HWRF Model Diagnostics Vijay Tallapragada HWRF Team Lead, NCEP/EMC HFIP Stream 1 Regional Hurricane Model Diagnostics Planning Meeting, 11/08/2010 1 Outline Diagnostics for systematic evaluation of storm track and intensity forecasts Complement the HFIP diagnostics team efforts ◦ Make use of preliminary diagnostic products to identify specific model behavior ◦ Provide more in-depth analysis of important model deficiencies and methods to improve model physics/dynamics/initialization Additional requirements for operational support ◦ Diagnostics tailored for operational needs ◦ Specific focus on HWRF model improvement 2 End of the season forecast verification statistics provide first order information on model performance (important measure of forecast skill) •Difficult to identify the source of errors •Storm to storm variability •Statistical properties of the forecast errors •Inter-dependency of intensity and track errors •Impact of initialization and model physics 3 Track forecasts are strongly linked to large-scale environment • • • • • • • Evolution of large-scale flow in HWRF domain Location and strength of troughs and ridges Steering currents, effects of shear Influence of lateral boundary conditions Sensitivity to physical parameterizations Impact of data analysis Non-physical issues (nest motion, interaction with topography, tracker issues etc.) 4 How good is the evolution of large-scale atmosphere in HWRF? Issues with nest interaction 6 Operational HWRF GFS Analysis HWRF fails to simulate increased shear in the environment 1/11/2010 Persistent northward bias, but reversal of zonal track bias 8 Intensity forecasts are dictated by several factors…. • Impact of Vortex initialization and data assimilation • Large-scale affects of shear, storm size, vertical structure of winds, temperature and moisture, asymmetries, scale interactions • Sensitivity to microphysics, PBL, land surface, surface physics and convection parameterizations • Effects of horizontal diffusion, divergence damping, momentum mixing, surface enthalpy coefficients etc. • Effects of air-sea interaction – MLD, OHC, SST, loop current, warm and cold core eddies • Effects of resolution, nest feedback 9 Rapid growth of intensity errors Pronounced negative bias 10 Systematic nature of HWRF Initialization Positive bias for weaker storms, negative for stronger storms (spinup/spindown) 64 kts threshold for storm size/intensity correction (based on more than 1200 forecasts) 11 Vortex Initialization Issues/Challenges H210 model produced Fewer strong storms Than observed HWRF 2009 production has too many strong storms Insufficient vortex size correction Broad initial vertical structure 12 t=00 t=06 Some of the maximum wind speed values are not associated with the storms. t=12 Band 3 Loop Simulated Band 3 Loop Real Convection and moisture transport in HWRF – systematic dryness 14 Outer Domain (27 km) Nest (9 km) Convective precipitation Outer Domain (27 km) Nest (9 km) Total precipitation Grid-scale Precipitation dominates convective precipitation at 9 km resolution 15 Impact of horizontal diffusion Impact of surface physics changes HWRF Track forecasts are somewhat less sensitive to changes in changes in horizontal diffusion and surface physics 16 Impact of data assimilation Analysis increments in the outer domain 17 Heat Fluxes (for ocean coupling) Adversely impacts SST fields in HyCOM POM vs HYCOM IKE example ◦ Model SST minus daily average GOES / AVHRR data. ◦ POM on left, too warm under hurricane. ◦ HYCOM on right, too cool in general. Existing Tools for Hurricane Model Diagnostics at EMC (I) Operational HWRF Model Output (archived): Track and intensity forecasts in ATCF format (6-hrly interval) 6-hourly model output on native model grid (horizontal, rotated lat/lon, E-grid) interpolated to standard pressure levels Model output in GRIB-1 and GRIB-2 format for selected 3-D variables on standard pressure levels, and a suite of surface variables Hourly max. 10m wind and rainfall (ascii) HWRF input and analysis files (binary) for initial time 6-hourly ocean model output for standard variables (T, S, U, V, W, MLD, OHC) Real-time automated graphical display of selected model forecast variables Track and intensity forecasts, max wind and rainfall swaths Animations of 850 hPa wind circulation (combined domain and nested domain) 10m winds and MSLP (nested domain) Vertical cross-sections of zonal and meridional wind (along the nest center) Simulated GOES IR and WV imagery Simulated Radar Reflectivity Real-time statistics for track and intensity forecast errors and biases 20 Existing Tools for Hurricane Model Diagnostics at EMC (II) HPLOT