Vol.32, NO.1, 2021

Display Method:
Reviews
Application of Artificial Intelligence Technology to Numerical Weather Prediction
Sun Jian, Cao Zhuo, Li Heng, Qian Simeng, Wang Xin, Yan Limin, Xue Wei
2021, 32(1): 1-11. DOI: 10.11898/1001-7313.20210101
Abstract:
Numerical weather prediction technology plays an increasingly important role in improving accuracy and service level of modern weather forecast. With the development of observation system and higher resolution and complexity of the numerical weather prediction model, the products of numerical weather forecast have been greatly improved in quantity and quality, and can offer rich information at high spatial-temporal frequency. However, such a large amount of prediction data are not fully explored. Artificial intelligence has achieved great success in many fields, such as pattern recognition and natural language processing, which provides an opportunity for further improving numerical weather prediction. It's also employed in initialization, numerical model and production of weather forecast service, involving observation system, data assimilation, model integration, ensemble forecast and high-performance computing methods. Both the accuracy of forecast results and computational efficiency have been improved by using error correction, parameter estimation, local surrogate model and so on. In addition, some end-to-end neural network models also show the potential of pure data-driven weather forecast. These models use spatial-temporal observation data as input and directly output the prediction results in terms of deterministic results or probabilities. Some of them perform well in short-term severe convective weather, precipitation, and long-term climate forecast. Existing works employ various artificial intelligence technology methods, mainly including large-scale calculation of neural network, feature analysis, interpretability, and customized loss function. However, there are still some challenges, the potential of artificial intelligence needs to be further explored. Some issues should be carefully considered, including weak interpretability, uncertainty analysis and the coupling with conventional numerical models, and how to use physical knowledge to guide the design of artificial intelligence model is also worth addressing. To deal with these challenges, some promising suggestions are proposed. Bayesian network and causal network will help to establish more comprehensive and profound feature engineering. Using Bayesian inference to generate distribution characteristics of current meteorological states may be an alternative to efficient and effective uncertainty quantification. The development of some standard workflow and framework will contribute to the coupling of conventional numerical model and artificial intelligence module. Successful artificial intelligence applications in weather forecast require deep cooperation between meteorological experts and computer experts who focus on artificial intelligence and high-performance computing.
A Review on Stable Precipitation Type Forecast in Winter
Zhao Linna, Mu Xiuxiang, Ma Cuiping, Wang Xiujuan, Li Dihua
2021, 32(1): 12-24. DOI: 10.11898/1001-7313.20210102
Abstract:
The accurate identification of precipitation type at ground level is one of the greatest difficulties for forecasters during winter. Special types of precipitation can be a threat to public safety and human health and can disrupt transportation and commerce, causing seriously loss of the economy. Winter precipitation may also cause serious disasters to aircraft navigation. In those situations, the consequences can be catastrophic, with heavy and prolonged freezing precipitation, collapsed power lines causing prolonged power outages, transportation networks of many types completely paralyzed, and even major long-term damage to infrastructure and vegetation in the most severe cases. Therefore, the accuracy of precipitation type is crucial for winter precipitation forecast. Accurate predictions from weather forecast models of timing (onset and duration), intensity, spatial extent, and phase (i.e., precipitation type) are crucial for decision-making and can help minimize the potential impacts. The research progress of precipitation type forecast in recent decades is investigated. The methods and techniques for predicting precipitation phases are reviewed systematically, which can be roughly divided into three categories. The index criterion methods are based on observations, numerical weather prediction weather predictions on thickness, area of warm atmosphere, significance level temperature, regression equation for vertical temperature profile, and model diagnosis. Some of those methods are highly dependent on the accuracy of the numerical model. The second type of methods are based on the microphysical processing scheme of numerical weather prediction model and ensemble prediction system. It includes microphysical scheme method and ensemble prediction method. The last type is the artificial intelligence prediction method including decision tree, artificial neural network, and deep learning etc. In recent years, sophisticated microphysical parameterizations schemes are widely used in high resolution regional forecast models, which help with precipitation-type prediction. The forecast accuracy of precipitation type model has been improved, which has become an important product support in precipitation type forecast. For instance, the precipitation type prediction product of ECMWF and the probabilistic prediction of precipitation type by ECMWF ensemble prediction. The probabilistic prediction has further improved the prediction skills compared with the deterministic model. However, even with such complex algorithms of NWP, correctly predicting what phase of precipitation ends up at the ground remains a challenging task. Besides this, many researches on the formation mechanism of microphysical processes are difficult to be applied to the precipitation type prediction, so it still needs continuous efforts to apply these achievements to improve the precipitation type prediction skill of numerical prediction model and increase the accuracy of precipitation type prediction by artificial intelligence and other technologies.
