Browse
Abstract
To mitigate the impacts associated with adverse weather conditions, meteorological services issue weather warnings to the general public. These warnings rely heavily on forecasts issued by underlying prediction systems. When deciding which prediction system(s) to utilize when constructing warnings, it is important to compare systems in their ability to forecast the occurrence and severity of high-impact weather events. However, evaluating forecasts for particular outcomes is known to be a challenging task. This is exacerbated further by the fact that high-impact weather often manifests as a result of several confounding features, a realization that has led to considerable research on so-called compound weather events. Both univariate and multivariate methods are therefore required to evaluate forecasts for high-impact weather. In this paper, we discuss weighted verification tools, which allow particular outcomes to be emphasized during forecast evaluation. We review and compare different approaches to construct weighted scoring rules, both in a univariate and multivariate setting, and we leverage existing results on weighted scores to introduce conditional probability integral transform (PIT) histograms, allowing forecast calibration to be assessed conditionally on particular outcomes having occurred. To illustrate the practical benefit afforded by these weighted verification tools, they are employed in a case study to evaluate probabilistic forecasts for extreme heat events issued by the Swiss Federal Office of Meteorology and Climatology (MeteoSwiss).
Abstract
To mitigate the impacts associated with adverse weather conditions, meteorological services issue weather warnings to the general public. These warnings rely heavily on forecasts issued by underlying prediction systems. When deciding which prediction system(s) to utilize when constructing warnings, it is important to compare systems in their ability to forecast the occurrence and severity of high-impact weather events. However, evaluating forecasts for particular outcomes is known to be a challenging task. This is exacerbated further by the fact that high-impact weather often manifests as a result of several confounding features, a realization that has led to considerable research on so-called compound weather events. Both univariate and multivariate methods are therefore required to evaluate forecasts for high-impact weather. In this paper, we discuss weighted verification tools, which allow particular outcomes to be emphasized during forecast evaluation. We review and compare different approaches to construct weighted scoring rules, both in a univariate and multivariate setting, and we leverage existing results on weighted scores to introduce conditional probability integral transform (PIT) histograms, allowing forecast calibration to be assessed conditionally on particular outcomes having occurred. To illustrate the practical benefit afforded by these weighted verification tools, they are employed in a case study to evaluate probabilistic forecasts for extreme heat events issued by the Swiss Federal Office of Meteorology and Climatology (MeteoSwiss).
Abstract
This study analyzes the potential of deep learning using probabilistic artificial neural networks (ANNs) for postprocessing ensemble precipitation forecasts at four observation locations. We split the precipitation forecast problem into two tasks: estimating the probability of precipitation and predicting the hourly precipitation. We then compare the performance with classical statistical postprocessing (logistical regression and GLM). ANNs show a higher performance at three of the four stations for estimating the probability of precipitation and at all stations for predicting the hourly precipitation. Further, two more general ANN models are trained using the merged data from all four stations. These general ANNs exhibit an increase in performance compared to the station-specific ANNs at most stations. However, they show a significant decay in performance at one of the stations at estimating the hourly precipitation. The general models seem capable of learning meaningful interactions in the data and generalizing these to improve the performance at other sites, which also causes the loss of local information at one station. Thus, this study indicates the potential of deep learning in weather forecasting workflows.
Abstract
This study analyzes the potential of deep learning using probabilistic artificial neural networks (ANNs) for postprocessing ensemble precipitation forecasts at four observation locations. We split the precipitation forecast problem into two tasks: estimating the probability of precipitation and predicting the hourly precipitation. We then compare the performance with classical statistical postprocessing (logistical regression and GLM). ANNs show a higher performance at three of the four stations for estimating the probability of precipitation and at all stations for predicting the hourly precipitation. Further, two more general ANN models are trained using the merged data from all four stations. These general ANNs exhibit an increase in performance compared to the station-specific ANNs at most stations. However, they show a significant decay in performance at one of the stations at estimating the hourly precipitation. The general models seem capable of learning meaningful interactions in the data and generalizing these to improve the performance at other sites, which also causes the loss of local information at one station. Thus, this study indicates the potential of deep learning in weather forecasting workflows.
