Based on multi-source observation data and data fusion and assimilation technology, National Meteorological Information Center has developed and released real-time analysis products that has passed the industry admittance review. In order to ensure the objectivity and authenticity of the evaluation results of the real-time analysis products, the representativeness of the data source—ground station data participating in the inspection and evaluation of the products was studied. 10 precipitation impact indicators such as longitude and latitude, slope, slope direction and data availability of ground meteorological stations from May to August 2020 were selected. On the basis of indicator screening, the weight of each impact indicator to form a comprehensive impact indicator of each station was determined by using correlation analysis, principal component analysis, and grading inspection of the comprehensive impact indicator were conducted. The results show that five of the 10 impact indicators are retained after screening, and their weights from large to small are data availability, equipment stability, slope variability, surface roughness and altitude. The comprehensive impact value of stations in Sichuan Basin is mostly above 0.9, and the stations with low comprehensive impact values are mainly distributed in Ganzi, Aba and Liangshan prefectures of Sichuan Province, which is closely related to the complex terrain and poor representativeness of station in these areas. Through the grading evaluation of the comprehensive impact index, it is reasonable to take the data of stations with the comprehensive impact index value above 0.8 as the "true value" data source for the evaluation of precipitation real-time analysis products.
In order to learn more about the performance of convective-scale ensemble forecast system for precipitation prediction in the Sichuan-Chongqing region, the control forecast (CNTL), the ensemble mean (MEAN) and the probability-matched ensemble mean (PM) of convective-scale ensemble prediction system are comprehensively analyzed based on daily precipitation data collected at 7 213 stations in the Sichuan-Chongqing region in warm season (from May to September) from 2020 to 2021, and differences between rainfall forecasts starting at 08:00 and 20:00 are compared. Results are as follows: (1) The forecast performance of PM and MEAN is better than that of CNTL. MEAN is skillful at forecasting moderate rain and heavy rain, and PM has obvious advantages for large rainfall. (2) Positive forecast deviations of light rainfall frequency are obvious in the whole research region, while for moderate rain and above, positive deviations are concentrated in high-altitude mountains such as the Daba Mountain, the Huaying Mountain and the Wuling Mountain, and negative deviations are mainly located in the Sichuan Basin and hilly areas. Positive (negative) deviations of light rain and moderate rain (heavy rain and rainstorm) predicted by MEAN are more obvious than those predicted by CNTL and PM. (3) The critical success index (CSI) and probability of detection (POD) scores with lead time of 36 h for the forecasts starting at 08:00 are higher than those with lead time of 48 h for the forecasts starting at 20:00, but the overestimation of rainfall frequency starting at 08:00 is more obvious in high-altitude mountains. (4) Compared with CNTL, PM and MEAN are better for the rainfall area of the heavy rain process from September 4 to 7, 2021 in the Sichuan Basin, which is related to the fact that ensemble forecast can better capture the position and morphology of the weather system.
Based on the daily 2 m maximum and minimum temperature data from 1990 to 2019 in Sichuan Province, the temperature transitional weather processes have been analyzed statistically. Then a correction model of temperature change during transitional processes of temperature has been performed by using of NCEP/NCAR (National Center for Environmental Prediction/National Center for Atmospheric Research) daily reanalysis data and the LightGBM (Light Gradient Boosting Machine) algorithm.The results show that the area with the most temperature transitional processes is the slope transition zone between the plateau and the basin, while the least is in the basin. The number of temperature transitional processes in each region has an obviously seasonal differences with the most in spring and the least in winter, and the temperature transitional processes in spring is significantly more than those in the other three seasons. For the training set from 1990 to 2019,the LightGBM model has good performances with an overall accuracy of 78.64% and a mean absolute error of 1.35 ℃. For the independent testing set in 2020,the LightGBM model has an overall accuracy of 53.60% and a mean absolute error of 2.19 ℃, which are better than those of ECMWF (European Centre for Medium-Range Weather Forecasting), SCMOC and SPCO models.