Articles
Evaluation Index Construction and Hazard Risk Assessment on Apple Drought in Northern China
Yang Jianying, Huo Zhiguo, Wang Peijuan, Wu Dingrong, Mao Hongdan, Kong Rui
2021, 32(1): 25-37. DOI: 10.11898/1001-7313.20210103
Abstract:
It is of great merit to construct apple drought index and analyze its hazard risk so as to support apple drought monitoring, prevention and mitigation, as well as agricultural disaster insurance management. Based on meteorological data, associated with historical disaster and phenophase data, apple drought index (DI) in northern China is firstly constructed, which fully considered previous and current water demand and precipitation supply. Afterwards, historical disaster remodeling, disaster sample reconstruction and process-based historical disaster analysis are comprehensively used as key technologies in evaluating the level of apple drought, integrating the independent sample T-test, Kolmogorov-Smirnov(K-S) test, cumulative probability inverse method, etc. Meanwhile, the apple drought risk is consequently estimated to seek the characteristics of apple drought hazard in detail. The results show that there are significant differences by independent T-test of drought index (DI) between historical recorded disaster samples and the non-disaster samples in three apple tree phenophases, i.e., the tree germinating to flower budding period, flower budding to full bloom and full bloom to mature periods (passing the test of 0.05 level). Therefore, the apple drought index constructed can effectively represent the drought disaster in different stages of apple development. The threshold of DI in the same level is higher in tree germinating to flower budding period, followed by flower budding to full bloom and full bloom to mature periods. The drought risk during flower budding to full bloom is high with regional average apple drought hazard index (M) of 0.44, followed by tree germinating to flower budding period and full bloom to mature period, with reginal average M of 0.40 and 0.25, respectively. Furthermore, the Bohai Bay region and northern Loess Plateau are detected as high-risk areas of apple drought. The evaluation method of apple drought based on historical disaster processing and re-analysis can provide new ideas for economic forest and fruit meteorological disaster research. The results of apple drought hazard risk assessment could provide evidence for the prevention and mitigation of apple drought in northern China.
WOFOST Model Parameter Calibration Based on Agro-climatic Division of Winter Wheat
Li Ying, Zhao Guoqiang, Chen Huailiang, Yu Weidong, Su Wei, Cheng Yaoda
2021, 32(1): 38-51. DOI: 10.11898/1001-7313.20210104
Abstract:
Crop model parameter calibration is an important work of extending point-scale crop model to regional application.Using K-means method with the main meteorological factors affecting the growth and yield formation of winter wheat obtained from 113 meteorological stations from 1981 to 2010 as zoning indicators, Henan Province is divided into five different agro-climatic ecological zones and the cumulative temperature parameters are calculated for each zone. Based on the observations during 2013-2017, nine sensitive parameters are obtained by using Sobol global sensitivity analysis method to analyze and select crop parameters with total sensitivity index greater than 0.01. The sensitive parameters selected from different agro-climatic ecological zones of different winter wheat varieties are highly consistent. A cost function is constructed with yield and leaf area index(LAI), and each partition is calibrated for sensitive parameters using Differential Evolution Markov Chain method. The results show that the simulated leaf area index in the different agro-climatic ecological zones are in good agreement with the observed values, the root mean square error (RMSE) using the posterior mean value of regional parameters adjustment to simulate the LAI of key growth periods is 0.655, which is obviously higher than that of using default parameters or using the same set of optimized parameters in the whole study area. Results show that the WOFOST model based on agro-climatic division can accurately simulate the growth process of crops. In terms of yield estimation accuracy, the yield simulation accuracy of regional parameter adjustment is also significantly improved. The best accuracy of simulated yield is achieved by using the posterior mean of regional parameters and RMSE is 672.016 kg·hm-2, 70.55% reduction than the yield simulation error when using the default parameters, or 48.75% reduction than the yield simulation error when the same set of optimized parameters (posterior mean) are used for the entire area. The method takes advantage of the knowledge of agro-climatology with the scientific and efficient Differential Evolution Markov Chain optimization algorithm to provide a scientific and theoretical basis for the application of crop models and optimization of regional model parameters through rational and efficient zonal calibration of the study area.