Abstract
Realistic ocean initial conditions are essential for coupled hurricane forecasts. This study focuses on the impact of assimilating high-resolution ocean observations for initialization of the Modular Ocean Model (MOM6) in a coupled configuration with the Hurricane Analysis and Forecast System (HAFS). Based on the Joint Effort for Data Assimilation Integration (JEDI) framework, numerical experiments were performed for the Hurricane Isaias (2020) case, a Category One hurricane, with use of underwater glider data sets and satellite observations. Assimilation of ocean glider data together with satellite observations provides opportunity to further advance understanding of ocean conditions and air-sea interactions in coupled model initialization and Hurricane forecast systems. This comprehensive data assimilation approach has led to a more accurate prediction of the salinity-induced barrier layer thickness that suppresses vertical mixing and sea surface temperature cooling during the storm. Increased barrier layer thickness enhances ocean enthalpy flux into the lower atmosphere and potentially increases tropical cyclone intensity. Assimilation of satellite observations demonstrates improvement in Hurricane Isaias’ intensity forecast. Assimilating glider observations with broad spatial and temporal coverage along Isaias’ track in addition to satellite observations further increase Isaias’ intensity forecast. Overall this case study demonstrates the importance of assimilating comprehensive marine observations to a more robust ocean and hurricane forecast under a unified JEDI-HAFS hurricane forecast system.
Abstract
Realistic ocean initial conditions are essential for coupled hurricane forecasts. This study focuses on the impact of assimilating high-resolution ocean observations for initialization of the Modular Ocean Model (MOM6) in a coupled configuration with the Hurricane Analysis and Forecast System (HAFS). Based on the Joint Effort for Data Assimilation Integration (JEDI) framework, numerical experiments were performed for the Hurricane Isaias (2020) case, a Category One hurricane, with use of underwater glider data sets and satellite observations. Assimilation of ocean glider data together with satellite observations provides opportunity to further advance understanding of ocean conditions and air-sea interactions in coupled model initialization and Hurricane forecast systems. This comprehensive data assimilation approach has led to a more accurate prediction of the salinity-induced barrier layer thickness that suppresses vertical mixing and sea surface temperature cooling during the storm. Increased barrier layer thickness enhances ocean enthalpy flux into the lower atmosphere and potentially increases tropical cyclone intensity. Assimilation of satellite observations demonstrates improvement in Hurricane Isaias’ intensity forecast. Assimilating glider observations with broad spatial and temporal coverage along Isaias’ track in addition to satellite observations further increase Isaias’ intensity forecast. Overall this case study demonstrates the importance of assimilating comprehensive marine observations to a more robust ocean and hurricane forecast under a unified JEDI-HAFS hurricane forecast system.
Abstract
Flash flooding is the most damaging and deadly type of flooding event in the continental United States (CONUS), and one of the deadliest hazards worldwide. The Weather Prediction Center’s (WPC) Excessive Rainfall Outlook (ERO) is used to highlight regions at risk of receiving excessive rainfall that can lead to flash flooding. While EROs have been validated by the WPC across several metrics, an analysis of flash floods that were not forecast by EROs, which we define as missed flash floods, has not been performed, nor have damages associated with missed flash floods been examined. Using EROs, flash flood data from the Unified Flooding Verification System (UFVS), and flash flood damage data from the National Centers for Environmental Information (NCEI) Storm Events Database, this research investigates the characteristics of missed flash floods. We find that missed damages occur most frequently in summer and least frequently in winter, but that only 9.2% of flash flood damages and 23.2% of damaging events are missed by the ERO. Missed damaging flash flood events occur frequently in the Southwest US. In this region, missed damages and missed events are primarily attributed to the North American monsoon (NAM). We also investigate forecast flash floods by ERO risk category. High Risks incur the most damages despite having the fewest number of damaging events, indicating that the ERO is able to distinguish higher impact events. The ERO predicts damaging flash floods well, although the Southwest and urban areas broadly could be investigated further to ensure that the potential of damaging flash floods are accurately forecast.
Abstract
Flash flooding is the most damaging and deadly type of flooding event in the continental United States (CONUS), and one of the deadliest hazards worldwide. The Weather Prediction Center’s (WPC) Excessive Rainfall Outlook (ERO) is used to highlight regions at risk of receiving excessive rainfall that can lead to flash flooding. While EROs have been validated by the WPC across several metrics, an analysis of flash floods that were not forecast by EROs, which we define as missed flash floods, has not been performed, nor have damages associated with missed flash floods been examined. Using EROs, flash flood data from the Unified Flooding Verification System (UFVS), and flash flood damage data from the National Centers for Environmental Information (NCEI) Storm Events Database, this research investigates the characteristics of missed flash floods. We find that missed damages occur most frequently in summer and least frequently in winter, but that only 9.2% of flash flood damages and 23.2% of damaging events are missed by the ERO. Missed damaging flash flood events occur frequently in the Southwest US. In this region, missed damages and missed events are primarily attributed to the North American monsoon (NAM). We also investigate forecast flash floods by ERO risk category. High Risks incur the most damages despite having the fewest number of damaging events, indicating that the ERO is able to distinguish higher impact events. The ERO predicts damaging flash floods well, although the Southwest and urban areas broadly could be investigated further to ensure that the potential of damaging flash floods are accurately forecast.