Diagnostics and visualization software GrADS/FORTRAN based software to plot model forecast variables in GRIB format Point and click vertical cross-sections, area-averaged quantities Model to model and model to analysis comparison (and difference plots) Variable level vertical wind shear and mean-layer wind Skew-T plots for thermodynamic structure Preliminary statistics for RMSE, ACC and Bias TC Vortex Diagnostics Generates azimutially averaged structures of the TC vortex Computes vortex center for each level separately (based on geopotential height) Determines departure from gradient wind balance for each instance BINPLOT software for visualization of model output on native grid User developed MATLAB tools for ocean model evaluation 21 Diagnostic Efforts at EMC – Vortex initialization Description Theory/Hypothesis Methodology Initialization issues HWRF model is quite sensitive to the initialization procedure, especially w.r.t. specification of storm size, cycling, surface pressure calculation, composite vortex, filter strength and vertical structure. Describe the problems with HWRF initialization. Sensitivity to GFDL filtering changes Consider changing the filter strength following GFDL approach Sensitivity to storm size specification Use of second parameter (34-kt wind radius and or radius of storm circulation Evaluate non-linear balance issues Explicit formulation for post-balance Gradient wind and hydrostatic balance Describe if there is a systematic bias in the gradient wind balance by examining different storms and different initial conditions. If this is a persistent feature of HWRF ICs, identify methods to correct this and impose dynamic and thermodynamic balance consistent with the model. Test the sensitivity of different approaches. Sensitivity to Composite Vortex Reconstruct axi-symmetric composite vortex using current operational HWRF configuration 22 Diagnostic Efforts at EMC – Vortex Evolution Description Theory/Hypothesis Methodology Vortex scale diagnostics Intensity forecast skill is closely linked to HWRF's ability to properly describe the evolution of vortex through organization of convection, primary and secondary circulations, distribution of heating and vertical motions, air-sea exchange of fluxes for heat and momentum and radiation, BL processes etc. 1. Identify cases with the following scenarios: a) HWRF tends to intensify the storm while in reality the storm weakens and vice-versa; b) reasonably good intensity forecast (2-3 day and 4-5 day) 2. Identify available datasets that can be used to evaluate model forecasts for these selected cases. Sensitivity to Microphysics Identify coding errors (if any) in the Ferrier Scheme and test their impacts on HWRF forecasts; Analyze cloud-radiation feedbacks (large-scale as well as storm scale); Sensitivity to BL parameterization 1. Sensitivity to specification of Ri; 2. Describe the BL processes through budget analysis of heat, momentum and moisture Sensitivity to initialization changes Analyze changes to vortex size and structure with reference to initialization experiments Sensitivity to changes in the ocean components Identify areas in HWRF/HYCOM/POM that have potential impact on improving (reducing) negative bias of HWRF intensity forecasts. Diagnostics Efforts at EMC - Track forecasts and sensitivity to initialization Description Theory/Hypothesis Methodology Effect of storm relocation and vortex initialization on track forecasts HWRF model is quite sensitive to the initialization procedure, especially w.r.t. specification of storm size, cycling, surface pressure calculation, composite vortex, filter strength and vertical structure. 1. Storm size correction using RMW and 34-kt wind radius 2. Post-balance method for masswind balance Sensitivity to GFDL filtering changes Consider changing the filter strength following GFDL approach Gradient wind and hydrostatic balance Impact of post-balanced vortex on track forecasts Sensitivity to Composite Vortex Reconstruct axi-symmetric composite vortex using current operational HWRF configuration Diagnostic Efforts at EMC – Large Scale Description Theory/Hypothesis Methodology Large scale Diagnostics It has been widely observed that evolution of large-scale features in HWRF tend to deviate from GFS analysis/forecast. Particularly the strengths of troughs and ridges and their location tend to show considerable differences. 