Research Collection Doctoral Thesis Performance and error diagnosis of global and regional NWP models Author(s): Didone, Marco Publication Date: 2006 Permanent Link: https://doi.org/10.3929/ethz-a-005294422 Rights / License: In Copyright - Non-Commercial Use Permitted This page was generated automatically upon download from the ETH Zurich Research Collection. For more information please consult the Terms of use. ETH Library Diss. ETH No. 1C597 Performance and Error Diagnosis of Global and Regional NWP Models A dissertation submitted to the Swiss Federal Institute of Technology (ETH) Zürich For the degree of Doctor of Sciences Presented by Marco Didone Dipl. Phys. ETH born 18 May, 1977 citizen of Carabbia TI, Switzerland Accepted on the recommendation of Prof. Dr. Huw C. Davies, ETH Zürich, examiner Prof. Dr. Dino Zardi, University of Trento, co-examiner Dr. Daniel Lüthi, ETH Zürich, co-examiner April 2006 Lontano lontano come un cieco m'hanno portato per mano Uiigaretti, 1917 Contents Abstract viii Riassunto x 1 Theorie, Models and Data 1 1.1 Theoretical Aspects 1 1.1.1 Basic Notions 1 1.1.2 Thermal Wind 2 1.1.3 Potential Vorticity 2 1.1.4 Properties of PV 4 1.1.5 Isentropic PV 5 1.1.6 Vertically Integrated PV 5 1.2 The 'Lokal Modell' G 1.3 Other Numerical Models 9 1.3.1 ECMWF Model and Data Set 9 1.3.2 The 'Climate High Resolution Model' 10 1.4 Lagrangian Trajectories 10 1 Errors of Large Scale Models 12 2 Case Study: ECMWF Model Errors 13 iii iv Contents 2.1 Introduction 13 2.2 The approach 14 2.3 Pacific Case Study 15 2.4 North Atlantic Case Study 17 2.5 Rationale 18 3 Nature of the 'Forecast - Analysis' Difference Fields 19 3.1 The Potential Vorticity Perspective 19 3.2 Prevalent Features of Error Fields 20 3.3 Time Evolution of the Forecast Errors 21 3.4 F-A Dynamics 26 3.5 High Resolution Simulations in the Pacific 27 3.6 Summary and Conclusion 28 4 Backtrajectories 31 4.1 Approach 31 4.2 Pacific Case 33 4.3 Atlantic Case 36 4.4 Summary and Conclusions 38 5 Idealized Simulations 41 5.1 Introduction to Countour Dynamics 41 5.2 Strong Negative Vorticity 46 5.3 Weak Negative Vorticity 47 5.4 Summary and Conclusions 49 6 Error Climatology 51 6.1 The approach 51 6.2 A Seasonal Climatology 52 Contents v 6.2.1 The Tiaditional View 52 6.2.2 The PV Perspective 54 6.3 The Nine Months Climatology 56 6.4 Evolution of PV Error Growth 58 II Studies with a Regional Model 59 7 Introductory Remarks and Case Description 61 7.1 Introduction 61 7.2 MAP IOP 8 64 7.3 MAP IOP 15 65 7.4 ECMWF Forecast 66 7.5 The Analysis Dataset 67 8 Horizontal vs. Vertical Resolution 69 8.1 Horizontal Resolution 69 8.2 Setting the Vertical Levels 70 8.3 Results - Comments 71 9 Error in Precipitation Field 73 9.1 IOP-15 74 9.2 IOP-8 75 9.3 Conclusions 78 10 Domain Study 81 10.1 Intioduction and Procedure 81 10.2 IOP-15 Case 82 10.3 IOP-8 Case 84 10.4 Conclusions 85 vi Contents 11 The Storm Lothar 87 11.1 Aim and Approach 88 11.2 Sensitivity to the Initial Conditions 89 11.3 A3 Dimensional PV Perspective 92 11.4 Conclusions 94 A Appendix to Chapter 5 97 B Appendix to Chapter 6 101 Bibliography 103 Acknowledgements 109 Curriculum Vitae 111 Seite Leer / Blank leaf i Abstract The new generation of supercomputers developed in the last years allow more and more sophisticated global and regional weather forecast models. Nonetheless deficiencies in modeling the atmosphere behavior will always be present due to its strong unpredictability. Thus, fundamental numerical modeling tasks include (i) the calibration of a model configuration to maximize the simulations quality and (ii) the investigation of systematic errors that lead to constant biases in the forecasts. These two aspects are considered in this thesis, where the latter is explored within a global numerical weather prediction suite, and the former on a regional scale at very high resolution. The first part of the thesis is devoted to the analysis of forecast 'errors' of the ECMWF model (defined as the differences between forecast and analysis). Var¬ ious statistical methods have been applied in studies on systematic model er¬ rors, where the main interest was to quantify the deviation on classical fields as geopotential and temperature. Here the PV perspective is adopted, as intrinsic properties of this perpective have direct implications for the study of forecast error fields. Moreover the dynamics of rapid error growth is linked to distinctive PV features. Two cases have been selected for the analysis of the forecasts de¬ viations, one occurring in the Atlantic and the other one in the Pacific ocean. Single model runs as well as the time evolution of the forecasts are taken into account to identify the features characterising these fields and their dynamics. Back trajectories are calculated to detect differences in the flow advection gener¬ ated by the model and the analysis. Hints of an undorstimation in