In recent years, the artificial intelligence has made a breakthrough in image identification. In order to find out the practical value of artificial intelligence models in radar echo nowcasting in Wuhan City, the radar echo and precipitation observation data in Wuhan from 2015 to 2020 are used to train four deep learning models (PredRNN++, MIM, CrevNet and PhyDNet), then these trained models and radar echo observation data in flood season of 2021 are used to do nowcasting of radar echo. And on this basis, the precipitation processes are selected by using precipitation intensity and area indexes in Wuhan, and the performance of four deep learning models and optical flow method in radar echo nowcasting are tested and evaluated in Wuhan in flood season of 2021 in terms of mean square error (MSE), structural similarity index measurement (SSIM), probability of detection (POD), false alarm rate (FAR) and critical success index (CSI). The results are as follows: (1) On the whole, MSE of MIM model is the smallest, while its POD is the highest, and SSIM of MIM and PredRNN++ models are the highest. FAR of four deep learning models is lower than that of optical flow method, and it is the lowest for PhyDNet model. Except for CrevNet model, CSI of other three deep learning models is higher than that of optical flow method, and it is the highest for MIM model. (2) CSI of optical flow method is the highest during 0-12 minutes of forecast, while that of MIM model is the highest from 18 to 120 minutes, which shows the advantage of deep learning model for long prediction time. (3) With the increase of echo intensity, POD and CSI of four deep learning models and optical flow method decrease rapidly, while the variation characteristics of FAR of optical flow method and deep learning models are different. (4) For the regional precipitation processes, the prediction ability of deep learning models firstly reduces and then enhances significantly with the increase of precipitation intensity, while the optical flow method is insensitive to the change of precipitation intensity, so the increments of CSI of deep learning models are the highest under the strong precipitation processes compared with optical flow method. For the local convective precipitation processes with general intensity, the prediction ability of all models and optical flow method significantly reduces. (5) The analysis results of a rainstorm case show that deep learning models not only have prediction ability to the change of echo intensity to a certain extent, but also have better prediction ability to echo movement than optical flow method, so they have a good operational prospect.
Weighted mean temperature (Tm) is a key parameter in the retrieval of atmospheric precipitable water (PW) from ground-based Global Positioning System (GPS). In order to improve the accuracy and reliability of the retrieval of PW in Hainan Island, temporal variation characteristics of Tm calculated based on Haikou radiosonde data during 2008-2010 and the relation with meteorological factors at Haikou station are analyzed. On this basis, based on radiosonde and surface observation data during 2008-2012, single-factor and multi-factor Tm regression equations and Tm regression models with day of year factor are established at Haikou, and the models are validated by using radiosonde and surface observation data during 2013-2014. Based on the local Tm regression models, the ground-based GPS PW retrieval of Haikou is performed from May to October 2012, and the retrieval accuracy is verified. The results show that: by comparison of the true Tm, the RMSE of single-factor and two-factor local Tm models are 2.000 and 1.978 K, superior to Bevis and constant model. The local model of Tm has good consistency with Tm calculated by radiosonde data. The GPS PW from single-factor Tm model exhibits much stronger correlations with radiosonde PW than GPS PW based on Bevis model, and the RMSE of GPS PW by single-factor Tm model is lower than that based on Bevis model. Compared with the multi-factor linear Tm model, GPS PW based on the Tm model with day of year factor has significantly improved accuracy. The local models could meet the accuracy requirements of the PW from ground-based GPS data of Haikou.
