Profiling observations in a horizontal spacing of 80 km provide fewer benefits for convective forecasting, while reducing the spacing to 40 km can dramatically improve model analysis and forecasts. Joint assimilation of multiple profiling observations in a 20-km horizontal spacing with radar data exhibits a beneficial synergistic effect and mitigates "the ramp-down issue" during the forecast stage. Assimilating profiling observations with an update interval less than 30 min does not have as pronounced an effect on convective forecasts as horizontal spacing. Furthermore, assimilating profiling ob-servations at a 20-km horizontal spacing can obtain accurate mesoscale background environment and forecast storms with an ability comparable to radar data assimilation. This work emphasizes the need to consider implementing a joint mesoscale detection system that incorporates weather radars and profiling observations for leveraging convective storm forecasting.

Retrieving cloud vertical structures with satellite remote-sensing measurements is highly desirable and technically challenging. In this paper, the conditional adversarial neural network (CGAN) for retrieving the equivalent cloud radar reflectivity at 94 GHz of the Cloud Profile Radar (CPR) onboard CloudSat is extended and evaluated comprehensively for different cloud types and geographical regions. The CGAN-based retrieval model was extended with additional data samples and improved with a new normalization adjustment. The model was trained with the labeled datasets of the moderate-resolution imaging spectroradiometer (MODIS) cloud top pressure, cloud water path, cloud optical thickness, and effective particle radius data, and the CloudSat/CPR reflectivity from 2010 to 2017 over the global oceans. The test dataset, containing 24,427 cloud samples, was statistically analyzed to assess the performance of the model for eight cloud types and three latitude zones with multiple verification metrics. The results show that the CGAN model possesses good reliability for retrieving clouds with reflectivity > −25 dBZ. The model performed the best for deep convective systems, followed by nimbostratus, altostratus, and cumulus, but presented a very limited ability for stratus, cirrus, and altocumulus. The model performs better in the low and middle latitudes than in the high latitudes. This work demonstrated that the CGAN model can be used to retrieve vertical structures of deep convective clouds and nimbostratus with great confidence in the mid- and lower latitude region, laying the ground for retrieving reliable 3D cloud structures of the deep convective systems including convective storms and hurricanes from MODIS cloud products and used for predicting these storms.

Forecasts of numerical weather prediction models unavoidably contain errors, and it is a common practice to post-process the model output and correct the error for the proper use of the forecasts. This study develops a grid-to-multipoint (G2N) model output error correction scheme which extracts model spatial features and corrects multistation forecasts simultaneously. The model was tested for an operational high-resolution model system, the precision rapid update forecasting system (PRUFS) model, running for East China at 3 km grid intervals. The variables studied include 2 m temperature, 2 m relative humidity, and 10 m wind speed at 311 standard ground-based weather stations. The dataset for training G2N is a year of historical PRUFS model outputs and the surface observations of the same period and the assessment of the G2N performance are based on the output of two months of real-time G2N. The verification of the real-time results shows that G2N reduced RMSEs of the 2 m temperature, 2 m relative humidity, and 10 m wind speed forecast errors of the PRUFS model by 19%, 24%, and 42%, respectively. Sensitivity analysis reveals that increasing the number of the target stations for simultaneous correction helps to improve the model performance and reduces the computational cost as well indicating that enhancing the loss function with spatial regional meteorological structure is helpful. On the other hand, adequately selecting the size of influencing grid areas of the model input is also important for G2N to incorporate enough spatial features of model forecasts but not to include the information from the grids far from the correcting areas. G2N is a highly efficient and effective tool that can be readily implemented for real-time regional NWP models.

