Convective storms and lightning are among the most important weather phenomena that are challenging to forecast.In this study,a novel multi-task learning(MTL)encoder-decoder U-net neural network was developed to forec...Convective storms and lightning are among the most important weather phenomena that are challenging to forecast.In this study,a novel multi-task learning(MTL)encoder-decoder U-net neural network was developed to forecast convective storms and lightning with lead times for up to 90 min,using GOES-16 geostationary satellite infrared brightness temperatures(IRBTs),lightning flashes from Geostationary Lightning Mapper(GLM),and vertically integrated liquid(VIL)from Next Generation Weather Radar(NEXRAD).To cope with the heavily skewed distribution of lightning data,a spatiotemporal exponent-weighted loss function and log-transformed lightning normalization approach were developed.The effects of MTL,single-task learning(STL),and IRBTs as auxiliary input features on convection and lightning nowcasting were investigated.The results showed that normalizing the heavily skew-distributed lightning data along with a log-transformation dramatically outperforms the min-max normalization method for nowcasting an intense lightning event.The MTL model significantly outperformed the STL model for both lightning nowcasting and VIL nowcasting,particularly for intense lightning events.The MTL also helped delay the lightning forecast performance decay with the lead times.Furthermore,incorporating satellite IRBTs as auxiliary input features substantially improved lightning nowcasting,but produced little difference in VIL forecasting.Finally,the MTL model performed better for forecasting both lightning and the VIL of organized convective storms than for isolated cells.展开更多
Monitoring and predicting highly localized weather events over a very short-term period,typically ranging from minutes to a few hours,are very important for decision makers and public action.Nowcasting these events us...Monitoring and predicting highly localized weather events over a very short-term period,typically ranging from minutes to a few hours,are very important for decision makers and public action.Nowcasting these events usually relies on radar observations through monitoring and extrapolation.With advanced high-resolution imaging and sounding observations from weather satellites,nowcasting can be enhanced by combining radar,satellite,and other data,while quantitative applications of those data for nowcasting are advanced through using machine learning techniques.Those applications include monitoring the location,impact area,intensity,water vapor,atmospheric instability,precipitation,physical properties,and optical properties of the severe storm at different stages(pre-convection,initiation,development,and decaying),identification of storm types(wind,snow,hail,etc.),and predicting the occurrence and evolution of the storm.Satellite observations can provide information on the environmental characteristics in the preconvection stage and are very useful for situational awareness and storm warning.This paper provides an overview of recent progress on quantitative applications of satellite data in nowcasting and its challenges,and future perspectives are also addressed and discussed.展开更多
Sudden precipitations may bring troubles or even huge harm to people’s daily lives.Hence a timely and accurate precipitation nowcasting is expected to be an indispensable part of our modern life.Traditionally,the rai...Sudden precipitations may bring troubles or even huge harm to people’s daily lives.Hence a timely and accurate precipitation nowcasting is expected to be an indispensable part of our modern life.Traditionally,the rainfall intensity estimation from weather radar is based on the relationship between radar reflectivity factor(Z)and rainfall rate(R),which is typically estimated by location-dependent experiential formula and arguably uncertain.Therefore,in this paper,we propose a deep learning-based method to model the ZR relation.To evaluate,we conducted our experiment with the Shenzhen precipitation dataset.We proposed a combined method of deep learning and the ZR relationship,and compared it with a traditional ZR equation,a ZR equation with its parameters estimated by the least square method,and a pure deep learning model.The experimental results show that our combined model performsmuch better than the equation-based ZRformula and has the similar performance with a pure deep learning nowcasting model,both for all level precipitation and heavy ones only.展开更多