In order to better exploit the detection advantages of wind profile radar in upper layer, the detection data of ST wind profile radar during 2014-2017 from Huainan Climate and Environment Observatory (HCEO), Institute of Atmospheric Physics, Chinese Academy of Sciences, were used to evaluate the detection performance of ST wind profile radar under different detection mode combinations, and the influence of meteorological environment on detection performance was explored. On this basis, the applicability of detection mode combinations was discriminated in practical operations. The results show that ST wind profile radar can achieve different detection purposes by combining high or low modes with switching altitude. However, there are differences in detection performance under different combinations of mode. On the one hand, the detection performance of radar gradually decreases before the mode conversion and rapidly increases after the mode conversion, while it gradually decreases with height under the high mode. On the other hand, the detection performance of radar doesn’t change significantly in process of the conversion, while it gradually decreases after the high mode to a certain height. In addition, the reducing degree of detection performance gradually increases as the transition altitude approach in autumn and winter. The precipitation reduces the detection performance in lower and middle layers of troposphere. So, we select suitable mode combination based on the detection performance of radar to atmospheric boundary layer, troposphere and stratosphere.
Based on rainfall enhancement operation records in Zhejiang Province and Doppler radar data, sounding data and hourly rainfall observations from automatic stations as well as MICAPS weather chart from June to November of 2018-2020, the indexes of Doppler radar echo for rainfall enhancement in summer and autumn in Zhejiang Province were researched with the inverse method after evaluation of rainfall enhancement effect in order to guide cloud seeding operation more scientific and reasonable. The results show that the shear line, upper trough and typhoon are the most favorable weather systems for artificial rainfall enhancement in Zhejiang Province, accounting for 28.6%, 21.4% and 21.4%, respectively. According to radar echo and precipitation characteristics, cloud types can be divided into stratiform cloud, cumuliform cloud, mixed cloud giving priority to stratiform cloud and mixed cloud giving priority to cumuliform cloud. Mixed cloud is the most common type for rainfall enhancement, accounting for up to 82.5%. The number of samples with obvious rainfall enhancement effect is small, only accounting for 13.4%. Radar echo intensity, radar echo top, vertical integrated liquid water and thickness of negative temperature layer are valid criteria for operating conditions. The indexes of Doppler radar echo for rainfall enhancement are different in different seasons and for different cloud types. So the indexes of Doppler radar echo and the discriminant equation of operating conditions should be established separately. Unreasonable operation was the main reason why we failed to get positive effect of rainfall enhancement, which accounted for 49.2% of all the samples. Many other reasons leading to failure of rainfall enhancement included but not limited to inappropriate time, position and object. The indexes of Doppler radar echo for rainfall enhancement established in this article are scientific and easy-to-use. These studies have evident significance to command cloud seeding operation.
In order to study the main geographical influencing factors of summer precipitation and the best interpolation method of precipitation in the complex Sichuan Basin, especially the mountainous area around the basin, Sichuan was divided into four regions by using cluster analysis based on 10 years (2010-2019) summer precipitation data of 157 automatic meteorological stations in Sichuan Province. The correlation analysis and the multiple regression analysis methods were used to screen out the geographical influencing factors of precipitation in each region. In addition to using the cooperative Kriging interpolation method, the traditional interpolation method is used to compare. The interpolation results are tested by cross-validation method. The results are as follows: (1) The geographic influencing factors that can be used to characterize the summer precipitation in Sichuan were mainly longitude, latitude, altitude, slope and normalized difference vegetation index. (2) Due to the diversity and complexity of the topography in Sichuan, the effect of precipitation interpolation after the division was better than that before the division.(3) When the number of precipitation influencing factors in the selected area was moderate, the coKriging interpolation method was better, and when the number of precipitation characterization factors in the selected area was single or too many, the radial Basis function interpolation method or empirical Bayesian Kriging interpolation method were more effective.