Satellite visible radiance data that contain rich cloud and precipitation information are increasingly assimilated to improve the forecasts of numerical weather prediction models. This study evaluates the Data Assimilation Research Testbed (DART, Manhattan release v9.8.0), coupled with the Weather Research and Forecasting (WRF) model (ARW v4.1.1) and the Radiative Transfer for TOVS (RTTOV, v12.3) package, for assimilating the simulated visible imagery of the FY-4A geostationary satellite located over Asia in an Observing System Simulation Experiment (OSSE) framework. The OSSE was performed for the tropical storm Higos that occurred in 2020 and contains multi-layer mixed-phase cloud and precipitation processes. The advantages and limitations of DART for assimilating FY-4A visible imagery were evaluated. Both single-observation experiments and cycled data assimilation (DA) experiments were performed to study the impact of different filter algorithms available in DART, variables being cycled, observation outlier thresholds, observation errors, and observation thinning. The results show that assimilating visible radiance data significantly improves the analysis of the cloud water path (CWP) and cloud coverage (CFC) from first-guess forecasts. The rank histogram filter (RHF) allows WRF to more accurately simulate CWP and CFC compared with the ensemble adjustment Kalman filter (EAKF) although it took roughly twice as long as the latter. By cycling both cloud and non-cloud variables, specifying large outlier threshold values, or setting smaller observation errors without thinning of observations, WRF achieved a better simulation of CWP and CFC. With model integration, DA of the visible radiance data also generated a slightly positive impact on non-cloud variables as they were adjusted through the model dynamics and physics related to cloud processes. In addition, the DA improved the representation of precipitation. However, the impact on the rain rate is limited by the inabilities of the DA to improve cloud vertical structures and cloud phases. A negative impact of the DA on cloud variables was found due to the nature of the non-linear forward operator and the non-Gaussian distribution of the prior. Future works should explore faster and more accurate forward operators suitable for assimilating FY-4A visible imagery, techniques to reduce the non-linear and non-Gaussian errors, and methods to correct the location errors which correspond to the clouds underestimated by the first guess.

The terrain effect on atmospheric environment is poorly understood in particular for the polluted region with underlying complex topography. Therefore, this study targeted the Sichuan Basin (SCB), a deep basin with severe PM2.5 pollution enclosed by the eastern Tibetan Plateau (TP), Yunnan-Guizhou Plateaus (YGP) and mountains over Southwest China, and we investigated the terrain effect on seasonal PM2.5 distribution and the meteorological mechanism based on the WRF-Chem simulation with stuffing the basin topography. It is characterized that the three-dimensional distribution of topography-induced PM2.5 concentrations over the SCB with the seasonal shift of regional PM2.5 averages from approximately 30 mu g mxe213; 3 in summer to 90 mu g mxe213; 3 in winter at surface layer and from summertime 10 mu g mxe213; 3 to wintertime 30 mu g mxe213; 3 in the lower free troposphere. Such basin-forced PM2.5 changes presented the vertically monotonical declines concentrated within the lower troposphere below 3.6 km in spring, 2.3 km in summer, 2.6 km in autumn and 4.8 km in winter. Impacts of deep basin aggravated PM2.5 accumulation within the SCB and transport toward the surrounding plateaus contributing approximately 50-90% to PM2.5 levels over the regions of eastern TP and northern YGP. In the SCB, atmospheric thermal structure in the lower troposphere could build a vertical convergence layer between the boundary layer and free troposphere, acting as a lid inhibiting air diffusion, which was regulated by the terrain effects on interactions of westerlies and Asian monsoons, especially the wintertime strong warm lid deteriorating air pollution in the SCB. Furthermore, warm and humid air conditions within the basin prompted sulfur oxidation ratio by +0.02 and nitrogen oxidation ratio by +0.22 effectively producing the secondary PM2.5 in atmospheric environment.

Sea breezes are one of the most important weather processes affecting the environmental and climatic features over coastal areas, and the sea breeze from the Pearl River Estuary (PRE) has significant effects on the Pearl River Delta (PRD) region. We simulated a typical sea breeze process that occurred on 27 December 2020 in the PRD region using the Weather Research and Forecasting (WRF) model to quantify the effects of topography and city clusters on the development of the sea breeze circulation. The results show that: (1) the topography on the west coast of the PRD tends to block the intrusion of the sea breeze and detour it along the eastern part of the terrain in the southeast of Jiangmen. The depth of sea breeze along the position of the detour is increased by 120 m, the penetration distance is increased by 40 km, the maximum intensity of sea breeze decreases by ~0.4 m/s, and the time of maximum speed delays for 4 h. However, on the east coast, the topography promotes the sea breeze, resulting in an occurrence about 4 h earlier due to the heating effects. The depth and the speed of the sea breeze are increased by 466 m and 1.2 m/s, respectively. (2) Under the influence of Urban Heat Island Circulation (UHIC), the sea breezes reach cities near the coast an hour earlier and are later inhibited from propagating further inland. Moreover, a wind convergence zone with a speed of 3–5 m/s and a width of about 25 km is formed along the boundary of suburbs and cities in the PRD region. As a result, two important convergence areas: Foshan–Guangzhou, and Dongguan–Shenzhen are formed. (3) Overall, the topography has a more remarkable impact on the mesoscale wind field especially in the mountain and bay areas, resulting in an average speed disturbance of 2.8 m/s. The urban heat island effect is relatively small and on average it causes only ± 0.9–1.8 m/s wind speed perturbations in the periphery of two convergence areas and over PRE.