The Boreal forest is a terrestrial ecosystem highly vulnerable to the impacts of short-term climate and weather variabilities. Detecting abrupt, rapid climate-induced changes in fire weather and related changes in fir...The Boreal forest is a terrestrial ecosystem highly vulnerable to the impacts of short-term climate and weather variabilities. Detecting abrupt, rapid climate-induced changes in fire weather and related changes in fire seasonality can provide important insights to assessing impacts of climate change on forestry. This paper, taking the Sakha Republic of Russia as study area, aims to suggest an approach for detecting signals indicating climate-induced changes in fire weather to express recent fire weather variability by using short-term ranks of major meteorological parameters such as air temperature and atmospheric precipitation. Climate data from the “Global Summary of the Day Product” of NOAA (the United States National Oceanic and Atmospheric Administration) for 1996 to 2018 were used to investigate meteorological parameters that drive fire activity. The detection of the climate change signals is made through a 4-step analysis. First, we used descriptive statistics to grasp monthly, annual, seasonal and peak fire period characteristics of fire weather. Then we computed historical normals for WMO reference period, 1961-1990, and the most recent 30-year period for comparison with the current means. The variability of fire weather is analyzed using standard deviation, coefficient of variation, percentage departures from historical normals, percentage departures from the mean, and precipitation concentration index. Inconsistency and abrupt changes in the evolution of fire weather are assessed using homogeneity analysis whilst a Mann-Kendall test is used to detect significant trends in the time series. The results indicate a significant increase of temperature during spring and fall months, which extends the fire season and potentially contributes to increase of burned areas. We again detected a significant rainfall shortage in September which extended the fire season. Furthermore, this study suggests a new approach in statistical methods appropriate for the detection of climate change signals on fire weather variability using short-term climate ranks and evaluation of its impact on fire seasonality and activity.展开更多
Strong convective systems and the associated heavy rainfall events can trig-ger floods and landslides with severe detrimental consequences.These events have a high spatio-temporal variability,being difficult to predic...Strong convective systems and the associated heavy rainfall events can trig-ger floods and landslides with severe detrimental consequences.These events have a high spatio-temporal variability,being difficult to predict by standard meteorological numerical models.This work proposes the M5Images method for performing the very short-term prediction(nowcasting)of heavy convective rainfall using weather radar data by means of a convolutional recurrent neural network.The recurrent part of it is a Long Short-Term Memory(LSTM)neural network.Prediction tests were performed for the city and surroundings of Campinas,located in the Southeastern Brazil.The convolutional recurrent neural network was trained using time series of rainfall rate images derived from weather radar data for a selected set of heavy rainfall events.The attained pre-diction performance was better than that given by the persistence forecasting method for different prediction times.展开更多
The sparse and uneven placement of rain gauges across the Tibetan Plateau(TP) impedes the acquisition of precise,high-resolution precipitation measurements,thus challenging the reliability of forecast data.To address ...The sparse and uneven placement of rain gauges across the Tibetan Plateau(TP) impedes the acquisition of precise,high-resolution precipitation measurements,thus challenging the reliability of forecast data.To address such a challenge,we introduce a model called Multisource Generative Adversarial Network-Convolutional Long Short-Term Memory(GAN-ConvLSTM) for Precipitation Nowcasting(MGCPN),which utilizes data products from the Integrated Multi-satellite Retrievals for global precipitation measurement(IMERG) data,offering high spatiotemporal resolution precipitation forecasts for upcoming periods ranging from 30 to 300 min.The results of our study confirm that the implementation of the MGCPN model successfully addresses the problem of underestimating and blurring precipitation results that often arise with increasing forecast time.This issue is a common challenge in precipitation forecasting models.Furthermore,we have used multisource spatiotemporal datasets with integrated geographic elements for training and prediction to improve model accuracy.The model demonstrates its competence in generating precise precipitation nowcasting with IMERG data,offering valuable support for precipitation research and forecasting in the TP region.The metrics results obtained from our study further emphasize the notable advantages of the MGCPN model;it outperforms the other considered models in the probability of detection(POD),critical success index,Heidke Skill Score,and mean absolute error,especially showing improvements in POD by approximately 33%,19%,and 8% compared to Convolutional Gated Recurrent Unit(ConvGRU),ConvLSTM,and small Attention-UNet(SmaAt-UNet) models.展开更多
