Vol.31, NO.3, 2020

Display Method:
Review of Machine Learning Approaches for Modern Agrometeorology
Li Ying, Chen Huailiang
2020, 31(3): 257-266. DOI: 10.11898/1001-7313.20200301
With the development of smart meteorology and precision agriculture, modern agrometeorology tasks demand for efficient analyzing and processing of massive agricultural and meteorological data, including multi-source remote sensing images. Machine learning technology can powerfully contribute to the development of agrometeorology and the innovation of agrometeorological service mode. A targeted overview on the related work of machine learning in modern agrometeorology domains is given, including mapping and zoning, detection and observation, yield prediction, and parameter prediction, with specially focuses on deep learning approaches for agrometeorology and the latest research progress in recent years. From the aspect of mapping and zoning, machine learning technology can be combined with remote sensing images to map land cover and crop types in different scales, and can also be combined with remote sensing data, soil data and statistical data to make thematic maps of crop growth and vegetation quality and to zone crop management areas. From the aspect of detection and observation, machine learning technology is successfully used to detect weeds in field images. Deep learning technology is used in plant phenotype observation, disease and pest detection, obstacles and anomaly detection, fruit counting and so on with high accuracy, which could greatly improve the level of agrometeorology automatic observation. From the aspect of yield prediction, machine leaning technology combined with remote sensing time series data, meteorological data and soil data is successfully used to predict the yield of different crops in different scales. Machine learning technology also has great application potential in loss assessment for agrometeorological disasters. From the aspect of parameter prediction, the hydrological, soil and crop parameters concerned by agrometeorology tasks such as evapotranspiration, leaf area index, soil moisture and nitrogen can be accurately inverted and predicted by the combination of machine learning technology, meteorological data and remote sensing data. Overall, among the traditional machine learning approaches, support vector machine and artificial neural network are the most widely used and the most ideal methods. In recent years, ensemble-based methods such as random forest and gradient boosting machine have generally achieved higher accuracy than kernel methods, while deep learning approaches have achieved higher accuracy than ensemble-based methods in some tasks. In the future, it is necessary to verify the applicability and advancement of more different machine learning approaches, especially deep learning approaches in more different agrometeorological tasks, and choose the most suitable machine learning technology for each specific task in modern agrometeorological services according to the data using, which will help to meet new challenges and opportunities of the modern agrometeorology development.
Construction of the Model for Soil Moisture Effects on Leaf Photosynthesis Rate of Winter Wheat
Wang Peijuan, Ma Yuping, Huo Zhiguo, Yang Jianying, Wu Dingrong
2020, 31(3): 267-279. DOI: 10.11898/1001-7313.20200302
The rate of leaf photosynthesis, which is sensitive to soil moisture, is one of the most important parameters to characterize the photosynthetic capacity of plants. Constructing a model which can reveal effects of soil moisture on leaf photosynthesis rate of winter wheat will be helpful to accurately understand the photosynthesis and yield formation. A total of 310 photosynthesis rate samples under different soil moistures, including 227 drought stress samples in 50 tests and 83 waterlogging stress samples in 14 tests, are jointly collected from 17 winter wheat cultivars at 11 experimental sites via the published references. However, photosynthesis rates of winter wheat are quite different between different cultivars, different developmental stages, and different experimental sites. Normalized photosynthesis rate coefficients for winter wheat are derived by calculating the ratio of leaf photosynthesis rate under different water stresses and CK. And then, segmental and exponential models are established for effects of drought and waterlogging stresses on leaf photosynthesis rate of winter wheat, respectively. The model for soil moisture effects on leaf photosynthesis rate of winter wheat (SMEP) is correspondingly constructed. Photosynthesis rate coefficients of winter wheat leaves show the trend of "stable low value-linear increase-stable high value-slow decrease" with the increase of soil relative moisture. Meanwhile, photosynthesis rate coefficients exhibit characteristics of "slow decline-rapid decline" with the prolongation of waterlogging stress. Four tests, including back-training test, extrapolation test, single-site test and certain developmental stage test, are also done to validate the SMEP model. Generally, the results simulated by the SMEP model are in good agreement with the records in the literatures. The linear regression coefficients are all around 1.0, and the regression equations all pass the significant test of 0.01. SMEP model will be coupled to Chinese Agro-Meteorological Model (CAMM1.0), providing scientific and technological supports for the continuous improvement of CAMM1.0.
