Search Results
You are looking at 1 - 7 of 7 items for
- Author or Editor: Stephan Rasp x
- Refine by Access: Content accessible to me x
Abstract
Ensemble weather predictions require statistical postprocessing of systematic errors to obtain reliable and accurate probabilistic forecasts. Traditionally, this is accomplished with distributional regression models in which the parameters of a predictive distribution are estimated from a training period. We propose a flexible alternative based on neural networks that can incorporate nonlinear relationships between arbitrary predictor variables and forecast distribution parameters that are automatically learned in a data-driven way rather than requiring prespecified link functions. In a case study of 2-m temperature forecasts at surface stations in Germany, the neural network approach significantly outperforms benchmark postprocessing methods while being computationally more affordable. Key components to this improvement are the use of auxiliary predictor variables and station-specific information with the help of embeddings. Furthermore, the trained neural network can be used to gain insight into the importance of meteorological variables, thereby challenging the notion of neural networks as uninterpretable black boxes. Our approach can easily be extended to other statistical postprocessing and forecasting problems. We anticipate that recent advances in deep learning combined with the ever-increasing amounts of model and observation data will transform the postprocessing of numerical weather forecasts in the coming decade.
Abstract
Ensemble weather predictions require statistical postprocessing of systematic errors to obtain reliable and accurate probabilistic forecasts. Traditionally, this is accomplished with distributional regression models in which the parameters of a predictive distribution are estimated from a training period. We propose a flexible alternative based on neural networks that can incorporate nonlinear relationships between arbitrary predictor variables and forecast distribution parameters that are automatically learned in a data-driven way rather than requiring prespecified link functions. In a case study of 2-m temperature forecasts at surface stations in Germany, the neural network approach significantly outperforms benchmark postprocessing methods while being computationally more affordable. Key components to this improvement are the use of auxiliary predictor variables and station-specific information with the help of embeddings. Furthermore, the trained neural network can be used to gain insight into the importance of meteorological variables, thereby challenging the notion of neural networks as uninterpretable black boxes. Our approach can easily be extended to other statistical postprocessing and forecasting problems. We anticipate that recent advances in deep learning combined with the ever-increasing amounts of model and observation data will transform the postprocessing of numerical weather forecasts in the coming decade.
Abstract
The statistical theory of convective variability developed by Craig and Cohen in 2006 has provided a promising foundation for the design of stochastic parameterizations. The simplifying assumptions of this theory, however, were made with tropical equilibrium convection in mind. This study investigates the predictions of the statistical theory in real-weather case studies of nonequilibrium summertime convection over land. For this purpose, a convection-permitting ensemble is used in which all members share the same large-scale weather conditions but the convection is displaced using stochastic boundary layer perturbations. The results show that the standard deviation of the domain-integrated mass flux is proportional to the square root of its mean over a wide range of scales. This confirms the general applicability and scale adaptivity of the Craig and Cohen theory for complex weather. However, clouds tend to cluster on scales of around 100 km, particularly in the morning and evening. This strongly impacts the theoretical predictions of the variability, which does not include clustering. Furthermore, the mass flux per cloud closely follows an exponential distribution if all clouds are considered together and if overlapping cloud objects are separated. The nonseparated cloud mass flux distribution resembles a power law. These findings support the use of the theory for stochastic parameterizations but also highlight areas for improvement.
Abstract
The statistical theory of convective variability developed by Craig and Cohen in 2006 has provided a promising foundation for the design of stochastic parameterizations. The simplifying assumptions of this theory, however, were made with tropical equilibrium convection in mind. This study investigates the predictions of the statistical theory in real-weather case studies of nonequilibrium summertime convection over land. For this purpose, a convection-permitting ensemble is used in which all members share the same large-scale weather conditions but the convection is displaced using stochastic boundary layer perturbations. The results show that the standard deviation of the domain-integrated mass flux is proportional to the square root of its mean over a wide range of scales. This confirms the general applicability and scale adaptivity of the Craig and Cohen theory for complex weather. However, clouds tend to cluster on scales of around 100 km, particularly in the morning and evening. This strongly impacts the theoretical predictions of the variability, which does not include clustering. Furthermore, the mass flux per cloud closely follows an exponential distribution if all clouds are considered together and if overlapping cloud objects are separated. The nonseparated cloud mass flux distribution resembles a power law. These findings support the use of the theory for stochastic parameterizations but also highlight areas for improvement.