[Objective] This study aimed to analyze the cause of the generation of short-term heavy precipitations in a regional heavy rainstorm in Shannxi Province. [Method] Taking a heavy rainstorm covering most parts of Shaanx...[Objective] This study aimed to analyze the cause of the generation of short-term heavy precipitations in a regional heavy rainstorm in Shannxi Province. [Method] Taking a heavy rainstorm covering most parts of Shaanxi Province in late July 2010 as an example, data of five Doppler weather radars in Shaanxi Province were employed for a detailed analysis of the evolution of the heavy rainstorm pro- cess. [Result] Besides the good large-scale weather background conditions, the de- velopment and evolution of some mesoscale and small-scale weather systems direct- ly led to short-term heavy precipitations during the heavy rainstorm process, involv- ing the intrusion of moderate IS-scale weak cold air and presence of small-scale wind shear, convergence and adverse wind area. In addition, small-scale convection echoes were arranged in lines and formed a ＂train effect＂, which would also con- tribute to the generation of short-term heavy precipitation. [Conclusion] This study provided basic information for more clear and in-depth analysis of the formation mechanism of short-term heavy precipitations.展开更多
This paper describes the procedure and methodology to formulate the convective weather potential （CWP） algorithm. The data used in the development of the algorithm are the radar echoes at 0.5° elevation from Gu...This paper describes the procedure and methodology to formulate the convective weather potential （CWP） algorithm. The data used in the development of the algorithm are the radar echoes at 0.5° elevation from Guangzhou Doppler Radar Station, surface observations from automatic weather stations （AWS） and outputs of numeric weather prediction （NWP） models. The procedure to develop the CWP algorithm consists of two steps： （1） identification of thunderstorm cells in accordance with specified statistical criteria; and （2） development of the algorithm based on multiple linear regression. The thunderstorm cells were automatically identified by radar echoes with intensity greater than or equal to 50 dB（Z） and of an area over 64 square kilometers. These cells are generally related to severe convective weather occurrences such as thunderstorm wind gusts, hail and tornados. In the development of the CWP algorithm, both echo- and environment-based predictors are used. The predictand is the probability of a thunderstorm cell to generate severe convective weather events. The predictor-predictand relationship is established through a stepwise multiple linear regression approach. Verification with an independent dataset shows that the CWP algorithm is skillful in detecting thunderstorm-related severe convective weather occurrences in the Pearl River Delta （PRD） region of South China. An example of a nowcasting case for a thunderstorm process is illustrated.展开更多
Based on conventional meteorological observation data and Doppler radar data,the occurrence and development mechanism of mixed severe convective weather and evolution of convective storm in Guangxi on March 4,2018 wer...Based on conventional meteorological observation data and Doppler radar data,the occurrence and development mechanism of mixed severe convective weather and evolution of convective storm in Guangxi on March 4,2018 were analyzed. The results showed that the dry line was the main trigger mechanism of this severe convective weather. Instable convection stratification of cold advection at middle layer and warm advection at low layer and abundant water vapor from low-level jet provided favorable stratification and water vapor conditions for the occurrence and development of severe convection. Cold trough at middle layer,low pressure and strong vertical wind shear at middle and lower layers may be main factors for the development and maintenance of strong storm system. Squall line developed along ground convergence line,and there was bow echo on reflectivity factor chart. Moving velocity of convective system was quick,and there was gale core and velocity ambiguity on velocity map.展开更多