Construction of Forecasting Model of Meteorological Suitability for Wheat Aphids in the Northern China
Wang Chunzhi, Huo Zhiguo, Zhang Lei, Guo Anhong, Huang Chong, Lu Minghong
2020, 31(3): 280-289. DOI: 10.11898/1001-7313.20200303
The forecasting and early warning technology of meteorological suitability of wheat aphids in the main growing areas can provide a scientific basis for disaster prevention and high yield. Based on data of the occurrence area of wheat aphids, winter wheat growth period and daily meteorological data at 601 observation stations from 1958 to 2015 in 8 main wheat production provinces of the northern China, relationships between surface meteorological factors and the occurrence area of wheat aphids for every province in North China and Huanghuai Area are fully analyzed using methods of correlation analysis, principal component analysis and stepwise regression analysis in various time-periods from last December to 10 June. Results indicate that the key meteorological factors which affect the occurrence area of wheat aphids in North China are average air temperature of last winter and in the first ten days of April, temperature-precipitation coefficients and the number of days with maximum air temperature(no less than 25℃) in March, sunshine hours in the third ten days in March, the number of days with daily maximum air temperature(no less than 28℃) in the third ten days of April, the number of heavy rain days(no less than 25 mm) in April, the number of days with relative air humidity between 40% and 80% in the first ten days in May. The key meteorological factors which affect the occurrence area of wheat aphids in Huanghuai Area are average air temperature of last winter and in March, precipitation in the third ten days of January, the number of days with relative air humidity (more than 80%) in the first ten days in March, temperature-precipitation coefficients in April, the number of rainless days in the third ten days in April. The meteorological suitability forecasting models of wheat aphids are established based on the normalized key meteorological factors in North China and Huanghuai Area. Hindcast validation results show that the forecasting accuracy for meteorological suitability models is 91.2%, 93.1% in North China and Huanghuai Area. The accuracy of extrapolation forecasting in 2016-2018 is higher than 75% in the former two areas respectively. The average accuracy of extrapolation forecasting from 2016 to 2018 is 100% in Anhui and Jiangsu using the meteorological suitability forecasting model in Huanghuai Area. Models can be put into operational application in Huang-Huai-Hai region of China.
Extremity Analysis on the Precipitation and Environmental Field of Typhoon Rumbia in 2018
Yang Shunan, Duan Yihong
2020, 31(3): 290-302. DOI: 10.11898/1001-7313.20200304
Using a variety of observational and analytical data, the evolution of heavy precipitation, features of extreme precipitation and physical characteristics of environmental field causing extreme precipitation of Typhoon Rumbia in 2018 are analyzed. Obvious extremity can be seen with the daily rainfall at many national observational stations breaking historic records. The rainstorm process of Typhoon Rumbia can be divided into three stages: Landing, moving further inland and turning, and extratropical transition due to cold air intrusion. In the second stage, the typhoon moves very slowly and the circulation of Typhoon Rumbia stays over Henan Province for very long time, which makes the second stage the strongest precipitation period. Influenced by Typhoon Rumbia, historical daily precipitation records of many national stations in eastern Henan, southwestern Shandong and northern Shandong are broken. The observed maximum hourly precipitation is 127.7 mm, and as many as 74 stations experience an hourly precipitation of more than 80 mm. Furthermore, short-term heavy rainfall feature is very obvious, for more than 14 hours, the hourly accumulated precipitation exceeds 20 mm. Influenced by both the high precipitation efficiency and the long duration, extreme precipitation happens. Due to the atmospheric circulation characteristics of Typhoon Rumbia, there is an abnormal low pressure circulation with the standardized anomaly smaller than -4 times climate standard deviation in middle and lower troposphere, which results in extreme low-level convergence. At the same time, extreme upper divergence, induced by the combined actions of both upper tropospheric jet and high-pressure edge, can be seen in upper level. Therefore, there are obvious extreme characteristics in dynamic conditions. Compared with top 30 precipitation days in recent 30 years, 200 hPa divergence and 850 hPa convergence of Typhoon Rumbia are either close to or far beyond the historical maximum. Substantial water vapor is brought and converged in the rainstorm area continuously through the water vapor transport belt on the east side of typhoon, resulting in an extreme water vapor environmental condition. Significant extremity can be seen in many moisture-related physical parameters, such as pseudo-equivalent potential temperature, atmospheric precipitable water and vapor flux divergence, compared with both climatic mean state and historical heavy precipitation days in recent 30 years, and the extreme vapor condition lasts for up to 30 hours.
