Uncertainties in COSMO-DE precipitation forecasts introduced by model perturbations and variation of lateral boundaries

https://doi.org/10.1016/j.atmosres.2010.12.008Get rights and content

Abstract

As a first step towards a convection-permitting ensemble prediction system (EPS), this study explores the use of perturbation methods within the numerical weather prediction (NWP) model COSMO-DE. The study isolates uncertainty sources so that their impact can be separately studied. The focus is set on uncertainties in model physics and lateral boundary conditions which are represented by a multi-parameter and a multi-boundary approach. Experimental ensemble forecasts of precipitation with a lead time of 24 h are generated. Three ensemble setups are constructed: one with model perturbations, one with variations of boundaries and one with combined perturbations. The investigation period comprises 15 days in summer 2007. Deterministic verification shows that each individual member leads to quantitative precipitation forecasts (QPFs) within a reasonable quality range. Verification shows that the probabilistic precipitation forecasts of the experimental ensembles are superior to the deterministic forecasts. Measures of ensemble dispersion show that the impact of the perturbations on the forecast strongly varies with lead time, with model perturbations always dominating the first few hours and variations of lateral boundaries often dominating the following forecast hours. The study concludes that the applied perturbation methods lead to potentially useful probabilistic precipitation forecasts and should be considered as part of a future EPS design.

Research Highlights

► We examine perturbation techniques for an atmospheric ensemble prediction system. ► Model physics and lateral boundary conditions are varied. ► We focus on precipitation forecasts in a convection-permitting model. ► There is a quality gain compared to a single unperturbed forecast. ► Relative effect of perturbation types on ensemble dispersion depends on lead time.

Introduction

QPFs based on NWP are an essential input for hydrological predictions. For example, high-quality precipitation forecasts are critical for flash flood warnings, river flood predictions and water resources management.

Unfortunately, QPFs are one of the least accurate products in NWP (Ebert et al., 2003). Advances are expected to be realized by an improved representation of convection (Fritsch and Carbone, 2004). Many operational weather forecasting centres are now developing or operating high-resolution convection-permitting numerical weather prediction models. For example, Deutscher Wetterdienst (DWD) is operating a 2.8 km grid-spacing configuration of the COSMO model, named COSMO-DE, formerly known as LM-K (Baldauf et al., 2006). These new high-resolution convection-permitting models attempt to explicitly simulate key processes at mesoscale, such as small-scale orographic effects, severe convection, and heavy precipitation events.

Since the explicit simulation of more and more atmospheric processes is generally beneficial for such forecasts, hydrological applications should particularly benefit from the step towards convection-permitting models. Many precipitation events which are crucial in hydrological risk management are intense and very localized in space and time. Atmospheric models with parameterized convection cannot properly simulate this kind of event, due to their natural limitations in describing convective processes. Convection-permitting models are expected to provide better forecasts of such events in terms of intensity, spatial structure and diurnal cycle.

However, convective processes are highly nonlinear and associated with very short life times. As a consequence, their forecasts are strongly affected by uncertainties, e.g. emerging from the initial state of the model, from the formulation of the model and from the lateral boundary conditions which are provided by the driving model. This meteorological uncertainty can also propagate to the uncertainty of hydrological forecasts, because hydrological forecasts are often strongly linked to the precipitation forecast (Zappa et al., 2010).

These uncertainties in combination with highly nonlinear processes severely limit the deterministic predictability of mesoscale precipitation features. Deterministic predictability can be measured by the extent to which model runs, which are undistinguishable with respect to the uncertainties in producing the forecast, diverge with forecast lead time.

The decreased grid-spacing is therefore not necessarily expected to increase the quality of single model simulations as measured by “grid-based” metrics (Mass et al., 2002). It is rather expected to allow an improved spatial/temporal representation of statistical properties of convective precipitation (Ament et al., 2011-this issue, Weusthoff et al., 2010).

However, many applications still have to use grid-based NWP model output. Therefore, instead of using the grid-based, convection-permitting forecasts in a purely deterministic sense, there is an urgent need to provide hydrologists and other users with probabilistic guidance from high-resolution NWP model output, so that the potential benefit of these model simulations can also become evident in the applications (Fritsch and Carbone, 2004).

EPSs have advanced to a standard technique for dealing with limited predictability and uncertainties in NWP (Lewis, 2005, and references within). Therefore, DWD is developing an EPS based on the high-resolution convection-permitting model COSMO-DE. Early steps in this development have already been accompanied by hydrological considerations (Gebhardt et al., 2008). Whereas many experimental and operational mesoscale EPS have been developed in the past 10 years, only very few experience has been gained with regard to convection-permitting EPS. Due to computational limitations, mesoscale EPS are still restricted to a grid-spacing of approximately 10 km or larger, and, therefore, use parametrization of deep convection (e.g., Du et al., 2004, Marsigli et al., 2008, Bowler et al., 2008). With the rapid advancement of computer technology, rising interest in convection-permitting ensembles can be expected. The construction of a convection-permitting EPS is not necessarily a straightforward analogue of a mesoscale EPS with parametrized convection, because error growth differently and more rapidly occurs at smaller scales (Hohenegger and Schär, 2007). Therefore, the effectiveness of known perturbation generation techniques must be explored again when switching over to convection-permitting ensembles.