针对业务运行中雷达观测存在遮挡和雷达产品延迟,提出利用带噪声基于密度的空间聚类(density-based spatial clustering of applications with noise,DBSC AN)算法对闪电数据的聚类结果替代雷达产品,并分别利用北京三维闪电定位网(Beiji...针对业务运行中雷达观测存在遮挡和雷达产品延迟,提出利用带噪声基于密度的空间聚类(density-based spatial clustering of applications with noise,DBSC AN)算法对闪电数据的聚类结果替代雷达产品,并分别利用北京三维闪电定位网(Beijing Total Lightning System,BJTLS)和升级后的国家闪电定位网(DDW1)总闪数据,应用2σ闪电跃增算法对北京2022年6月4日和12日两次强对流致灾过程进行临近预警,对比强对流单体识别法和DBSCAN聚类法的预警效果。结果表明:两种算法和两种闪电数据均能有效预警北京地区的灾害性天气,基于BJTLS总闪数据的预警效果较优;对于BJTLS总闪数据,两种方法的预警效果相当,预警命中率、误报率、临近成功指数和平均预警提前时间依次分别为100%,11.9%,88.1%,38.9 min和100%,13.3%,86.7%,42.8 min;仅利用闪电数据并应用2σ闪电跃增算法可对灾害性天气进行临近预警,摆脱对雷达产品的依赖。展开更多
基金supported by the Science and Technology Grant No.520120210003,Jibei Electric Power Company of the State Grid Corporation of China。
文摘Convective storms and lightning are among the most important weather phenomena that are challenging to forecast.In this study,a novel multi-task learning(MTL)encoder-decoder U-net neural network was developed to forecast convective storms and lightning with lead times for up to 90 min,using GOES-16 geostationary satellite infrared brightness temperatures(IRBTs),lightning flashes from Geostationary Lightning Mapper(GLM),and vertically integrated liquid(VIL)from Next Generation Weather Radar(NEXRAD).To cope with the heavily skewed distribution of lightning data,a spatiotemporal exponent-weighted loss function and log-transformed lightning normalization approach were developed.The effects of MTL,single-task learning(STL),and IRBTs as auxiliary input features on convection and lightning nowcasting were investigated.The results showed that normalizing the heavily skew-distributed lightning data along with a log-transformation dramatically outperforms the min-max normalization method for nowcasting an intense lightning event.The MTL model significantly outperformed the STL model for both lightning nowcasting and VIL nowcasting,particularly for intense lightning events.The MTL also helped delay the lightning forecast performance decay with the lead times.Furthermore,incorporating satellite IRBTs as auxiliary input features substantially improved lightning nowcasting,but produced little difference in VIL forecasting.Finally,the MTL model performed better for forecasting both lightning and the VIL of organized convective storms than for isolated cells.
基金Supported by the National Natural Science Foundation of China(U2142201 and 42175086).
文摘Monitoring and predicting highly localized weather events over a very short-term period,typically ranging from minutes to a few hours,are very important for decision makers and public action.Nowcasting these events usually relies on radar observations through monitoring and extrapolation.With advanced high-resolution imaging and sounding observations from weather satellites,nowcasting can be enhanced by combining radar,satellite,and other data,while quantitative applications of those data for nowcasting are advanced through using machine learning techniques.Those applications include monitoring the location,impact area,intensity,water vapor,atmospheric instability,precipitation,physical properties,and optical properties of the severe storm at different stages(pre-convection,initiation,development,and decaying),identification of storm types(wind,snow,hail,etc.),and predicting the occurrence and evolution of the storm.Satellite observations can provide information on the environmental characteristics in the preconvection stage and are very useful for situational awareness and storm warning.This paper provides an overview of recent progress on quantitative applications of satellite data in nowcasting and its challenges,and future perspectives are also addressed and discussed.
基金supported by Sichuan Provincial Key Research and Development Program(No.2021YFG0345,to J.Ma)the National Key Research and Development Program of China(No.2020YFA0608001,to J.Ma).
文摘Sudden precipitations may bring troubles or even huge harm to people’s daily lives.Hence a timely and accurate precipitation nowcasting is expected to be an indispensable part of our modern life.Traditionally,the rainfall intensity estimation from weather radar is based on the relationship between radar reflectivity factor(Z)and rainfall rate(R),which is typically estimated by location-dependent experiential formula and arguably uncertain.Therefore,in this paper,we propose a deep learning-based method to model the ZR relation.To evaluate,we conducted our experiment with the Shenzhen precipitation dataset.We proposed a combined method of deep learning and the ZR relationship,and compared it with a traditional ZR equation,a ZR equation with its parameters estimated by the least square method,and a pure deep learning model.The experimental results show that our combined model performsmuch better than the equation-based ZRformula and has the similar performance with a pure deep learning nowcasting model,both for all level precipitation and heavy ones only.