For precipitation forecast products with different methods and time, a large number of evaluation results often exist together. At present, we’re still lacking effective measures on how to analyze comprehensively and systematically these results. In this study, the agglomerative hierarchical cluster analysis is introduced to classify and analyze the different evaluation results of different forecast products, based on a grid precipitation forecast dataset of each member of the national forecast technology and method competition of CMA from June to September 2019, the central station guide forecast (SCMOC) of the National Meteorological Center, the seamless analysis and forecasting leading-edge system forecast of Chinese Academy of Meteorological Sciences and objective forecast products of 31 provinces (municipalities and autonomous regions), the global modelforecast of ECMWF (European Centre for Medium-Range Weather Forecasts) and NCEP (National Centers for Environmental Prediction). The results show that the agglomerative hierarchical clustering results can clearly distinguish their similarities and differences between different forecast products. The different evaluation indicators lead to different clustering results, but the forecast products with high similarity are still divided into a same subclass. The identification effect of four different inter-class similarity measurement methods on categories characteristics was different, and the Ward method was followed by Complete, Average and Single method from clear to fuzzy. In addition, the precipitation prediction ability for different administrative regions and forecast products was different, the accuracy of rain probability forecast in North China and East China was better than that in other regions, and most objective forecasts to rain probability and precipitation relative error were better than model forecast of ECMWF, while they to heavy precipitation were worse than ECMWF model, there are still greater difficulties in interpretation to heavy precipitation forecast.
With continuous growth of forecast service demand and increasingly refined forecast content, the forecast of short-term heavy precipitation above 20 mm·h-1 can not meet the forecast service demand fully. It is very necessary to carry out research on forecast methods about short-term heavy precipitation with different rainfall intensity. The 51 355 samples of short-term heavy rainfall from national and regional meteorological stations in nine provinces and one city in southern China from June to August during 2016-2019 were divided into four rainfall grades according to their rainfall intensity (R), namely, I: 20≤R<30 mm·h-1, II: 30≤R<50 mm·h-1, III: 50≤R<80 mm·h-1, and IV: R≥80 mm·h-1. The samples of all rainfall grades were spatiotemporal matched with the initial field of CMA-MESO (China Meteorological Administration mesoscale model) in the same period, and the percentile statistics were applied to 22 physical quantities extracted from these samples. The XGBoost (extreme gradient boosting) machine learning method was used to rank importance of those 22 physical quantities to determine their weight coefficients. Based on the continuous probability prediction method, the ascending and descending half ridge functions were selected as the membership function, the probability prediction models of short-term heavy precipitation with different rainfall grades were established. The real-time operational prediction was carried out in flood season of 2020 using these prediction medels, and the hourly probability prediction products of short-term heavy precipitation with different rainfall grades for 0-36 h prediction time during 15 heavy rainstorm precesses in Hubei Province from June to August 2020 were tested. The results show that for the grade I probability prediction products, the TS score (0.145) using 60% as threshold works best, with a corresponding hit rate of 55.7%; for the grade II probability prediction products, the TS score (0.083) using 65% as threshold works best, with a corresponding hit rate of 39.1%; for the grade III probability prediction products, the TS score (0.03) using 70% as threshold works best, with a corresponding hit rate of 21.7%; for the grade IV probability prediction products, the TS score (0.005) using 80% as threshold works best, with a corresponding hit rate of 5.8%.The results also suggest that probability prediction products help to correct the CMA-MESO model in predicting short-term heavy precipitation with different rainfall grades at the same time. The hourly prediction test of three heavy precipitation processes shows the hit rate of 40%-80%, the false rate of 50%-90%, and 36 h prediction time for the grade I probability forecast products, which are generally better than CMA-MESO precipitation forecast at the same time. A model was established to forecast short-term heavy precipitation with different grades in this study, and it outperforms existing numerical models and can be a good reference for meteorologists to forecast short-term heavy precipitation and correct precipitation forecast biases in CMA-MESO.