Abstract. Satellite visible (VIS) radiance data contain rich cloud information that are increasingly assimilated for improving cloud and precipitation forecasting of numerical weather prediction models. Recently, the Data Assimilation Research Testbed (DART), a widely used data assimilation resource that supports the Weather Research and Forecasting (WRF) model, was facilitated with an interface for the Radiative Transfer for TOVS (RTTOV), which supports radiance assimilation from visible (VIS) to microwave wavelength channels. This study evaluates the WRF (ARW v4.1.1)/DART (Manhattan release v9.8.0)-RTTOV (v12.3) system for assimilating the radiance data of channel 2 (0.55~0.75 μm) of the Advanced Geostationary Radiation Imager (AGRI) onboard FY-4. Observing System Simulation Experiments (OSSEs) were performed for a cyclone case. The results indicate that assimilating VIS radiance data improves cloud forecast skills in general. Best results were achieved for the data assimilation (DA) experiment with dense observations and high updating frequency. The best results could capture the “eye” structure of the cyclone system and significantly improves cloud water path and cloud coverage simulations. Nevertheless, three main problems were revealed. The first is its inability to improve cloud vertical distribution such as layered structures and cloud phases; The second is the its inability to influence atmosphere thermodynamic state variables positively; The third is a waste of up to 50 % observations during the filtering processes.

The model_codes.tar.gz file includes the source code of WRF-ARW (v4.1.1), WPS (v4.1), RTTOV (v12.3), and DART (Manhattan release v9.8.0). The input_file.tar.gz file includes the input namelist files for the WRF/WPS/DART models (tool) for the nature run, control run, and six data assimilation experiments. In addition, the input observation sequence file for the DART tool at each analysis time are provided for the six data assimilation experiments. More details are provided in the readme.txt file after unzipping input_file.tar.gz. The PROCESSED.tar.gz file includes the processed output files for the nature run, the control run, and six data assimilation experiments. More details are in the readme.txt file after unzipping PROCESSED.tar.gz. The Figure_script.tar.gz file includes the visualization scripts of figure 1~figure 12 in the manuscript. Each figure corresponds to a subdirectory which includes a readme.txt file for more detailed description.

In this study, collocated filter samples of particulate matter with aerodynamic diameter less than 2.5 μm (PM2.5) from northern Nanjing were extracted using water and methanol, followed by analysis of light absorption. A backup quartz filter was used to correct sampling artifacts caused by adsorption of gaseous organics. The collocated precision of light‐absorbing properties of water‐soluble organic carbon (WSOC) and methanol‐extractable organic carbon (MEOC) were parameterized using correlation coefficient (r), coefficient of divergence (COD), and average relative percent difference (ARPD, %). In general, the light absorption of WSOC and MEOC showed good agreement (r > 0.80, COD < 0.20) between collocated samples. Performing artifact correction is necessary and will increase the heterogeneity between collocated measurements. The duplicate‐derived ARPD values of MEOC absorption were more than 60% higher than those of WSOC absorption. Then, it would be inappropriate to assume a uniform uncertainty fraction (e.g., ∼10%) for WSOC and MEOC absorption in future studies on their climate effects and source apportionment. To apportion artifact‐corrected absorption of aerosol extracts to specific emission sources or formation pathways, positive matrix factorization was performed by using concentration data of selected bulk species and organic molecular markers. Among the nine identified factor/sources, the biomass burning factor had the highest average contributions to the absorption of both WSOC (31.6%) and MEOC (48.0%), followed by dust resuspension and coal combustion factors. Unlike combustion‐related primary emissions, the factors containing influences from atmospheric processing (e.g., secondary nitrate) contributed more fractions of WSOC absorption than MEOC absorption.