文摘The Boreal forest is a terrestrial ecosystem highly vulnerable to the impacts of short-term climate and weather variabilities. Detecting abrupt, rapid climate-induced changes in fire weather and related changes in fire seasonality can provide important insights to assessing impacts of climate change on forestry. This paper, taking the Sakha Republic of Russia as study area, aims to suggest an approach for detecting signals indicating climate-induced changes in fire weather to express recent fire weather variability by using short-term ranks of major meteorological parameters such as air temperature and atmospheric precipitation. Climate data from the “Global Summary of the Day Product” of NOAA (the United States National Oceanic and Atmospheric Administration) for 1996 to 2018 were used to investigate meteorological parameters that drive fire activity. The detection of the climate change signals is made through a 4-step analysis. First, we used descriptive statistics to grasp monthly, annual, seasonal and peak fire period characteristics of fire weather. Then we computed historical normals for WMO reference period, 1961-1990, and the most recent 30-year period for comparison with the current means. The variability of fire weather is analyzed using standard deviation, coefficient of variation, percentage departures from historical normals, percentage departures from the mean, and precipitation concentration index. Inconsistency and abrupt changes in the evolution of fire weather are assessed using homogeneity analysis whilst a Mann-Kendall test is used to detect significant trends in the time series. The results indicate a significant increase of temperature during spring and fall months, which extends the fire season and potentially contributes to increase of burned areas. We again detected a significant rainfall shortage in September which extended the fire season. Furthermore, this study suggests a new approach in statistical methods appropriate for the detection of climate change signals on fire weather variability using short-term climate ranks and evaluation of its impact on fire seasonality and activity.
文摘Strong convective systems and the associated heavy rainfall events can trig-ger floods and landslides with severe detrimental consequences.These events have a high spatio-temporal variability,being difficult to predict by standard meteorological numerical models.This work proposes the M5Images method for performing the very short-term prediction(nowcasting)of heavy convective rainfall using weather radar data by means of a convolutional recurrent neural network.The recurrent part of it is a Long Short-Term Memory(LSTM)neural network.Prediction tests were performed for the city and surroundings of Campinas,located in the Southeastern Brazil.The convolutional recurrent neural network was trained using time series of rainfall rate images derived from weather radar data for a selected set of heavy rainfall events.The attained pre-diction performance was better than that given by the persistence forecasting method for different prediction times.
基金Supported by the National Natural Science Foundation of China (41871285 and 52104158)。
文摘The sparse and uneven placement of rain gauges across the Tibetan Plateau(TP) impedes the acquisition of precise,high-resolution precipitation measurements,thus challenging the reliability of forecast data.To address such a challenge,we introduce a model called Multisource Generative Adversarial Network-Convolutional Long Short-Term Memory(GAN-ConvLSTM) for Precipitation Nowcasting(MGCPN),which utilizes data products from the Integrated Multi-satellite Retrievals for global precipitation measurement(IMERG) data,offering high spatiotemporal resolution precipitation forecasts for upcoming periods ranging from 30 to 300 min.The results of our study confirm that the implementation of the MGCPN model successfully addresses the problem of underestimating and blurring precipitation results that often arise with increasing forecast time.This issue is a common challenge in precipitation forecasting models.Furthermore,we have used multisource spatiotemporal datasets with integrated geographic elements for training and prediction to improve model accuracy.The model demonstrates its competence in generating precise precipitation nowcasting with IMERG data,offering valuable support for precipitation research and forecasting in the TP region.The metrics results obtained from our study further emphasize the notable advantages of the MGCPN model;it outperforms the other considered models in the probability of detection(POD),critical success index,Heidke Skill Score,and mean absolute error,especially showing improvements in POD by approximately 33%,19%,and 8% compared to Convolutional Gated Recurrent Unit(ConvGRU),ConvLSTM,and small Attention-UNet(SmaAt-UNet) models.