1. Systematic bias in HWRF output for the following fields: (a) 850/700/500/200 hPa heights (b) wind fields at 850 hPa, 500 hPa and 200 hPa (c) Mean-layer winds (steering currents) (d) Vertical wind shear (e) Mean-layer thickness (f) Stream function at 850 hPa (g) Velocity Potential at 200 hPa Track bias • Systematic biases in HWRF model for track forecasts (phase and directional) 25 Diagnostic Efforts at EMC – Large Scale evolution Description Theory/Hypothesis Methodology Large scale Diagnostics Sensitivity to radiation parameterization • Replace GFDL radiation with RRTM (or simple Dudhia scheme) and analyze largescale features Sensitivity to LSM parameterization • NOAH LSM replacing slab model Sensitivity to convection • parameterization • Effects of momentum mixing on track forecasts Evaluation of heating rates and precipitation • Sensitivity to specification of Ri and mixing length Describe the BL processes through budget analysis of heat, momentum and moisture Sensitivity to PBL parameterization • Requirements for Stream-1 Model Diagnostics (I) What is required to assess the model performance and provide input to improve model formulation: • Define specific evaluation metrics along with suggestions for associated observed datasets. • Provide guidance on specific case studies that will be used in extended model diagnostics. • Identification of systematic errors and model biases • Development of objective techniques for comparing model fields (temperature, moisture etc) to various well-validated satellite-derived fields • PDFs and structure functions comparing statistical properties of model output with those from satellite derived observations, soundings from dropsondes, aircraft data and radar observations (ground based and aircraft based). • Derived 3-D wind and temperature fields from these remotely sensed and insitu observations for model evaluation. 27 Requirements for Stream-1 Model Diagnostics (II) • Computational methods for partitioning of forecast errors arising due to various components of model dynamics and physics • Heat/energy budget studies (to include contributions from diabatic heating terms, PV budget, moisture transport etc.) • Large scale track and intensity predictors, interactions among large scale (synoptic) and vortex scale (mesoscale) circulations • Off-line programs to study the impact of changes to physics and initialization (idealized simulations) • Event specific diagnostic methods to estimate the impacts of interaction with topography, impact of various features of the ocean where applicable • Multi-dimensional visualization tools • Automated model diagnostics tools to assist real-time and post-season evaluation of model performance • Diagnostics tailored to assist pre-implementation T&E • Develop capability to conduct sensitivity experiments in a common framework where all models can be initialized with the same initial vortex and have the same background environmental flow. 28 Requirements for Stream-1 Model Diagnostics (III) Vortex Scale Observations •Observations that describe details of vortex evolution (3-D structure) in different environments (thermodynamic/kinematic) Particularly in development stages What is hurricane scale circulation? Do we have comprehensive data sets that describe the intensity change process?? Microphysics – missing data above melting level, e.g. hydrometeor fields – nucleation is not well understood Large Scale: •GFS Analysis, gridded satellite derived datasets (atmosphere and ocean), composite 3-D analysis of the hurricane environment using aircraft dropsonde/radar etc. Requirements for Stream-1 Model Diagnostics (IV) Diagnostic Methods: •Compute derived variables using a common methodology using scientifically established procedures. •Identify standard as well as derived variables needed for diagnostics •Develop set of programs that can be used to compute extended diagnostic variables from respective model forecast and observational datasets. •Establish a code repository with a collection of such software Visualization Tools: •Common software that can incorporate forecast datasets from different models in assisting model inter-comparisons and evaluation with observations. •Integration of the capabilities of HPLOT, DIAPOST and TC Diagnostics •Incorporating MATLAB, BINPLOT and other applications •TC Diagnostics file and SHIPS predictors (Mark DeMaria); Initial Vortex Comparison (Wallace Hogsett); and Post-Balance evaluation (In-Hyuk Kwon) HWRF Annual Upgrade T&E and Process Automation for diagnostics: •HWRF team at EMC has a very ambitious and aggressive T&E for annual upgrade implementations. Diagnostics efforts should be geared to fit in the timeline. Need to develop a framework for process automation of standard as well as advanced diagnostics. •Bring the experience of experts in the community in analyzing these forecasts and provide feedback and suggestions for model improvement.