Download

Use of High-Resolution WRF Simulations to Forecast Lightning Threat






Advertisement
/ 25 []
Download Presentation
Comments
elewa
From:
|  
(1156) |   (0) |   (0)
Views: 93 | Added:
Rate Presentation: 5 0
Description:
Use of High-Resolution WRF Simulations to Forecast Lightning Threat. E. W. McCaul, Jr. USRA Huntsville/SPoRT National Space Science and Technology Center Huntsville, Alabama, USA. Science Advisory Council 13 June 2007. Photo, David Blankenship Guntersville, Alabama.
Use of High-Resolution WRF Simulations to Forecast Lightning Threat

An Image/Link below is provided (as is) to

Download Policy: Content on the Website is provided to you AS IS for your information and personal use only and may not be sold or licensed nor shared on other sites. SlideServe reserves the right to change this policy at anytime. While downloading, If for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.











- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -




Slide 1

Use of High-Resolution WRF Simulations to

Forecast Lightning Threat

E. W. McCaul, Jr.

USRA Huntsville/SPoRT

National Space Science and Technology Center

Huntsville, Alabama, USA

Science Advisory Council

13 June 2007

Photo, David Blankenship

Guntersville, Alabama

Slide 2

Premises and Objectives

Given:

Precipitating ice aloft is correlated with LTG rates

Mesoscale CRMs are being used to forecast convection

CRMs can represent many ice hydrometeors (crudely)

Goals:

Create WRF forecasts of LTG threat, based on ice flux

in layers near -15 C, and on simulated reflectivity aloft.

WRF: Weather Research and Forecast Model

CRM: Cloud Resolving Model

Additional Forecast Interests

CI - convective initiation

Ti - First lightning (35 dBZ at -15C, glaciation)

Tp - Peak flash rate

Tf - Final lightning

Slide 3

0 oC

Flash Rate Coupled to Mass in the Mixed Phase Region

Cecil et al., Mon. Wea. Rev. 2005 (from TRMM Observations)

Slide 4

WRF Lightning Threat Forecasts:Methodology

Use high-resolution (2-km) WRF simulations to prognose

convection for a series of case studies

Develop diagnostics from model output fields to serve as

proxies for LTG: graupel fluxes, reflectivity profiles, etc.

Calibrate WRF LTG proxies with actual total LTG rates from HSV LMA and reflectivity/LTG from TRMM PR/LIS

4. Draw maps of WRF LTG threat based on calibrated proxies

Assess WRF capabilities for forecasting LTG threat

Slide 5

Compute field of graupel flux at -15C

Plot peak values against observed peak LTG flash rates on the same grid, for high-LTG, low-LTG cases; empirically estimate the nature of the relationship between observed LTG, forecast proxy; estimate any coefficients needed to convert gridded proxy to gridded LTG rate

Here, we find good fit from F1 = 0.06 w qg (see next page)

With histogram truncation, could match areal coverage to observations; not done here because of loss of some simulated storms which appear to be important

After applying conversion function to WRF proxy field, obtain field of LTG flash rates F1 in units of fl/(5 min)/grid column

LTG Threat 1 derived from WRF graupel flux at -15C

Slide 6

Prototypical Calibration Curves

Threat 1 (Graupel flux)

Slide 7

Compute field of reflectivity Z1 at -15C

Compute reflectivity field Z2 at 3 km above -15C

Regress LIS flash rates against TRMM PR Z1 and Z2 for SE USA warm season; obtain three coefficients, one for Z1, one for Z2, and a threshold; this yields an uncalibrated reflectivity profile shape parameter Z that can be evaluated in WRF output:

Z = aZ1 + bZ2 + c, where a=1.0856, b= 0.5196, c = -49.49

Plot peak values of LMA flash rate density vs peak gridded Z, for low-LTG and high-LTG cases; empirically estimate functional relation; plot shows possibly quadratic functional relationship, F2 = 0.005Z2 (see next page)

Apply functional equation to Z data to obtain flash rate density field F2 in terms of fl/(5 min)/grid column

LTG Threat 2 derived from WRF reflectivity profile shape

Slide 8

Prototypical Calibration Curves

Threat 2 (dBZ profile)

Slide 9

Methods based on LTG physics, and should be robustly useful

Methods supported by solid observational evidence

Can be used to obtain quantitative estimates of flash rates

Methods are fast and simple; based on fundamental model output fields; no need for complex electrification modules

Methods can be revised to include thresholds, which will allow truncation of proxy histograms so as to allow matching of areal coverages of predicted and observed LTG activity

LTG Threat Methodology: Advantages

Slide 10

Methods are only as good as the numerical model output; model used here (WRF) has shortcomings in several areas

Inherent ambiguity in interpretation of threats (extent density, etc.)