the analysis of the negative PV regions along the southern flank of the jet that may enhance and even trigger the 'errors' are identified. Idealized simulations are generated with a contour-dynamics based model. They show the behavior of a positive vorticity region patially surrounded by a negative vorticity line. Strong similari¬ ties with the 'forecast minus analysis' fields are highlighted. Finally, to obtain a geographical distribution of the systematical deficiencies of the forecasts and to identify regions where the observation and/or data assimilations may have to be improved, a climatology of the 'errors' is produced. The aim of the second part of the dissertation is to investigate the potential benefits and the problems involved with the increase in resolution of a regional high resolution model. Particular attention is devoted to the precipitation field, a key mesoscale phenomenon when forecasting extreme weather events. The Spe¬ cial Observing Period (SOP) of the Mesoscale Alpine Programme (MAP) that has taken place in autumn 1999 had among the scientific objectives to improve understanding of orographic precipitation, gap flows and foehn, providing also a Abstract ix useful database for validating numerical models. For this occasion the ECMWF produced in addition to the operational analysis, a second dataset assimilating extra observations to be used as initial and boundary conditions: This opportu¬ nity is used here and two cases are selected among the MAP as case studies for the verification. The model adopted in this part is the 'Lokal Modell' developed by COSMO. Dif¬ ferent set-ups are tested, with particular interest in the horizontal and vertical resolution. Two distinct configurations are then identified: one for the coarse simulation, where the convection in parametrised, and the second for the nested convection resolving simulations, and are adopted for the sequent investigations. A first application is to compare the observed data of the MAP database to the 24hrs precipitation accumulation over the Alpine Area of the coarse and of the high resolution simulations, using operational analysis and reanalysis as ini¬ tial and boundary conditions. The observed differences in the flow advection that lead to disagreement in precipitation amounts between model and analysis trigger the next study, where the influence of the domain size of the coarser sim¬ ulation on the large scale flow of the nested integration is examined on different fields. In the last application in the framework of high resolution modeling three hindcast simulations of the storm Lothar initialized at different times are con¬ ducted to evidence the extreme importance of the initial and lateral data time resolution in case of explosive cyclogenesis. The integration is performed only at 7km resolution due to the exceptional phase velocity of the storm. Moreover a dry simulation is also carried out to highlight (i) the fundamental contribution of diabatic processes to the explosive evolution of the storm and (ii) the difficulties of the model to simulate a violent condensation rate and therefore PV production to keep up with the observed intensification. Riassunto Lo sviluppo dei modelli numerici a scala globale e regionale, dovuto alia nuova generazione di supercomputers costruiti negli ultimi anni, permette previsioni del tempo sempre pin particolareggiate e a lungo termine. Nonstante cid inesattezze e imprecisioni riscontrate cercando di modellare il comportamento deh"atmosfera sono e saranno sempre presenti. Per questo motivo tra i compiti fondamentali di chi lavora nel campo dei modelli meteorologici si distinguono sprattutto (i) quello di configurare il modello numerico in modo da massimizzare la qualità della previsione e (ii) analizzare gli errori sistematici che portano ad una costante polarizzazione dei risultati. Nel corso di quest a, tesi sono affrontati entrambi gli aspetti: il primo su scala regionale ad alt a risoluzione, il secondo nell'ambito di una previsione numerica globale. La prima parte della dissertazione è dedicata aH'analisi degli 'errori' di pre¬ visione (definiti corne la differenza tra modello ed analisi) del modello IFS del centro meteorologico europeo ECMWF. Vari metodi statistici sono stati appli- cati, in diversi studi, agli errori sistematici, cercando di quantificare la differenza di previsione attraverso grandezze classiche quali il geopotenziale e la temper- atura. In questo studio è adottata invece la prospettiva della vorticita Potenziale (PV) in quanto alcune sue propriété intrinsiche sono direttamente implicate nello studio degli 'errori' di previsione. Questa prospettiva evidenzia inoltre partico- lari caratteristiche collegate alla teoria degli errori ('rapid error growth'). Pr l'analisi di queste grandezze sono stati selezionati due casi avvenuti nell'oceano Atlantico e rispettivamente nel Pacifico.