Heavy Precipitation Forecasts Based on Multi-model Ensemble Members
Zhi Xiefei, Zhao Chen
2020, 31(3): 303-314. DOI: 10.11898/1001-7313.20200305
Based on the daily 24-168 h ensemble precipitation forecasts over China from 1 May to 31 August in 2016 from the global ensemble models of ECMWF, JMA, NCEP, CMA and UKMO extracted from the TIGGE archives, the frequency matching method is tested to calibrate the precipitation frequency of each ensemble member. Then results of multi-model ensemble forecasts before and after calibration, including Kalman filter(KF), multi-model super-ensemble (SUP) and bias-removed ensemble mean(BREM), are analyzed in order to improve the prediction of precipitation based on numerical weather forecast data. Results show that precipitation forecasts calibrated by the frequency matching method, which uses the moderate precipitation to correct light and heavy precipitation, can effectively improve the problem of the underestimation of heavy precipitation caused by ensemble mean forecast and improve the positive deviation of the ensemble forecasting system, so that the precipitation forecast category is closer to the observation. However, the frequency matching method can barely improve the prediction of precipitation area. Different from frequency matching method, multi-model ensemble forecasts can extract and consider features of each model, therefore the prediction of precipitation area is more accurate than each single model, but the result is not as good as the frequency matching method in terms of the prediction of precipitation category. Among different multi-model methods, because of the updated weights over time, the result of Kalman filter forecast is superior to SUP and BREM in terms of threat scores, root mean square error (RMSE) and anomaly correlation coefficient (ACC). Furthermore, combining advantages of the above two methods, the multi-model ensemble precipitation after calibration based on ensemble members is more effective in the prediction of heavy precipitation category and area, which is closer to the observation. Results improve the threat score (TS) of the precipitation in all forecast lead times, especially in the heavy precipitation with the TS of 24 h forecast reaching 0.26, indicating a lower false alarm rate and missing rate compared with single model. Results also improve ACC and RMSE of the heavy precipitation and this method produces the best results among all the other methods, especially in the coastal areas in the south of China. In terms of the prediction of precipitation area, results effectively optimize the area of heavy and light precipitation, making the multi-model ensemble precipitation after calibration best in predicting heavy precipitation processes.
Impact of Arctic Extreme Cyclones on Cold Spells in China During Early 2015
Zhang Lin, Lü Junmei, Junmei Ding
2020, 31(3): 315-327. DOI: 10.11898/1001-7313.20200306
Extreme cyclones entering the Arctic from the mid-high latitude North Atlantic can cause anomalous warming in the Arctic region, which is closely related to extreme weather in mid-high latitudes and is very harmful. However, there are few studies on the impact of extreme cyclones upon weather and climate in China. Physical processes and mechanisms of two extreme cyclones (C1 and C2) affecting cold spells in China during January and February of 2015 are explored, ERA-Interim reanalysis data and observations from China meteorological stations are used. Results show that extreme cyclones occur on the mid-high latitude North Atlantic and meanwhile the anomalous warming appears near the extreme cyclones center in the lower and upper atmosphere. When extreme cyclones move northward, circulations in mid-high latitudes are shown as the formation and maintenance of Ural Blocking and the break-up of the polar vortex. Then the mid-high latitude trough deepens and moves southward over China, and the cold air intrudes southward into China driven by the northerly flow behind the trough, which results in cold weather occurring in China. Furthermore, the mechanism of the apparent adjustment in the polar vortex and general circulation are explored. It shows that the anomalous warming accompanied by extreme cyclones promotes the development of the mid-high latitude troughs and ridges through the energy dispersion of anomalous Rossby waves. In addition, comparison analysis in terms of their occurrence and track shows that there are significant differences between C1 and C2. Compared with C2, C1 occurs in higher latitudes and has an eastward track. Because of these differences, anomalous Rossby waves of C1 are divided into two branches which disperse energy upstream along high latitude and middle latitude, but C2 only has one branch of the anomalous Rossby wave dispersing upstream along the mid-latitude westerlies. Due to the discrepancy of anomalous Rossby waves between C1 and C2, adjustments of circulation in two events are different. Affected by these factors, C1 corresponds to a wider range of cold weather, and under effects of a little trough the duration of 5 d is longer than C2's which only lasts for 3 d. Cold spells corresponding with C2 only affects Northeast China, but apparently its intensity is slightly stronger than C1's. These results indicate that extreme cyclones are one of the important causes of cold spells in China. It also should be pointed out that more extreme cyclone events will be analyzed to draw more definite conclusions in future and provide a new reference for forecasting cold spells in China.
