The impact of observations in convective-scale numerical weather prediction

Necker, Tobias (2019): The impact of observations in convective-scale numerical weather prediction. Dissertation, LMU München: Fakultät für Physik

Abstract

The accuracy of the initial conditions strongly determines the skill of numerical weather prediction (NWP). Data assimilation systems combine millions of observations with the latest short-range forecast to provide optimal initial conditions. Only recently, NWP centers are capable of performing high-resolution, convection-permitting forecasts on a regional scale. However, moving to a higher model resolution involves several challenges concerning observations and the underlying data assimilation algorithm. The chaotic nature and limited predictability of convection calls for spatially and temporally high resolved observations. However, limited knowledge exists on which observations are most important for high-resolution NWP. Hence, a better understanding of the impact of different observations on these scales is required to improve current data assimilation, forecasting, and observing systems. Furthermore, knowledge of the potential impact of observations is needed to develop advanced observation and data assimilation strategies for future convective-scale NWP. This thesis, therefore, investigates the impact of observations in convective-scale ensemble forecasting. The impact of assimilated observation and the potential impact of future observations is evaluated by applying two complementary ensemble-based methods. Both methods rely on sample correlations that are estimated with an ensemble. However, state of the art ensemble prediction systems usually provide ensembles with only 20-250 members for estimating the uncertainty of the forecast and its spatial and temporal covariance. Given that the degrees of freedom of atmospheric models are several magnitudes higher, sample correlations are significantly affected by sampling errors. Therefore, sampling errors pose an issue for the impact assessment and in many other ensemble applications. Thus, it is essential to quantify sampling errors on convective-scales and to find methods to mitigate sampling errors. To address the previously discussed challenges, this dissertation aims to estimate the impact of observations and to reduce the issue of sampling error in convective-scale modeling and ensemble diagnostics. The first part of this thesis evaluates the impact of about 3 million conventional observations in the regional ensemble forecasting system of Deutscher Wetterdienst. This study presents the first evaluation of ensemble-based estimates of observation impact over an extended period of six weeks in a convection-permitting modeling system. Nearly all previous observation impact studies used the difference between the forecast and subsequent analysis of the same modeling system for verification. However, this kind of verification does not adequately reflect relevant forecast aspects of convective-scale forecasting. Hence, the observation impact is examined for different observation-based verification norms. The second part introduces an approach for estimating the relative potential impact of different observable quantities in convective-scale modeling. The approach is based on ensemble sensitivity analysis and uses accumulated squared spatiotemporal correlations as a proxy for the potential impact. To obtain reliable spatiotemporal correlations, a very large ensemble is required. Therefore, an unprecedented convective-scale 1000-member ensemble was computed in collaboration with the RIKEN Institute for computational science. This simulation allows to examine the sensitivity of the approach to the ensemble size. The present study further highlights the scale dependence of the potential impact and provides the basis for developing better observation and data assimilation strategies. The third part uses the 1000-member ensemble simulation as truth to quantify sampling errors on convective-scales and to evaluate a statistical sampling error correction. The sampling error correction is a simple look-up table based approach and aims to reduce spurious correlations. A detailed evaluation for spatiotemporal correlations shows that the sampling error correction significantly reduces sampling errors in sample correlations that are required for estimating the impact of observations. Additionally, the study demonstrates the great potential of the sampling error correction method for data assimilation where it could replace distance-based localization techniques and thereby increase the impact of observations.

• Dokumententyp: Dissertationen (Dissertation, LMU München)
• Keywords: numerical weather prediction, observation impact, potential impact, sampling error, 1000-member ensemble
• Themengebiete: 500 Naturwissenschaften und Mathematik ; 500 Naturwissenschaften und Mathematik > 530 Physik
• Fakultäten: Fakultät für Physik
• Sprache der Hochschulschrift: Englisch
• Datum der mündlichen Prüfung: 29. Juli 2019
• 1. Bericht­erstatter:in: Weissmann, Martin