Small number of cases means uncertainty in calibrations; choices of coefficient values may be off by factor of 2, but such errors are small compared to gross model errors in storm placement

Calibration requires successful simulation of both high- and low-LTG cases; the latter can be hard to find; in most cases, there are at least a few moderate or strong-LTG storms present in the domain

Calibrations must be redone whenever model is changed or upgraded

LTG Threat Methodology: Disadvantages

Slide 11

WRF Lightning Threat Forecasts:10 December 2004Cold-season Hailstorms, Little LTG

Slide 12

2-km horizontal grid mesh

51 vertical sigma levels

Dynamics and physics:

Eulerian mass core

Dudhia SW radiation

RRTM LW radiation

YSU PBL scheme

Noah LSM

WSM 6-class microphysics scheme (graupel; no hail)

8h forecast initialized at 12 UTC 10 December 2004 with AWIP212 NCEP EDAS analysis;

Also used METAR, ACARS, and WSR-88D radial velocities at 12 UTC;

Eta 3-h forecasts used for LBC’s

WRF Configuration (typical)10 December 2004 Case Study

Slide 13

Ground truth: LTG flash extent density, dBZ10 December 2004, 19Z

Slide 14

WRF fcst: LTG Threat 1, SBCAPE10 December 2004, 19Z

Slide 15

WRF fcst: LTG Threat 2, dBZ at -15C10 December 2004, 19Z

Slide 16

WRF Lightning Threat Forecasts:30 March 2002Squall Line plus Isolated Supercell

Slide 17

Ground truth: LTG flash extent density, dBZ30 March 2002, 04Z

Slide 18

WRF fcst: LTG Threat 1, SBCAPE30 March 2002, 04Z

Slide 19

WRF fcst: LTG Threat 2, dBZ at -15C30 March 2002, 04Z

Slide 20

Domainwide Peak Flash Density Time Series

Slide 21

Implications of time series:1. WRF LTG threat 1 peak values may have small pos. bias2. WRF LTG threat 1 peak values have proper t variability 3. WRF LTG threat 2 peak values may have bigger pos. bias4. WRF LTG threat 2 peak values have insufficient t variability5. WRF LTG threat mean biases are positive because our method of calibrating was designed to capture peak flash rates correctly, not mean flash rates6. WRF LTG threat t variability is too small for Threat 2, because of integrating effect of multi-layer methods7. WRF LTG threat 1 should give best performance

Slide 22

Conclusions 1:1. WRF forecasts of convection are useful, but of variable quality2. Timing of initiation of convection is depicted fairly well; however WRF convection is sometimes too widespread 3. Inclusion of WSR-88D velocity data helps; so does dBZ if storms already up at t=04. WRF wmax, dBZ values on 2-km grids sometimes too weak, relative to observed weather and 0.5-km RAMS simulations5. WSM6 microphysics still too simple; need more ice categories6. Finer model mesh may improve updraft representation, and hydrometeor amounts7. Biggest limitation is likely errors in initial mesoscale fields

Slide 23

Conclusions 2:1. Current LTG threat parameters provide more spatial coverage of threat as compared to actual LTG, but less than standard weather forecasts and less than area covered by CAPE>02. Coverage of graupel-flux LTG threat could be reduced by using a threshold value of graupel flux, but doing so eliminates some WRF convection that might be important3. Because of location, timing errors in WRF convection field, excess LTG threat spatial coverage may be desirable for now4. Graupel-flux LTG threat 1 shows large time rms, like obs; dBZ LTG threat 2, based on 2 levels, has small time rms

Slide 24

Future Work:1. Expand catalog of simulation cases to obtain robust statistics; consider physics/IC mini-ensemble simulation approach2. Continue refinement of dBZ profile LTG threat 2 method3. Test newer versions of WRF, when available: - more hydrometeor species - double-moment microphysics4. Run on 1-km or finer grids; study PBL scheme response5. Examine sensitivity of calibration constants to model physics and IC diversity6. In future runs, examine fields of interval-cumulative wmax, and associated hydrometeor and reflectivity data, not just the instantaneous values; for save intervals of 15-30 min, events happening between saves may be important for LTG threat7. Publish findings and collaborate with NWS, others

Slide 25

Acknowledgments:This research was funded by the NASA Science Mission Directorate’s Earth Science Division in support of the Short-term Prediction and Research Transition (SPoRT) project at Marshall Space Flight Center, Huntsville, AL. Collaborators:Kate LaCasse, UAH/SPoRTSteve Goodman, NOAA/NESDISDan Cecil, UAH


Copyright © 2014 SlideServe. All rights reserved | Powered By DigitalOfficePro