Characteristics of the Waterspout in East Dongting Lake on 13 August 2017
Yang Wei, Fang Yang, Jiang Shuai, Yuan Quan, Lin Nan
2020, 31(3): 328-338. DOI: 10.11898/1001-7313.20200307

Based on conventional weather data, automatic weather station data, and the observation of Yueyang Doppler radar, a waterspout occurred in Bianshan waters of East Dongting Lake (Bianshan waterspout for short) on 13 August 2017 is analyzed. Results show that the cold and warm airflows converge in the East Dongting Lake area when the upper East Asian trough forces the cold air southward, and the subtropical high guides the southwestern warm moist flow northward. The quasi stationary front over the north central Hunan Province is northeast to southwest, forming an "S" curve, which is favorable for the convergence of frontal instability energy to the East Dongting Lake area. The special geographical environment is easy to trigger canyon effect, which often leads to increased wind speed and humidity. The strong divergence in front of the upper trough, the deep low-pressure shear from northeast to southwest in the middle and lower layers, strong cyclonic convergence in the boundary layer, and the special topography jointly form a strong convergent upwelling flow field. When three meso-gamma-scale low eddies on the ground move northward to Bianshan waters, influenced by combined effects of the above flow field and the front and back vortices, the second vortex strengthens rapidly and forms a waterspout. The meteorological factors such as wind speed, wind direction, air pressure and visibility recorded by the lighthouse automatic meteorological station in the lake center change significantly when the waterspout passes, while precipitation is only 0.2 mm. Yueyang Doppler radar shows that the centroid of heavy precipitation is low to the north of strong convergence zone, where shear of strong wind is moderate and the radial wind speed over the shear is low. Yueyang Doppler radar wind profiles show that mesocyclone at the height of 0.6 km and the convergent flow fields near the ground at the height of 0.3 km are superimposed when the waterspout formed at 0905 BT. Waterspouts in the southern convergence zone have no storm tracking information, mesocyclones or tornado-type vortices. However, heavy precipitation accompanied by strong subsidence and convergence at the middle and low altitudes often produce both rising and subsidence currents, which are obviously unfavorable for the formation and development of waterspouts that need huge upward pumping. Comparing and analyzing waterspout processes of the Shengjin Lake in Anhui Province and the Dongting Lake in Hunan Province, it is concluded that the funnel-shaped strong lift suction caused by large-scale divergence at high altitude and the deep low-pressure shear from northeast to southwest in the middle and lower layers, and the intense convergence of cyclones and surface cyclones in the boundary layer are the main causes of the waterspout formation.