Abstract
The extended-range forecast with a lead time of 10–30 days is the gap between weather (<10 days) and climate (>30 days) predictions. Improving the forecast skill of extreme weather events at the extended range is crucial for risk management of disastrous events. In this study, three deep learning (DL) models based on the methods of convolutional neural networks and gate recurrent units are constructed to predict the rainfall anomalies and associated extreme events in East China at lead times of 1–6 pentads. All DL models show skillful prediction of the temporal variation of rainfall anomalies (in terms of temporal correlation coefficient skill) over most regions in East China beyond 4 pentads, outperforming the dynamical models from the China Meteorological Administration (CMA) and the European Centre for Medium-Range Weather Forecasts (ECMWF). The spatial distribution of the rainfall anomalies is also better predicted by the DL models than the dynamical models; and the DL models show higher pattern correlation coefficients than the dynamical models at lead times of 3–6 pentads. The higher skill of DL models in predicting the rainfall anomalies will help to improve the accuracy of extreme-event predictions. The Heidke skill scores of the extreme rainfall event forecast performed by the DL models are also superior to those of the dynamical models at a lead time beyond about 4 pentads. Heat map analysis for the DL models shows that the predictability sources are mainly the large-scale factors modulating the East Asian monsoon rainfall.
Significance Statement
Improving the forecast skill for extreme weather events at the extended range (10–30 days in advance), particularly over populated regions such as East China, is crucial for risk management. This study aims to develop skillful models of the rainfall anomalies and associated extreme heavy rainfall events using deep learning techniques. The models constructed here benefit from the capability of deep learning to identify the predictability sources of rainfall variability, and outperform the current operational models, including the ECMWF and the CMA models, at forecast lead times beyond 3–4 pentads. These results reveal the promising application prospect of deep learning techniques in the extended-range forecast.
Abstract
The extended-range forecast with a lead time of 10–30 days is the gap between weather (<10 days) and climate (>30 days) predictions. Improving the forecast skill of extreme weather events at the extended range is crucial for risk management of disastrous events. In this study, three deep learning (DL) models based on the methods of convolutional neural networks and gate recurrent units are constructed to predict the rainfall anomalies and associated extreme events in East China at lead times of 1–6 pentads. All DL models show skillful prediction of the temporal variation of rainfall anomalies (in terms of temporal correlation coefficient skill) over most regions in East China beyond 4 pentads, outperforming the dynamical models from the China Meteorological Administration (CMA) and the European Centre for Medium-Range Weather Forecasts (ECMWF). The spatial distribution of the rainfall anomalies is also better predicted by the DL models than the dynamical models; and the DL models show higher pattern correlation coefficients than the dynamical models at lead times of 3–6 pentads. The higher skill of DL models in predicting the rainfall anomalies will help to improve the accuracy of extreme-event predictions. The Heidke skill scores of the extreme rainfall event forecast performed by the DL models are also superior to those of the dynamical models at a lead time beyond about 4 pentads. Heat map analysis for the DL models shows that the predictability sources are mainly the large-scale factors modulating the East Asian monsoon rainfall.
Significance Statement
Improving the forecast skill for extreme weather events at the extended range (10–30 days in advance), particularly over populated regions such as East China, is crucial for risk management. This study aims to develop skillful models of the rainfall anomalies and associated extreme heavy rainfall events using deep learning techniques. The models constructed here benefit from the capability of deep learning to identify the predictability sources of rainfall variability, and outperform the current operational models, including the ECMWF and the CMA models, at forecast lead times beyond 3–4 pentads. These results reveal the promising application prospect of deep learning techniques in the extended-range forecast.
