Monday, 29 January 2024: 5:45 PM
323 (The Baltimore Convention Center)
Valid time shifting (VTS) is a versatile, cost-effective method to increase the background
ensemble size for ensemble-based data assimilation (DA) systems. This is accomplished by
including ensemble member forecast output at multiple times surrounding the central analysis
time (before and after) within the background ensemble covariances for DA. This method
effectively increases the first guess ensemble size by a factor of 3 at a fraction of the added cost
to the system (approximately 50%). Another “cost-neutral” approach can approximately double
the ensemble size without incurring additional costs to the base system. VTS was systematically
tested during 2021 and 2022 Hazardous Weather Testbed (HWT) Spring Forecasting
Experiments (SFEs) within the University of Oklahoma (OU) Multiscale Data Assimilation and
Predictability (MAP) Laboratory. This 108-member VTS implementation is based off of a
configuration similar to the future Rapid Refresh Forecasting System (RRFS) with a base
ensemble size of 36. This RRFS-like system includes hourly sequential multiscale DA of
mesoscale in situ and convective-scale radar reflectivity observations in the GSI-based
ensemble-variational (EnVar) system, featuring direct assimilation of radar reflectivity
observations (Wang and Wang 2017) and coupled with the FV3-LAM. Results from these SFEs
have shown that multiscale VTS can systematically improve the analysis of convection and its
surrounding environment, leading to improved 0-15 h prediction of convective systems when
underlying model biases are not extreme.
In this study, blending of time-shifted subensembles is tested to further improve on the
previous real-time multiscale VTS configuration. Although systematic improvement was seen
from the SFEs, it is likely that the assumed 1-h time-shifting interval in the BothVTS
configuration of 2022 is suboptimal for all cases and both mesoscale and storm-scale DA
components. To account for more scales of phase uncertainty in the time-shifted subensembles,
two or more time-shiftings are blended into one VTS-blended subensemble. In this approach,
smaller scales from short-term time shifting (30-min) are blended together with larger scales taken
from a longer time-shifting (2-h) in order to capture optimal scales of uncertainty for storm-scale
and mesoscale DA components, respectively. Furthermore, the use of an external ensemble is
tested to capture the large-scale shiftings to further limit added computational costs and
potentially improve the practicality of VTS for an operational system such as RRFS. In this way,
the VTS blending can not only account for more phase-related scales of model uncertainty, but
may lend itself well to the future combination of VTS with more advanced simultaneous
multiscale DA methods that assimilate all scales of observations in one step.
ensemble size for ensemble-based data assimilation (DA) systems. This is accomplished by
including ensemble member forecast output at multiple times surrounding the central analysis
time (before and after) within the background ensemble covariances for DA. This method
effectively increases the first guess ensemble size by a factor of 3 at a fraction of the added cost
to the system (approximately 50%). Another “cost-neutral” approach can approximately double
the ensemble size without incurring additional costs to the base system. VTS was systematically
tested during 2021 and 2022 Hazardous Weather Testbed (HWT) Spring Forecasting
Experiments (SFEs) within the University of Oklahoma (OU) Multiscale Data Assimilation and
Predictability (MAP) Laboratory. This 108-member VTS implementation is based off of a
configuration similar to the future Rapid Refresh Forecasting System (RRFS) with a base
ensemble size of 36. This RRFS-like system includes hourly sequential multiscale DA of
mesoscale in situ and convective-scale radar reflectivity observations in the GSI-based
ensemble-variational (EnVar) system, featuring direct assimilation of radar reflectivity
observations (Wang and Wang 2017) and coupled with the FV3-LAM. Results from these SFEs
have shown that multiscale VTS can systematically improve the analysis of convection and its
surrounding environment, leading to improved 0-15 h prediction of convective systems when
underlying model biases are not extreme.
In this study, blending of time-shifted subensembles is tested to further improve on the
previous real-time multiscale VTS configuration. Although systematic improvement was seen
from the SFEs, it is likely that the assumed 1-h time-shifting interval in the BothVTS
configuration of 2022 is suboptimal for all cases and both mesoscale and storm-scale DA
components. To account for more scales of phase uncertainty in the time-shifted subensembles,
two or more time-shiftings are blended into one VTS-blended subensemble. In this approach,
smaller scales from short-term time shifting (30-min) are blended together with larger scales taken
from a longer time-shifting (2-h) in order to capture optimal scales of uncertainty for storm-scale
and mesoscale DA components, respectively. Furthermore, the use of an external ensemble is
tested to capture the large-scale shiftings to further limit added computational costs and
potentially improve the practicality of VTS for an operational system such as RRFS. In this way,
the VTS blending can not only account for more phase-related scales of model uncertainty, but
may lend itself well to the future combination of VTS with more advanced simultaneous
multiscale DA methods that assimilate all scales of observations in one step.