Effects of Improving Evapotranspiration Parameterization Scheme on WOFOST Model Performance in Simulating Maize Drought Stress Process
Cai Fu, Mi Na, Ming Huiqing, Zhang Shujie, Zhang Hui, Zhao Xianli, Zhang Yushu, Zhang Bingbing
2021, 32(1): 52-64. DOI: 10.11898/1001-7313.20210105
Abstract:
To solve the problem on poor performance of crop growth model in simulating crop growth process under water stress, three schemes including improving evapotranspiration parameterization scheme with Penman-Monteith method, building dynamic crop coefficient (Kc) and considering simultaneously two above-mentioned solutions which are respectively named as the PM, CC and PMCC schemes are used to improve WOFOST model. Their effects on model performance in simulating maize drought stress process are evaluated based on experiments of different sowing dates on 20 April, 30 April and 10 May conducted in Jinzhou in the year with normal precipitation (2012) and the dry year (2015 and 2018). The results show that compared with the default model, PM scheme plays a role in increasing potential evapotranspiration and transpiration rate in 2012, while CC scheme decreases (increases) transpiration rate as Kc is larger (smaller) than the model default value 1, respectively. Three schemes hardly affect simulation accuracies of soil moisture in rooted zone, total above ground production, leaf area index (ILA) and yield. In 2015, PM scheme makes ILA, total above ground production yield and soil moisture dramatically smaller than those with the default model after the jointing stage of maize. It also increases (decreases) transpiration rate before (after) the whorl stage of maize. After improving the model with CC scheme, ILA, total above ground production, yield and soil moisture are slightly larger than those simulated by the default model after the whorl stage, and the simulated transpiration rates are smaller (larger) than those simulated by the default model before (after) the whorl stage of maize. Nevertheless, the above-mentioned variables simulated by the improved model with PMCC scheme range between those simulated by the model with PM and CC schemes, and ILA, total above ground production and yield are obviously closer to the observations. Specifically, the mean increments of simulation accuracies for all growth periods in three sowing dates are 6%, 21% and 3% for ILA and 8%, 8% and 14% for total above ground production, respectively. The simulation accuracies of yield for three sowing dates increase by 66%, 63% and 66%, respectively. In 2018, the simulation accuracies of total above ground production and yield for the sowing dates on 20 April and 30 April are obviously improved by PMCC scheme, and increase by 5%, 1% in total above ground production as well as 32%, 5% in yield. Therefore, the model performance with PMCC scheme in simulating maize growth is significantly improved under water stress.
Chemical Characteristics of PM10 at Background Stations of Central and Eastern China in 2016-2017
Jiao Jian, Jia Xiaofang, Yan Peng, Cao Fang, Fang Dongqing, Ma Qianli, Yu Dajiang, Chu Jinhua
2021, 32(1): 65-77. DOI: 10.11898/1001-7313.20210106
Abstract:
The characteristics of PM10 and its chemical composition from December 2015 to December 2017 are studied at three regional atmospheric background stations (Longfengshan, Lin'an, and Jinsha) in central and eastern China. The water-soluble ions (F-, Cl-, NO3-, SO42-, PO43-, Na+, K+, NH4+, Ca2+, Mg2+), carbon-containing compounds (OC, EC) and main elements in PM10 samples are analyzed. The average PM10 concentration at Lin'an (62.2±36.6 μg·m-3) during the whole period is the highest, followed by Jinsha (57.6±31.8 μg·m-3), and that at Longfengshan (57.5±55.3 μg·m-3) is the lowest. At those stations, the annual PM10 mass concentrations are lower than the national second-level air quality standard. The concentrations of PM10 in 2016-2017 show a downward trend, with the reductions of concentrations about 29.3% at Lin'an and 26.2% at Jinsha compared with the results during the year of 2013.The concentrations of SO42-, NO3-, and NH4+ during the whole period are 9.9, 8.2 μg·m-3 and 3.7 μg·m-3 at Lin'an, and 10.2, 6.7 μg·m-3 and 2.6 μg·m-3 at Jinsha, respectively, which are all higher than those at Longfengshan, with the average concentrations of 5.9 μg·m-3 for SO42-, 4.9 μg·m-3 for NO3-, and 2.1 μg·m-3 for NH4+. However, the ratio of NO3-/SO42- at Longfengshan and Lin'an are relatively higher. The concentrations of OC and EC during the whole period are 10.1 μg·m-3 and 2.7 μg·m-3 at Longfengshan, 6.1 μg·m-3 and 3.1 μg·m-3 at Lin'an, and 4.7 μg·m-3 and 2.3 μg·m-3 at Jinsha. Compared to the year of 2013, the concentrations of SO42-, NH4+ and OC during the whole period decrease by 38.1%, 26.0%, and 55.6% at Lin'an, respectively, and decrease by 46.3%, 51.9%, and 44.7% at Jinsha, while the concentrations of NO3- increase by 12.3% and 15.5% at Lin'an and Jinsha, respectively. The concentrations of EC decreases by 27.9% at Lin'an, but increases by 4.5% at Jinsha. The ratio of NO3-/SO42- increase obviously at all the stations, which indicates the increases of nitrate aerosols, due to the continuously control of coal combustion emissions and increase of car ownership.The seasonal concentrations of PM10, NO3-, EC and the ratio of NO3-/SO42- are lowest in summer at three stations. Meanwhile, the mass concentrations of SO42-, NO3- and NH4+ are basically higher in winter than in other seasons.