Abstract
A new probabilistic tornado detection algorithm was developed to potentially replace the operational tornado detection algorithm (TDA) for the WSR-88D radar network. The tornado probability algorithm (TORP) uses a random forest machine learning technique to estimate a probability of tornado occurrence based on single-radar data, and is trained on 166 145 data points derived from 0.5°-tilt radar data and storm reports from 2011 to 2016, of which 10.4% are tornadic. A variety of performance evaluation metrics show a generally good model performance for discriminating between tornadic and nontornadic points. When using a 50% probability threshold to decide whether the model is predicting a tornado or not, the probability of detection and false alarm ratio are 57% and 50%, respectively, showing high skill by several metrics and vastly outperforming the TDA. The model weaknesses include false alarms associated with poor-quality radial velocity data and greatly reduced performance when used in the western United States. Overall, TORP can provide real-time guidance for tornado warning decisions, which can increase forecaster confidence and encourage swift decision-making. It has the ability to condense a multitude of radar data into a concise object-based information readout that can be displayed in visualization software used by the National Weather Service, core partners, and researchers.
Significance Statement
This study describes the tornado probability algorithm (TORP) and its performance. Operational forecasters can use TORP as real-time guidance when issuing tornado warnings, causing increased confidence in warning decisions, which in turn can extend tornado warning lead times.
Abstract
A new probabilistic tornado detection algorithm was developed to potentially replace the operational tornado detection algorithm (TDA) for the WSR-88D radar network. The tornado probability algorithm (TORP) uses a random forest machine learning technique to estimate a probability of tornado occurrence based on single-radar data, and is trained on 166 145 data points derived from 0.5°-tilt radar data and storm reports from 2011 to 2016, of which 10.4% are tornadic. A variety of performance evaluation metrics show a generally good model performance for discriminating between tornadic and nontornadic points. When using a 50% probability threshold to decide whether the model is predicting a tornado or not, the probability of detection and false alarm ratio are 57% and 50%, respectively, showing high skill by several metrics and vastly outperforming the TDA. The model weaknesses include false alarms associated with poor-quality radial velocity data and greatly reduced performance when used in the western United States. Overall, TORP can provide real-time guidance for tornado warning decisions, which can increase forecaster confidence and encourage swift decision-making. It has the ability to condense a multitude of radar data into a concise object-based information readout that can be displayed in visualization software used by the National Weather Service, core partners, and researchers.
Significance Statement
This study describes the tornado probability algorithm (TORP) and its performance. Operational forecasters can use TORP as real-time guidance when issuing tornado warnings, causing increased confidence in warning decisions, which in turn can extend tornado warning lead times.
Abstract
A hybrid three-dimensional ensemble–variational (En3D-Var) data assimilation system has been developed to explore incorporating information from an 11-member regional ensemble prediction system, which is dynamically downscaled from a global ensemble system, into a 3-hourly cycling convective-scale data assimilation system over the western Maritime Continent. From the ensemble, there exists small-scale ensemble perturbation structures associated with positional differences of tropical convection, but these structures are well represented only after the downscaled ensemble forecast has evolved for at least 6 h due to spinup. There was also a robust moderate negative correlation between total specific humidity and potential temperature background errors, presumably because of incorrect vertical motion in the presence of clouds. Time shifting of the ensemble perturbations, by using those available from adjacent cycles, helped to ameliorate the sampling error prevalent in their raw autocovariances. Monthlong hybrid En3D-Var trials were conducted using different weights assigned to the ensemble-derived and climatological background error covariances. The forecast fits to radiosonde relative humidity and wind observations were generally improved with hybrid En3D-Var, but in all experiments, the fits to surface observations were degraded compared to the baseline 3D-Var configuration. Over the Singapore radar domain, there was a general improvement in the precipitation forecasts, especially when the weighting toward the climatological background error covariance was larger, and with the additional application of time-shifted ensemble perturbations. Future work involves consolidating the ensemble prediction and deterministic system, by centering the ensemble prediction system on the hybrid analysis, to better represent the analysis and forecast uncertainties.
Abstract
A hybrid three-dimensional ensemble–variational (En3D-Var) data assimilation system has been developed to explore incorporating information from an 11-member regional ensemble prediction system, which is dynamically downscaled from a global ensemble system, into a 3-hourly cycling convective-scale data assimilation system over the western Maritime Continent. From the ensemble, there exists small-scale ensemble perturbation structures associated with positional differences of tropical convection, but these structures are well represented only after the downscaled ensemble forecast has evolved for at least 6 h due to spinup. There was also a robust moderate negative correlation between total specific humidity and potential temperature background errors, presumably because of incorrect vertical motion in the presence of clouds. Time shifting of the ensemble perturbations, by using those available from adjacent cycles, helped to ameliorate the sampling error prevalent in their raw autocovariances. Monthlong hybrid En3D-Var trials were conducted using different weights assigned to the ensemble-derived and climatological background error covariances. The forecast fits to radiosonde relative humidity and wind observations were generally improved with hybrid En3D-Var, but in all experiments, the fits to surface observations were degraded compared to the baseline 3D-Var configuration. Over the Singapore radar domain, there was a general improvement in the precipitation forecasts, especially when the weighting toward the climatological background error covariance was larger, and with the additional application of time-shifted ensemble perturbations. Future work involves consolidating the ensemble prediction and deterministic system, by centering the ensemble prediction system on the hybrid analysis, to better represent the analysis and forecast uncertainties.