基金Supported by Special Fund for National Weather Service Forecaster of China (CMAYBY2011-050)~~
文摘[Objective] This study aimed to analyze the cause of the generation of short-term heavy precipitations in a regional heavy rainstorm in Shannxi Province. [Method] Taking a heavy rainstorm covering most parts of Shaanxi Province in late July 2010 as an example, data of five Doppler weather radars in Shaanxi Province were employed for a detailed analysis of the evolution of the heavy rainstorm pro- cess. [Result] Besides the good large-scale weather background conditions, the de- velopment and evolution of some mesoscale and small-scale weather systems direct- ly led to short-term heavy precipitations during the heavy rainstorm process, involv- ing the intrusion of moderate IS-scale weak cold air and presence of small-scale wind shear, convergence and adverse wind area. In addition, small-scale convection echoes were arranged in lines and formed a ＂train effect＂, which would also con- tribute to the generation of short-term heavy precipitation. [Conclusion] This study provided basic information for more clear and in-depth analysis of the formation mechanism of short-term heavy precipitations.
文摘This paper describes the procedure and methodology to formulate the convective weather potential （CWP） algorithm. The data used in the development of the algorithm are the radar echoes at 0.5° elevation from Guangzhou Doppler Radar Station, surface observations from automatic weather stations （AWS） and outputs of numeric weather prediction （NWP） models. The procedure to develop the CWP algorithm consists of two steps： （1） identification of thunderstorm cells in accordance with specified statistical criteria; and （2） development of the algorithm based on multiple linear regression. The thunderstorm cells were automatically identified by radar echoes with intensity greater than or equal to 50 dB（Z） and of an area over 64 square kilometers. These cells are generally related to severe convective weather occurrences such as thunderstorm wind gusts, hail and tornados. In the development of the CWP algorithm, both echo- and environment-based predictors are used. The predictand is the probability of a thunderstorm cell to generate severe convective weather events. The predictor-predictand relationship is established through a stepwise multiple linear regression approach. Verification with an independent dataset shows that the CWP algorithm is skillful in detecting thunderstorm-related severe convective weather occurrences in the Pearl River Delta （PRD） region of South China. An example of a nowcasting case for a thunderstorm process is illustrated.
基金Supported by Special Project for Forecasters of China Meteorological Administration(CMAYBY2020-096)Meteorological Scientific Research Plan Project of Guangxi Meteorological Bureau(GUIQIKE2017Z06)。
文摘Based on conventional meteorological observation data and Doppler radar data,the occurrence and development mechanism of mixed severe convective weather and evolution of convective storm in Guangxi on March 4,2018 were analyzed. The results showed that the dry line was the main trigger mechanism of this severe convective weather. Instable convection stratification of cold advection at middle layer and warm advection at low layer and abundant water vapor from low-level jet provided favorable stratification and water vapor conditions for the occurrence and development of severe convection. Cold trough at middle layer,low pressure and strong vertical wind shear at middle and lower layers may be main factors for the development and maintenance of strong storm system. Squall line developed along ground convergence line,and there was bow echo on reflectivity factor chart. Moving velocity of convective system was quick,and there was gale core and velocity ambiguity on velocity map.
文摘针对业务运行中雷达观测存在遮挡和雷达产品延迟,提出利用带噪声基于密度的空间聚类(density-based spatial clustering of applications with noise,DBSC AN)算法对闪电数据的聚类结果替代雷达产品,并分别利用北京三维闪电定位网(Beijing Total Lightning System,BJTLS)和升级后的国家闪电定位网(DDW1)总闪数据,应用2σ闪电跃增算法对北京2022年6月4日和12日两次强对流致灾过程进行临近预警,对比强对流单体识别法和DBSCAN聚类法的预警效果。结果表明:两种算法和两种闪电数据均能有效预警北京地区的灾害性天气,基于BJTLS总闪数据的预警效果较优;对于BJTLS总闪数据,两种方法的预警效果相当,预警命中率、误报率、临近成功指数和平均预警提前时间依次分别为100%,11.9%,88.1%,38.9 min和100%,13.3%,86.7%,42.8 min;仅利用闪电数据并应用2σ闪电跃增算法可对灾害性天气进行临近预警,摆脱对雷达产品的依赖。