This study is one step in constructing a convection-permitting EPS based on the model COSMO-DE. Ensemble generation techniques are applied with the aim to study their impact on QPFs and their potential for producing reasonable forecasts. The study examines the deterministic forecasts of the individual ensemble members, the probabilistic forecasts from the entire ensemble and also the ensemble dispersion. The investigation is based on a time period of 15 days and includes forecast verification against radar estimations and rain gauge observations.

The method of this study is the isolation of various sources of uncertainty. The study focuses on the variation of lateral boundary conditions and on the perturbation of model physics. Variation of initial conditions is left to future work. Uncertainties in model physics and in lateral boundary conditions are first represented in two distinct COSMO-DE ensembles. The respective perturbation methods consist in a multi-parameter and in a multi-boundary technique. In addition, a third COSMO-DE ensemble represents the combination of the two perturbation techniques.

This paper investigates the effects of the perturbations in order to gain some knowledge about their general usefulness. Investigating the technical aspects of operational feasibility is beyond the scope of this paper.

Section 2 describes the model COSMO-DE, the ensemble specifications, the observational data and the examined time period. Section 3 describes the methods for ensemble evaluation and Section 4 presents the results. Section 5 provides a summary and suggestions for future work.

Section snippets

The model COSMO-DE

The COSMO-DE is a non-hydrostatic and convection-permitting model for very short-range forecasts which has been developed in the framework of COSMO at DWD (Baldauf et al., 2006). The COSMO-DE is a high-resolution configuration of the COSMO model. The COSMO-DE model domain (Fig. 1) is centred over Germany and uses a rotated regular longitude-latitude grid. The grid box size is approximately 2.8 km. There are 50 vertical levels up to 30 hPa, the lowest level is at 10 m height. There are 12 levels

Deterministic quality measures

The first step in the evaluation of the perturbation methods (Section 2) is a quality check of each individual ensemble member as a deterministic forecast. This aims at validating whether each new system configuration still leads to reasonable QPFs. Since the perturbation method simply gathers multiple boundary conditions and/or multiple physics configurations, systematic differences between the long-term error statistics of each individual ensemble member are expected.

Two frequently used

Quality check of each individual member

When investigating the use of perturbation methods (Section 2), the quality check of each individual member is an essential step. Since the individual members are produced by systematically different model setups, it is expected that the individual members differ in their long-term error statistics. However, the statistical differences should be fairly small, because the ensemble interpretation assumes that the individual members are equally likely. Therefore, the property of small statistical

Summary and suggestions for future work

This study examines the use of multi-boundary and multi-parameter perturbation techniques when constructing a convection-permitting EPS. The investigation looks into their impact on the forecast and into their potential to attain good forecast quality. Three ensemble setups are examined, with one setup representing uncertainties in model physics, the second setup representing uncertainties in lateral boundary conditions and the third setup combining the perturbation techniques of the first two.

Acknowledgements

We thank Michael Buchhold and two anonymous reviewers for their helpful and valuable comments. We are indebted to Chiara Marsigli and José A. Garcia-Moya for providing boundary data and sharing their expertise in ensemble prediction.

References (27)

  • F. Ament et al.

    Evaluation of MAP D-PHASE heavy precipitation alerts in Switzerland during summer 2007

    Atmos. Res.

    (2011)
  • M. Baldauf et al.

    The new very short range forecast model LMK for the convection-resolving scale

  • M.A. Ball et al.

    Comparison of stochastic parametrization approaches in a single-column model

    Philos. Trans. R. Soc. A

    (2008)
  • N.E. Bowler et al.

    The MOGREPS short-range ensemble prediction system

    Q. J. R. Meterol. Soc.

    (2008)
  • G.W. Brier

    Verification of forecasts expressed in terms of probability

    Mon. Wea. Rev.

    (1950)
  • R. Buizza

    Potential forecast skill of ensemble prediction, and spread and skill distributions of the ECMWF ensemble prediction system

    Mon. Wea. Rev.

    (1997)
  • R. Buizza et al.

    Stochastic representation of model uncertainties in the ECMWF ensemble prediction system

    Q. J. R. Meterol. Soc.

    (1999)
  • A.J. Clark et al.

    A comparison of precipitation forecast skill between small convection-allowing and large convection-parameterizing ensembles

    Wea. Forecasting

    (2009)
  • J. Du et al.

    The NOAA/NWS/NCEP short-range ensemble forecast (SREF) system: evaluation of an initial condition vs. multi-model physics ensemble approach

  • E.E. Ebert et al.

    The WGNE assessment of short-term quantitative precipitation forecasts

    Bull. Am. Meteorol. Soc.

    (2003)
  • J.M. Fritsch et al.

    Improving quantitative precipitation forecasts in the warm season

    Bull. Am. Meteorol. Soc.

    (2004)
  • J.A. Garcia-Moya et al.

    Predictability for Short-range Forecast: a Multi-model Approach

    (2007)
  • C. Gebhardt et al.

    Experimental ensemble forecasts of precipitation based on a convection-resolving model

    Atmos. Sci. Lett.

    (2008)
  • Cited by (0)

    View full text