Abstract
Air parcel ascent in midlatitude cyclones driven by latent heat release has been investigated using convection-permitting simulations together with an online trajectory calculation scheme. Three cyclones were simulated to represent different ascent regimes: one continental summer case, which developed strong convection organized along a cold front; one marine winter case representing a slantwise ascending warm conveyor belt; and one autumn case, which contains both ascent types as well as mesoscale convective systems. Distributions of ascent times differ significantly in mean and shape between the convective summertime case and the synoptic wintertime case, with the mean ascent time being one order of magnitude larger for the latter. For the autumn case the distribution is a superposition of both ascent types, which could be separated spatially and temporally in the simulation. In the slowly ascending airstreams a significant portion of the parcels still experienced short phases of convective ascent. These are linked to line convection in the boundary layer for the wintertime case and an elevated conditionally unstable layer in the autumn case. Potential vorticity (PV) modification during ascent has also been investigated. Despite the different ascent characteristics it was found that net PV change between inflow and outflow levels is very close to zero in all cases. The spread of individual PV values, however, is increased after the ascent. This effect is more pronounced for convective trajectories.
Abstract
Air parcel ascent in midlatitude cyclones driven by latent heat release has been investigated using convection-permitting simulations together with an online trajectory calculation scheme. Three cyclones were simulated to represent different ascent regimes: one continental summer case, which developed strong convection organized along a cold front; one marine winter case representing a slantwise ascending warm conveyor belt; and one autumn case, which contains both ascent types as well as mesoscale convective systems. Distributions of ascent times differ significantly in mean and shape between the convective summertime case and the synoptic wintertime case, with the mean ascent time being one order of magnitude larger for the latter. For the autumn case the distribution is a superposition of both ascent types, which could be separated spatially and temporally in the simulation. In the slowly ascending airstreams a significant portion of the parcels still experienced short phases of convective ascent. These are linked to line convection in the boundary layer for the wintertime case and an elevated conditionally unstable layer in the autumn case. Potential vorticity (PV) modification during ascent has also been investigated. Despite the different ascent characteristics it was found that net PV change between inflow and outflow levels is very close to zero in all cases. The spread of individual PV values, however, is increased after the ascent. This effect is more pronounced for convective trajectories.
Abstract
Humans excel at detecting interesting patterns in images, for example, those taken from satellites. This kind of anecdotal evidence can lead to the discovery of new phenomena. However, it is often difficult to gather enough data of subjective features for significant analysis. This paper presents an example of how two tools that have recently become accessible to a wide range of researchers, crowdsourcing and deep learning, can be combined to explore satellite imagery at scale. In particular, the focus is on the organization of shallow cumulus convection in the trade wind regions. Shallow clouds play a large role in the Earth’s radiation balance yet are poorly represented in climate models. For this project four subjective patterns of organization were defined: Sugar, Flower, Fish, and Gravel. On cloud-labeling days at two institutes, 67 scientists screened 10,000 satellite images on a crowdsourcing platform and classified almost 50,000 mesoscale cloud clusters. This dataset is then used as a training dataset for deep learning algorithms that make it possible to automate the pattern detection and create global climatologies of the four patterns. Analysis of the geographical distribution and large-scale environmental conditions indicates that the four patterns have some overlap with established modes of organization, such as open and closed cellular convection, but also differ in important ways. The results and dataset from this project suggest promising research questions. Further, this study illustrates that crowdsourcing and deep learning complement each other well for the exploration of image datasets.
Abstract
Humans excel at detecting interesting patterns in images, for example, those taken from satellites. This kind of anecdotal evidence can lead to the discovery of new phenomena. However, it is often difficult to gather enough data of subjective features for significant analysis. This paper presents an example of how two tools that have recently become accessible to a wide range of researchers, crowdsourcing and deep learning, can be combined to explore satellite imagery at scale. In particular, the focus is on the organization of shallow cumulus convection in the trade wind regions. Shallow clouds play a large role in the Earth’s radiation balance yet are poorly represented in climate models. For this project four subjective patterns of organization were defined: Sugar, Flower, Fish, and Gravel. On cloud-labeling days at two institutes, 67 scientists screened 10,000 satellite images on a crowdsourcing platform and classified almost 50,000 mesoscale cloud clusters. This dataset is then used as a training dataset for deep learning algorithms that make it possible to automate the pattern detection and create global climatologies of the four patterns. Analysis of the geographical distribution and large-scale environmental conditions indicates that the four patterns have some overlap with established modes of organization, such as open and closed cellular convection, but also differ in important ways. The results and dataset from this project suggest promising research questions. Further, this study illustrates that crowdsourcing and deep learning complement each other well for the exploration of image datasets.