The Geostationary Lightning Mapper (GLM) on the Geostationary Operational Environmental Satellite 16 (GOES-16) detects total lightning continuously, with a high spatial resolution and detection efficiency. Coincident data from the GLM and the Advanced Baseline Imager (ABI) are used to explore the correlation between the cloud top properties and flash activity across the continental United States (CONUS) sector from May to September 2020. A large number of collocated infrared (IR) brightness temperature (TBB), cloud top height (CTH) and lightning data provides robust statistics. Overall, the likelihood of lightning occurrence and high flash density is higher if the TBB is colder than 225 K. The higher CTH is observed to be correlated with a larger flash rate, a smaller flash size, stronger updraft, and larger optical energy. Furthermore, the cloud top updraft velocity (w) is estimated based on the decreasing rate of TBB, but it is smaller than the updraft velocity of the convective core. As a result, the relationship between CTH and lightning flash rate is investigated independently of w over the continental, oceanic and coastal regimes in the tropics and mid-latitudes. When the CTH is higher than 12 km, the flash rates of oceanic lightning are 38% smaller than those of both coastal and continental lightning. In addition, it should be noted that more studies are necessary to examine why the oceanic lightning with low clouds (CTH < 8 km) has higher flash rates than lightning over land and coast. Finally, the exponents of derived power relationship between CTH and lightning flash rate are smaller than four, which is underestimated due to the GLM detection efficiency and the difference between IR CTH and 20 dBZ CTH. The results from combining the ABI and GLM products suggest that merging multiple satellite datasets could benefit both lightning activity and parameterization studies, although the parallax corrections should be considered.

The spatiotemporal statistical characteristics of warm-season deep convective systems, particularly deep convective systems initiation (DCSI), over China and its vicinity are investigated using Himawari-8 geostationary satellite measurements collected during April-September from 2016 to 2020. Based on a satellite brightness temperature multiple-threshold convection identification and tracking method, a total of 47593 deep convective systems with lifetimes of at least 3 h were identified in the region. There are three outstanding local maxima in the region, located in the southwestern, central and eastern Tibetan Plateau and Yunnan-Guizhou Plateau, followed by a region of high convective activities in South China. Most convective systems are developed over the Tibetan Plateau, predominantly eastward-moving, while those developed in Yunnan-Guizhou Plateau and South China mostly move westward and southwestward. The DSCI occurrences become extremely active after the onset of the summer monsoon and tend to reach a maximum in July and August, with a diurnal peak at 11–13 LST in response to the enhanced solar heating and monsoon flows. Several DCSI hotspots are identified in the regions of inland mountains, tropical islands and coastal mountains during daytime, but in basins, plains and coastal areas during nighttime. DCSI over land and oceans exhibits significantly different sub-seasonal and diurnal variations. Oceanic DCSI has an ambiguous diurnal variation, although its sub-seasonal variation is similar to that over land. It is demonstrated that the high spatiotemporal resolution satellite dataset provides rich information for understanding the convective systems over China and vicinity, particularly the complex terrain and oceans where radar observations are sparse or none, which will help to improve the convective systems and initiation nowcasting.

In this study, collocated samples of particulate matter with aerodynamic diameter less than 2.5 mu m (PM2.5) were collected every sixth day on quartz filters (Q(f)) at a suburban site in northern Nanjing, China for one year. A backup quartz filter (Q(b)) was installed behind Q(f) to estimate positive artifacts. The analysis of gravimetric mass, water-soluble inorganic ions (WSIIs), total organic (OC) and elemental carbon (EC), water-soluble OC (WSOC), total nitrogen (WSTN) and organic nitrogen (WSON) were performed on both Q(f) and Q(b). Due to the high mass loadings on Q(f) (17.9 +/- 7.82 mg filter(-1), 4.71-38.5 mg filter(-1)), the collocated precision of gravimetric mass and dominant species (NH4+, NO3-, SO42-, OC, and EC) is better than that from previous work. Except K+, EC, and WSON, all other target components were detected on Q(b) with average Q(b)/Q(f) ratios ranging from 3.02 +/- 3.48-21.7 +/- 22.4%. The final concentrations and uncertainties of PM2.5 components were determined based on duplicate Q(f)-Q(b) data. The results suggest that using an error fraction of 10% will underestimate the uncertainty of less concentrated species (e.g., Ca2+ and Mg2+) in PM2.5. Due to the evaporation loss of semi-volatile materials from the Q(f), the Q(f)-Q(b) calculation would lead to an estimate of the lower limit for particulate NH4+, NO3-, and OC. Synchronized hourly data of PM2.5 mass and components were obtained at downtown Nanjing. The comparisons of gravimetric versus reconstructed PM2.5 and filter-based versus continuous measurements of PM2.5 components indicate that a substantial fraction of the unexplained gravimetric PM2.5 can be attributed to aerosol water.