Ground Clutter Detection Algorithm for Array Weather Radar at Changsha Airport
Wei Wanyi, Ma Shuqing, Yang Ling, Zhen Xiaoqiong, Lü Siwei
2020, 31(3): 339-349. DOI: 10.11898/1001-7313.20200308

In order to obtain more detailed small-scale weather system data, Meteorological Observation Center of China Meteorological Administration (CMA) designed and developed X-band array weather radar (AWR), cooperating with relevant manufacturers. In March of 2018, the first prototypeis deployed at Changsha Airport for field experiments. Combing advantages of networked radars and a distributed phased array technology, the AWR has a highly coordinated scanning mode and high spatial and temporal resolutions to acquire fine echo intensity and wind field data. Compared with conventional parabolic antenna weather radars, a phased array antenna has wider beams and stronger side lobes, so that more ground clutter will appear in radar echoes. If the ground clutter cannot be effectively detected and removed, the accuracy of radar products will be affected seriously.Data collected by the X-band AWR at Changsha Airport are used to study the ground clutter detection algorithm for the AWR. According to the research progress all over the world, characteristic parameters of reflectivity factors, radial velocity and velocity spectrum width are extracted. In addition, time variability of reflectivity factor (TVR), a new parameter, is added due to the high temporal and spatial resolution of the AWR. Based on analyzing statistical characteristics of each feature parameter, membership functions are determined. The contribution of TVR to the clutter detection algorithm and the performance of the algorithm on different weather conditions are analyzed. Results show that the accuracy of ground clutter detection for Changsha Airport AWR can be maximally increased by 4% by adding TVR, and the error rate of detecting the precipitation echo as clutter echo can be decreased by 2%. The accuracy of the proposed ground clutter detection algorithm reaches 96% in the detection of ground clutter when no precipitation processes happen. In the precipitation weather, the accuracy is 92%, and the error rate of detecting the precipitation echo as clutter echo is about 10%. The algorithm can basically detect and remove the ground clutter echoes from precipitation echoes.

Parameter Improvements of Hydrometeor Classification Algorithm for the Dual-polarimetric Radar
Xu Shuyang, Wu Chong, Liu Liping
2020, 31(3): 350-360. DOI: 10.11898/1001-7313.20200309

Most of recent hydrometeor classification schemes for dual-polarimtric radar are based on fuzzy logic. Due to the lack of true value of hydrometeors, it is difficult to verify whether classifications are right or not. Therefore, three methods are proposed, cumulative value test, algorithm sensitivity test and hydrometeors' distribution test. Cumulative value test is used to inspect the ability of classifying and distinguishing. When input data contains system deviation and noise, the output classification would also contain system deviation and noise. Hence, a method is proposed to test the sensitivity of system deviation and noise. Hydrometeor distribution test is to analyze whether the hydrometeor distribution is temporal and spatial continuous. Through these tests the reliability and stability of the algorithm are analyzed, and key factors which affect the classification can be found out. Using observations of precipitation processes in Guangzhou in spring and summer from 2016 to 2017, the efficiency of S-band dual-polarization doppler radar is examined. Main results show that the classifying hydrometeor relies on the membership function. Using cumulative value test, some hydrometeors are found out with inappropriate membership function. These membership functions are not consistent with real characteristics of hydrometeors, which are ground clutter (including that due to anomalous propagation), biological scatters, dry aggregated snow, crystals of carious orientations, light and moderate rain, and a mixture of rain and hail. The way to modify these membership functions is based on hydrometeor statistic characteristics. Another insufficiency of membership function is to distinguish similar hydrometeors, such as crystals of various orientations with dry aggregated snow and heavy rain with mixture of rain and hail. The method to modify membership function's distinguishing ability is increasing parametric weights which has stronger discriminating ability. To ensure every hydrometeor has more than 90% stable results, the error of Zh is between -0.5 dB and +0.5 dB, that of ZDR is between -0.1 dB and +0.1 dB, that of ρhv is less than 0.02, and that of KDP is between -0.3 dB and +0.9 dB. Moreover, data quality of Zh and ZDR is more important than other parameters, system deviation is more influencing than noise. After hydrometeor distribution test, it is found that heavy rain may be misclassified as mixture of rain and hail. The spatial consistency correction method is to check whether the distribution of rain and hail mixture in a certain space is continuous.