Abstract
Kilometer-scale models allow for an explicit simulation of deep convective overturning but many subgrid processes that are crucial for convective initiation are still poorly represented. This leads to biases such as insufficient convection triggering and late peak of summertime convection. A physically based stochastic perturbation scheme (PSP) for subgrid processes has been proposed (Kober and Craig) that targets the coupling between subgrid turbulence and resolved convection. The first part of this study presents four modifications to this PSP scheme for subgrid turbulence: an autoregressive, continuously evolving random field; a limitation of the perturbations to the boundary layer that removes artificial convection at night; a mask that turns off perturbations in precipitating columns to retain coherent structures; and nondivergent wind perturbations that drastically increase the effectiveness of the vertical velocity perturbations. In a revised version, PSP2, the combined modifications retain the physically based coupling to the boundary layer scheme of the original scheme while removing undesirable side effects. This has the potential to improve predictions of convective initiation in kilometer-scale models while minimizing other biases. The second part of the study focuses on perturbations to account for convective initiation by subgrid orography. Here the mechanical lifting effect is modeled by introducing vertical and horizontal wind perturbations of an orographically induced gravity wave. The resulting perturbations lead to enhanced convective initiation over mountainous terrain. However, the total benefit of this scheme is unclear and we do not adopt the scheme in our revised configuration.
Abstract
Kilometer-scale models allow for an explicit simulation of deep convective overturning but many subgrid processes that are crucial for convective initiation are still poorly represented. This leads to biases such as insufficient convection triggering and late peak of summertime convection. A physically based stochastic perturbation scheme (PSP) for subgrid processes has been proposed (Kober and Craig) that targets the coupling between subgrid turbulence and resolved convection. The first part of this study presents four modifications to this PSP scheme for subgrid turbulence: an autoregressive, continuously evolving random field; a limitation of the perturbations to the boundary layer that removes artificial convection at night; a mask that turns off perturbations in precipitating columns to retain coherent structures; and nondivergent wind perturbations that drastically increase the effectiveness of the vertical velocity perturbations. In a revised version, PSP2, the combined modifications retain the physically based coupling to the boundary layer scheme of the original scheme while removing undesirable side effects. This has the potential to improve predictions of convective initiation in kilometer-scale models while minimizing other biases. The second part of the study focuses on perturbations to account for convective initiation by subgrid orography. Here the mechanical lifting effect is modeled by introducing vertical and horizontal wind perturbations of an orographically induced gravity wave. The resulting perturbations lead to enhanced convective initiation over mountainous terrain. However, the total benefit of this scheme is unclear and we do not adopt the scheme in our revised configuration.
Abstract
Different approaches for representing model error due to unresolved scales and processes are compared in convective-scale data assimilation, including the physically based stochastic perturbation (PSP) scheme for turbulence, an advanced warm bubble approach that automatically detects and triggers absent convective cells, and additive noise based on model truncation error. The analysis of kinetic energy spectrum guides the understanding of differences in precipitation forecasts. It is found that the PSP scheme results in more ensemble spread in assimilation cycles, but its effects on the root-mean-square error (RMSE) are neutral. This leads to positive impacts on precipitation forecasts that last up to three hours. The warm bubble technique does not create more spread, but is effective in reducing the RMSE, and improving precipitation forecasts for up to 3 h. The additive noise approach contributes greatly to ensemble spread, but it results in a larger RMSE during assimilation cycles. Nevertheless, it considerably improves the skill of precipitation forecasts up to 6 h. Combining the additive noise with either the PSP scheme or the warm bubble technique reduces the RMSE within cycles and improves the skill of the precipitation forecasts, with the latter being more beneficial.
Abstract
Different approaches for representing model error due to unresolved scales and processes are compared in convective-scale data assimilation, including the physically based stochastic perturbation (PSP) scheme for turbulence, an advanced warm bubble approach that automatically detects and triggers absent convective cells, and additive noise based on model truncation error. The analysis of kinetic energy spectrum guides the understanding of differences in precipitation forecasts. It is found that the PSP scheme results in more ensemble spread in assimilation cycles, but its effects on the root-mean-square error (RMSE) are neutral. This leads to positive impacts on precipitation forecasts that last up to three hours. The warm bubble technique does not create more spread, but is effective in reducing the RMSE, and improving precipitation forecasts for up to 3 h. The additive noise approach contributes greatly to ensemble spread, but it results in a larger RMSE during assimilation cycles. Nevertheless, it considerably improves the skill of precipitation forecasts up to 6 h. Combining the additive noise with either the PSP scheme or the warm bubble technique reduces the RMSE within cycles and improves the skill of the precipitation forecasts, with the latter being more beneficial.