Radar data are essential to convection nowcasting and nudging-based radar data assimilation through diabatic initialization is one of the most effective approaches for forecasting convective systems with numerical weather prediction (NWP) models, used at several advanced global weather centers. It is desired to assess the uncertainty and physical consistency of this assimilation process. This paper investigated impacts of relaxation coefficient, radar data update intervals and continuous assimilation time duration and addressed the key issues and possible solutions of the radar data assimilation based on the WRF hydrometeor and latent heat nudging (HLHN) developed at the National Center for Atmospheric Research (NCAR). It is revealed that excessively large relaxation coefficient forced the model to observations with a tendency greater than the physical terms of the convection, causing the dynamic imbalances and serious convection “ramp-down” right after the free forecast starts. Assimilating high update frequency radar data can make the tendency terms moderate and sustained thereby maintaining the assimilation effect and reducing fortuitous convection. HLHN requires a minimum continuous assimilation duration to contain the initial forced disturbance of the model. For a summer Meiyu precipitation case studied, the minimum duration is ~1 h. Appropriate selection of the HLHN parameters is able to effectively improve the temperature, humidity, and dynamic fields of the model. In addition, several issues still remain to be solved to further enhance HLHN.

随着高分辨率数值天气模式以及新一代地球观测系统的发展,气象领域的数据量在迅速增加,为天气和气候的理论研究和业务应用提供了丰富的信息,同时也对传统的数据处理方法及天气分析和预报技术带来了新的挑战.深度学习具有从大量的高维时空分布气象数据中提取复杂时空特征的能力,且具备计算效率高、可迁移性强、协同性和灵活性优的特点.目前深度学习已经在对流短时临近预报、极端事件检测和改进数值天气模式及其预报误差订正等方面得到了较为广泛的研究.首先概述目前气象领域所应用的深度学习方法和模型,然后聚焦介绍和讨论数据驱动的深度学习在理论驱动的数值天气预报模式方面的应用.深度学习在数值天气预报模式的资料同化、次网格物理过程参数化、数值天气模式后处理等方面展现出很好的应用前景,但仍需要进一步改善深度学习模型的可解释性和不确定性量化问题.构建数据驱动的深度学习和理论驱动的数值天气模式混合模型,发挥深度学习和数值天气模式的协同作用,将是进一步改善数值天气预报能力的新途径.

Based on the Weather Research Forecasting (WRF) model that features charging and discharging parameterization, relationships between tornado, hail and lightning were investigated for a tornado-producing (EF4 intensity) supercell thunderstorm over Yancheng City in Jiangsu Province, China, on 23 June 2016. Based on a sounding at 0800, there was a low lifting condensation level, substantial convective available potential energy (CAPE), and strong vertical wind shear near Yancheng City, which promote supercell development. At 1400, observations revealed that hail production and a dramatic increase of positive cloud-to-ground flash rates occurred simultaneously, maximizing five minutes later. The tornado occurred 30 min after the hail production. The time of minimum positive cloud-to-ground flash rates was 15 min later. The simulation indicated that the tornadic supercell moved eastward and that positive cloud-to-ground flash rates increased dramatically at 1400, the same as observed, but their maximum was 5 min later than observed. The simulated updraft volume peaked at 1425 and the simulated downdraft volume maximized 5 min later, when the mesocyclone formed. Simulated reflectivities showed no hook echo and horizontal winds for different height at mid-low levels had a different cyclonic shear at 1430, favorable to mesocyclone formation. Based on the simulated results, the region of positively charged graupel ascended resulting from the region of high liquid water content was lifted by the strong updraft, forming a mid-level strong positive charge region. A lower negative charge region formed by the inductive charging mechanism of collisions between graupel and droplets at the bottom of the cloud, conducive to positive cloud-to-ground flashes.