An Objective Hailstorm Labeling Algorithm Based on Ground Observation
Liu Bojun, Zhang Yaping, Li Zhongju, Han Xiao, Lu Hua, Zhang Yong
2021, 32(1): 78-90. DOI: 10.11898/1001-7313.20210107
Abstract:
Data labeling makes the key foundation of building data sets for deep learning, especially in the intelligent forecasts of severe weather, such as hail, the observations of which are lacking. Disaster report is a kind of information that describes the details of meteorological disasters which is collected by meteorological information officer. Due to the high coverage rate of informants throughout villages and communities, disaster report is considered to have good consistency and high spatial resolution. However, the vague description of disaster occurrence time in disaster report limits its application. To solve this problem, 13 hail cases(divided into reference set and verification set) with accurate occurrence time in hail reports in Chongqing during 2008-2019 are selected, and an objective hailstorm labeling algorithm based on actual hail observations is developed using fuzzy logic algorithm. In order to obtain a reasonable match between hail occurrence location and convective storm, the distance between the centroid of the storm and initial guess location of hail occurrence, the maximum values of reflectivity, height of 45 dBZ reflectivity, vertical integral liquid water content and echo top are selected as discriminant factors, and the storm is identified by the storm cell identification and tracking (SCIT) algorithm. In reference set, 7 hail cases can be labeled correctly and only 1 case is failed to identify storms. The time bias between the labeling time and ground disaster report is less than 6 minutes during 5 cases. Inspected by verification set (5 cases in 2019), the algorithm labeling accuracy is 100% and the matching degree ranges from 0.887 to 1.000. Furthermore, the algorithm is applied to 22 hailstorm labeling cases lacking accurate time, and the results are compared with the manual labeling results by forecasters. Subjective and objective methods tend to identify the same storm cell and have little impact on data set construction. Forecasters tend to label the same storm cell 6-12 minutes ahead. Further analysis shows that the size of hail has no significant effects on the labeling result. The algorithm is not sensitive to the occurrence time of hail disaster, and it can give reliable labeling results for both long time living storm and local hail disaster. However, when the identification algorithm fails to figure out storms, or the initial guess location deviation is large, it will have a significant negative impact on the labeling results.
Application of QVP Method to Winter Precipitation Observation Based on Polarimetric Radar
Guan Li, Dai Jianhua, Tao Lan, Yin Chunguang, Meng Fanwang
2021, 32(1): 91-101. DOI: 10.11898/1001-7313.20210108
Abstract:
Winter precipitation events, especially those involving transitions of precipitation type, continue to pose a formidable forecasting and nowcasting challenge to operational meteorologists. The polarimetric radars provide unique insight into microphysical processes in clouds and precipitation. Using polarimetric radars in conjunction with thermodynamic information is a promising way for better winter precipitation detection. To explore the microphysical characteristic and the internal structure of winter precipitation over eastern coast of China, the data collected by the WSR-88D polarimetric radar at Nanhui, Shanghai are exploited by the quasi-vertical profile (QVP) method. The QVP method involves azimuthal averaging of radar reflectivity factor at horizontal polarization(ZH), differential reflectivity(ZDR) and the copular correlation coefficient (ρhv) at high antenna elevation, presenting QVPs in a height-time format. QVP generation is an efficient way to examine the temporal evolution of microphysical processes governing precipitation production and to display physical links between polarimetric signatures aloft in the ice-phase or mixed-phase parts of the clouds. In 3 different synoptic system governing snow cases affecting Shanghai, the QVPs retrieved from dual-polarization radars at elevations of 19.5ånd 9.9åre demonstrated to successfully monitor the evolution of melting layer and Bergeron process. They also provide opportunities to discriminate between the processes of snow aggregation and riming with the joint analysis of reanalysis data and observations of radio sounding, auto weather station, disdrometer and wind profiler radar. The vertical observation by cloud radar data is used to compare with QVP retrieved profile. Additionally, for discontinuous or multi-scale synoptic precipitation, a selected azimuthal averaging QVP technique is introduced to separate QVPs into before and after the synoptic system for detailed comparisons and monitoring microphysical processes leading to precipitation formation. The method is demonstrated to monitor important microphysical signatures as well as following precipitation development. In conclusion, the procedure for generating quasi-vertical profiles of polarimetric radar variables is very simple and straightforward, and the QVP plots in the height-time format can be produced in real time for operational polarimetric weather radars as a standard product, which is very easy to implement and very promising to use along with traditional weather radar products of PPIs and reconstructed RHIs. The QVP methodology is particularly effective because of its local coverage and high precision as well as its potential for nowcasting.