In the flood season (from June to August) of 2020, Gansu Province experienced intensive precipitation with long duration and wide ranges. The performances of three global models (ECMWF, GRAPES_GFS and NCEP_GFS) and four regional models (GRAPES_3 km, GRAPES_LZ10 km, GRAPES_LZ3 km and regional model SMS-WARMS in East China) for 24-hour accumulated precipitation forecast were evaluated in this paper. The main results are as follows: (1) The ECMWF model surpassed the other two global models in forecast performance, while among regional models, the GRAPES_3 km and the SMS-WARMS were better, and the latter was more stable. (2) The regional models had lower accuracy of rain probability forecast and TS, ETS, POD than those of global models for light and moderate rain, but for rainstorms they outperformed global models; the POD and Bias of regional models for heavy rain and rainstorms were significantly higher than those of global models. (3) According to the differences of 500 hPa circulation pattern, the precipitation in Gansu could be divided into two types including subtropical high marginal type and low trough type. Four subtropical high marginal precipitation processes and three low trough precipitation processes in flood season of 2020 were tested and evaluated. For global models and regional models, they all had better capability in predicting precipitation with different magnitudes for the former type than the latter one. The ECMWF model and regional models were better than the NCEP_GFS model and the GRAPES_GFS model in predicting heavy rain and rainstorm. Among global models, the ECMWF model had the best forecast effect for the two precipitation types, and the East China regional model had the best forecast effect for the two precipitation types among regional models. (4) All the seven models had good forecasting capability for the spatial orientation of moderate and heavy rain for both rainfall types, while the forecast effect of rainfall location for subtropical high marginal type was better than that of low-trough type, but the predicted precipitation intensity was stronger than observations, especially for the center of precipitation.
A forest fire occurred in Qinyuan County, Shanxi Province on June 5, 2020. Based on the analysis of weather situation, radar echo, lightning location and other multi-source meteorological data, and European Centre for Medium-Range Weather Forecasts (ECMWF) fifth-generation global atmospheric reanalysis (ERA5),the evolution characteristics of meteorological elements including temperature and precipitation were analyzed in the early stage and during the forest fire, and combined with the field investigation, the cause of the forest fire were given. The results show that the forest fire was caused by a positive lightning, the lightning point was at the edge of convective cloud and the lightning occurred at 15:39 BST on June 5, with the current intensity of 42.2 kA.There was no precipitation in the areas around fire site for 2 consecutive days in the early stage, the 2 m temperature in the areas around fire site increased significantly on the day of the forest fire, with the air temperature of 30-33 ℃, and the precipitation was less than 0.1 mm and it was breezy.
Based on the SRTM (shuttle radar topography mission)data, the ground clutter and other clutters around Tianshui radar station were filtered, then the Z-I function with localized parameters was established on the basis of six precipitation processes with three types in Southeast Gansu after filtering the ground clutter and other clutters of radar data, and at last the reflectivity factor of Xifeng new generation weather radar in Qingyang was compared with the one in Tianshui within the coincidence range. The results show that SRTM data can well simulate the distribution of ground clutter; radar reflectivity is ahead of precipitation; the Z-I function with localized parameters in Tianshui, which had a smaller A and bigger b, is significantly different to common ones; Tianshui new generation weather radar may have a systematic problem of low echo intensity.
The prediction based on dynamic downscaling prediction technology of the climate extension of weather research and forecasting (CWRF) model to summer precipitation has a certain deviation, so it is difficult to achieve accurate prediction. This paper analyzed the correlated meteorological elements with summer precipitation based on the climatic characteristics of summer precipitation in the main land of China. And on this basis, the reforecasts of summer precipitation by CWRF model in China during 1996-2019 were corrected by using the combined method of dendritic network (DD) and artificial neural network (ANN). Finally, the correction effect was tested by mean square error (MSE), anomaly correlation coefficient (ACC) and temporal correlation coefficient (TCC), etc. The results show that the correction effect to summer precipitation based on the artificial dendritic neural network (ADNN) algorithm model was better than the historical reforecasts of CWRF model in China. The ACC and TCC both increased by about 0.10, MSE dropped by about 26%, and the overall trend anomaly test scores improved by 6.55, which indicated that the ADNN machine learning method could achieve correction to summer precipitation forecasts of CWRF model to a certain extent, thus it could improve the accuracy of precipitation forecasts of CWRF model.