The Effect Assessment of Wind Field Inversion Based on WPR in Precipitation
Lin Xiaomeng, Wei Yinghua, Chen Hong, Wang Yanchun
2020, 31(3): 361-372. DOI: 10.11898/1001-7313.20200310

Wind profile radar(WPR), taking atmospheric turbulence of clear air as main detecting object, is the main reference tool currently for short-time forecast because of its high spatial and temporal resolution. In the past few decades, WPR spectral data processing mainly focuses on the wind spectrum estimation. In recent years, with the use of WPR data expansion, there are increasingly high demand for WPR data accuracy, but because of ground clutter and external noise, flying objects, the presence of disturbances such as precipitation and limitations of Fourier Transform method itself, there are often multiple overlapped peaks, which makes it difficult to judge the spectral meaning, resulting in large error detection products. WPR has a large dynamic reception range, so it can receive the echo of scattering of atmosphere turbulence and precipitation particles simultaneously. However, the superimposed spectrum of atmosphere and precipitation cannot be separated effectively. In the meantime, the wind field calculation is based on the hypothesis of local uniform and isotropy, which cannot be met during precipitation with great spatial variability and leads to data of WPR serious deficiency or distortion. It's of great importance to establish an effective spectral extraction programs under different weather conditions to improve the accuracy of spectral estimation for wind field data after the inversion, thereby enhancing the wind profile accuracy of radar detection.A method of WPR-HW is developed for the case of precipitation according to the principle of WPR detection and the feature of spectrum, and then the effectiveness of the method is tested using ECMWF ERA Interim data. 10 precipitation cases in Tianjin are investigated to verify the significance of wind field data processed by WPR-HW in severe convection prediction. Results show that the WPR-HW has significant advantage compared with the recent WIND method (the universal method of wind field inversion from WPR) in integrity and reliability. For the wind field data in 10 precipitation cases, the leakage rate of WIND is 25.4% while that of WPR-HW is 0. The root mean square error in wind speed of WPR-HW is 1.6 m·s-1 while that of WIND is 2.3 m·s-1. The RMSE in wind direction of WPR-HW is 22° while that of WIND is 45°. The wind field processed by WPR-HW is able to make up for the deficiency and distortion of WPR data effectively in precipitation, which thus benefit to improve the timeliness and accuracy in strong convective weather forecasting.

Retrieval of Aerosol Optical Properties by Skyradiometer over Urban Beijing
Yang Xianyi, Che Huizheng, Chen Quanliang, Liang Yuanxin
2020, 31(3): 373-384. DOI: 10.11898/1001-7313.20200311

Aerosol particles can scatter and absorb solar radiation and affect microphysical processes of clouds to change the earth's radiation budget. It is reported that aerosol particles not only have an impact on climate change, but also cause polluted environment and affect human health. Ground-based measurement networks such as AERONET and SKYNET are very useful and accurate ways to monitor the spatio-temporal distribution of aerosols using the sun-sky radiometric technique. Aerosol optical properties retrieved by a PREDE skyradiometer are used to analyze the variation of aerosol in Beijing from October 2018 to September 2019. Results show that aerosol optical depth at 500 nm is high from February to July, the highest value is 0.71 in June, the highest single scattering albedo is 0.96 in August and the lowest value is 0.89 in May, Ångström exponent in summer (1.11) is higher than that in spring (0.89), and the volume size distribution pattern shows typical bimodal in every month. According to the Chinese National Secondary Standards for PM2.5, pollution days are picked. It is found that pollution days only account for 17%, of which 62% are light pollution days. The statistical result of air quality in Beijing is good from October 2018 to September 2019. Aerosol optical properties and PM2.5 under pollution and clean weather conditions in Beijing are discussed. The value of PM2.5 under pollution weather condition is 2.27 times larger than that under clean weather condition, values of aerosol optical depth at 500 nm are 0.85 and 0.49 under pollution and clean weather conditions, respectively. Values of single scattering albedo are 0.96 and 0.92 under pollution and clean weather conditions, respectively. The value of Ångström exponent under pollution weather condition (1.02) is larger than that under clean weather condition (0.91) in winter while the value of Ångström exponent under pollution weather condition (0.87) is smaller than that under clean weather condition (0.90) in spring. Skyradiometer retrieved data, combined with lidar measurement and meteorological data are used to analyze a serious pollution event in winter over Beijing. The result suggests that poor meteorological conditions (low wind speed and high relative humidity), the hygroscopic growth of aerosol, aerosol secondary transformation, local emissions and regional transportation lead to this serious pollution event.