Automatic Recognition Algorithm of Convergence Region Based on Relative Storm Radial Velocity Field
Zhu Li, Kang Lan
2021, 32(1): 102-114. DOI: 10.11898/1001-7313.20210109
Abstract:
An algorithm for automatically identifying the mid-altitude radial convergence from the storm-relative radial velocity field is proposed. The algorithm first identifies the positive-negative velocity segments in each radial direction on the single-elevation radial velocity field, before pairing them to form a radial convergence segment. A two-dimensional radial convergence block is obtained through horizontal correlation analysis, and then three-dimensional radial convergence body of the storm is obtained through vertical correlation analysis. Thus, the parameters such as strength, thickness and center height are calculated.The algorithm is verified using two squall line radar data with a typical "positive-negative velocity zone pairs" radial convergence characteristics, and the results show that the radial convergence feature identified in the relative storm radial velocity field is more complete than the original radial velocity field. The flow field of the meso-small-scale weather system is mainly composed of rotation and translation combined with ascending motion. When the translational motion speed is greater than the rotational speed, the shear (rotation, convergence, or divergence) of the system in the basic radial velocity field may be affected, while using the relative storm radial velocity can overcome this to identify the mid-level convergence better. A batch experiment of 10 thunderstorms and strong convective weather indicates the recognition accuracy of this algorithm is 82.4%, including a typical MARC features.Statistical analysis of the correlation between characteristic parameters and strength of squall line winds shows that the average radial convergence strength, maximum radial convergence strength, thickness have good positive linear correlations with wind speed. The correlation coefficient between convergence intensity and wind speed is the largest, reaching 0.79. According to the radial convergence characteristic parameter value, the intensity of the ground gale can be roughly judged, which provides a certain reference for the monitoring and early warning of convective gale and disaster assessment. The radial convergence feature identified by the algorithm can alert squall line gale about 30 minutes in advance. Therefore, the application of this algorithm will effectively improve the advancement of the warning signal release time.
Correction of Precipitation Forecast Predicted by DERF2.0 During the Pre-flood Season in South China
Wang Juanhuai, Li Qingquan, Wang Fang, Yang Shoumao, Hu Yamin
2021, 32(1): 115-128. DOI: 10.11898/1001-7313.20210110
Abstract:
There are two main types of precipitation during the pre-flood season in South China, frontal precipitation in early period and monsoon precipitation in late period. It is related to not only the tropical system, but also the cold air in the middle and high latitudes. The extended range forecasting skills of the precipitation in the pre-flood season which depend on the atmosphere-ocean interaction and the internal changes of the atmosphere are still very low. There are biases in models compared to the observations, which makes it hard to directly use model in operational forecast. Therefore, in order to better apply model forecast data to extended period forecasts, the precipitation biases are corrected during the pre-flood season from 1983 to 2019 produced by the Dynamic Extended Range Forecast Operational System version 2.0 (DERF2.0) based on a non-parameter Quantile-Mapping (QM) correction method. Daily precipitation observation data from 261 stations in South China from 1983 to 2019 are selected for evaluation. On the basis of probability forecast of the original model outputs, the model biases are then corrected using monotone cubic spline interpolation combined with the observation. The models are established by cross samples and independent samples to validate the correction method's performance by absolute difference/percentage difference, root mean square error, temporal correlation coefficient and pattern correlation coefficient. It is found that the QM correction method can improve the model forecasting skills by effectively eliminating the systematic deviation of the model. It shows that the improvement of the method remains stable with different lead times and magnitudes of model biases. Further analysis shows that the main locations and average intensities of precipitation show better consistency with observation after correction. The QM correction method can generally capture the trend difference between the model and the observation, and effectively improve the inter-annual variability of model, but it has a poor ability for extreme events. On the other hand, the revised effect of the statistical scheme according to different percentile intervals is also significant. In addition, it shows that the correction performances of prediction are more consistent with the hindcast result.