Based on topography and forecasted 3-hour wind fields, relative humidity fields initialed from 20:00 BST and 08:00 BST by using the SWCWARMS (southwest center WRF ADAS real-time modeling system), the precipitation correction equation was constructed by calculating the terrain precipitation estimates combined with precipitation fields forecasted by SWCWARMS. The daily precipitation, precipitation processes in Sichuan Basin and in western Sichuan Basin during flood season from June to August during 2018-2020 are corrected, and the precipitation in the steep terrain transition zone from the eastern slope of western Sichuan Plateau to the western Sichuan Basin was tested and evalcated only. The results are as follows: (1) The TS of the precipitation correction value with each magnitude was improved compared with TS of forecasted precipitation by the SWCWARMS. The correction effect of precipitation forecasted initialed from 20:00 BST was better than that initialed from 08:00, and the correction effect of the precipitation processes in western Sichuan Province was the best for heavy rain and above. Compared with the SWCWARMS, the relative improvement rates of TS of corrected value of precipitation with heavy rain, torrential rain and heavy downpour were 19%, 25% and 37%, respectively, the hit ratio was higher, the false alarm rate and miss rate were decreased significantly. (2) The correction equations of precipitation had a good correction effect on both torrential rain and general precipitation cases of precipitation processes in western Sichuan Province occurring in the steep terrain transition zone, even for cases of precipitation area predicted by the SWCWARMS was far from the actual situation.
Using the radiation observational data at Hangzhou station and the hourly simulation output data by Zhejiang WRF ADAS real_time modeling system (ZJWARMS), the forecast effect of radiation was evaluated in 2019. On this basis, the MOS correction models under different weather conditions were built based on ten meteorological factors from ZJWARMS output, including shortwave radiation flux, cloud cover, surface temperature, humidity, etc. The results showed that ZJWARMS had a good ability in simulating the diurnal variations of surface solar radiation, the correlation coefficient between simulation value and observation value of radiation was 0.82. However, the system generally overestimated surface solar radiation. The lowest error appeared on sunny days, while there was a significant increase of error on overcast days or rainy days. The correction model could effectively improve the prediction effect of surface solar radiation. After the correction of MOS model, the mean absolute percentage error decreased from 273.4% to 46.3%, and the root mean square error dropped from 246.7 W·m-2 to 105.0 W·m-2. The correction effect exited difference in different months, and that in August was the best, and the mean absolute percentage error decreased from 126.6% to 26.3%, while the correction error was larger in April, the mean absolute percentage error was 56.6%.
Based on the multi-mode precipitation gridded forecast data, observation data at meteorological stations of Qinghai Province and precipitation gridded analysis product of CMA multi-source precipitation analysis system (CMPAS), the prediction performance of models to heavy precipitation cases in Qinghai Province from July to August 2020 were comparatively tested by using traditional verification method such as threat score (TS) and spatial verification method such as fraction skill score (FSS) of neighborhood method and object-oriented diagnostic evaluation method (MODE). The main conclusions are as follows: (1) The traditional TS scores of global model of European Center of Medium-range Weather Forecasts (ECMWF) and National Center for Environmental Prediction (NCEP), China Meteorological Administration global assimilation forecast system (CMA-GFS) and GRAPES regional mesoscale numerical prediction system (GRAPES-Meso) to light rain and above were higher, and the prediction performance difference of four models to light rain was little, but the models with the highest score under different verification methods were slightly different. (2) Compared with the observation, the forecasted locations of four models to moderate rain and above were generally to the west. The traditional TS scores of moderate rain and above were significantly different, but the performance score of models under different verification methods was relatively consistent. (3) Compared with the observation, the forecasted location of four models to heavy rain and above was generally to the north. The prediction ability of each model to heavy rain and above was poor, and the traditional TS scores of heavy rain and above were equal to 0, while FSS scores could effectively improve the evaluation ability to models difference, and MODE could give the specific performance of corresponding object attributes, which provides valuable reference for model application, but it was more sensitive to the selection of verification parameters.