Abstract
Herein, 14 severe quasi-linear convective systems (QLCS) covering a wide range of geographical locations and environmental conditions are simulated for both 1- and 3-km horizontal grid resolutions, to further clarify their comparative capabilities in representing convective system features associated with severe weather production. Emphasis is placed on validating the simulated reflectivity structures, cold pool strength, mesoscale vortex characteristics, and surface wind strength. As to the overall reflectivity characteristics, the basic leading-line trailing stratiform structure was often better defined at 1 versus 3 km, but both resolutions were capable of producing bow echo and line echo wave pattern type features. Cold pool characteristics for both the 1- and 3-km simulations were also well replicated for the differing environments, with the 1-km cold pools slightly colder and often a bit larger. Both resolutions captured the larger mesoscale vortices, such as line-end or bookend vortices, but smaller, leading-line mesoscale updraft vortices, that often promote QLCS tornadogenesis, were largely absent in the 3-km simulations. Finally, while maximum surface winds were only marginally well predicted for both resolutions, the simulations were able to reasonably differentiate the relative contributions of the cold pool versus mesoscale vortices. The present results suggest that while many QLCS characteristics can be reasonably represented at a grid scale of 3 km, some of the more detailed structures, such as overall reflectivity characteristics and the smaller leading-line mesoscale vortices would likely benefit from the finer 1-km grid spacing.
Significance Statement
High-resolution model forecasts using 3-km grid spacing have proven to offer significant forecast guidance enhancements for severe convective weather. However, it is unclear whether additional enhancements can be obtained by decreasing grid spacings further to 1 km. Herein, we compare forecasts of severe quasi-linear convective systems (QLCS) simulated using 1- versus 3-km grids to document the potential value added of such increases in grid resolutions. It is shown that some significant improvements can be obtained in the representation of many QLCS features, especially as regards reflectivity structure and in the development of small, leading-line mesoscale vortices that can contribute to both severe surface wind and tornado production.
Abstract
Herein, 14 severe quasi-linear convective systems (QLCS) covering a wide range of geographical locations and environmental conditions are simulated for both 1- and 3-km horizontal grid resolutions, to further clarify their comparative capabilities in representing convective system features associated with severe weather production. Emphasis is placed on validating the simulated reflectivity structures, cold pool strength, mesoscale vortex characteristics, and surface wind strength. As to the overall reflectivity characteristics, the basic leading-line trailing stratiform structure was often better defined at 1 versus 3 km, but both resolutions were capable of producing bow echo and line echo wave pattern type features. Cold pool characteristics for both the 1- and 3-km simulations were also well replicated for the differing environments, with the 1-km cold pools slightly colder and often a bit larger. Both resolutions captured the larger mesoscale vortices, such as line-end or bookend vortices, but smaller, leading-line mesoscale updraft vortices, that often promote QLCS tornadogenesis, were largely absent in the 3-km simulations. Finally, while maximum surface winds were only marginally well predicted for both resolutions, the simulations were able to reasonably differentiate the relative contributions of the cold pool versus mesoscale vortices. The present results suggest that while many QLCS characteristics can be reasonably represented at a grid scale of 3 km, some of the more detailed structures, such as overall reflectivity characteristics and the smaller leading-line mesoscale vortices would likely benefit from the finer 1-km grid spacing.
Significance Statement
High-resolution model forecasts using 3-km grid spacing have proven to offer significant forecast guidance enhancements for severe convective weather. However, it is unclear whether additional enhancements can be obtained by decreasing grid spacings further to 1 km. Herein, we compare forecasts of severe quasi-linear convective systems (QLCS) simulated using 1- versus 3-km grids to document the potential value added of such increases in grid resolutions. It is shown that some significant improvements can be obtained in the representation of many QLCS features, especially as regards reflectivity structure and in the development of small, leading-line mesoscale vortices that can contribute to both severe surface wind and tornado production.
