1. JCSDA Workshop Brock Burghardt August 5 2015

2. Model Configuration WRF-ARW v3.5.1 Forecasts integrated 30 hours Cycled every 6 hours (non-continuous boundaries) Hourly state output d01: dx=36 km d02: dx=12 km d03 : dx=4 km 38 vert. levels

3. Data Assimilation DART EAKF system (Kodiak; Anderson et al. 2009) 50 members Prior adaptive inflation NCEP observation error variances MADIS obs filtered into 6 h forecast background RAOB, ACARS, Cloud-Track winds, marine, metar, mesonet (except d01) Gaspari-Cohn cutoff function for localization covariance Half radiusD01D02D03 Vert. Local.4.5 km5 km2.5 km Horiz. Local.600 km300 km

4. Rank Histograms Image from: http://www.eumetcal.org/resources/ukmeteocal/temp/msgcal/www/english/msg/ver_prob _forec/uos4b/uos4b_ko1.htm

5. Ensemble Dispersion

6. Forcing spread  Multiplicative inflation  Prior  Posterior (tendency to be unstable)  Physics parameterizations  Explicit changes  Physics tendencies  Stochastic Kinetic-Energy Backscatter (SKEB) -Easy to implement (R2O; ESA) -Shown to improve mean forecast (Yussouf et al. 2013)

7. Varied Physics  Choose widely used, similar computational cost parameterizations that are sufficiently different  Vary physics across different domains (allowing ESA)  Applying (initially) to three convective events MicrophysicsCumulus Parameterization Planetary Boundary Layer Surface Layer Thompson*Kain-Fritsch*YSU*MM5 Monin-Obukhov* MorrisonBetts-Miller-JanjicMellor-Yamada- Janjic Monin-Obukhov (Janjic) WDM5Grell-FreitasACM2 (Pleim)Pleim-Xiu WSM5

8. Fixed vs Varied Physics

9. Defining Objects Specifically looking at the ‘initiation’ of convective rainfall objects (ideally dBZ). Defined by hourly rainfall rate thresholds Utilizing multi-sensor Stage IV analysis Visually this appears to well represent storm-scale aspects of the forecast. Limitations: Temporal discretization (hourly windows) Potential overproduction bias in Mexico Member 45 Objects 00z 20130516 Observed Objects 00z 20130516

10. Fixed vs Varied Physics Date is 20120430 Date is 20130515 Date is 20130520 Physics is fixedPhysics is variedPhysics is fixedPhysics is variedPhysics is fixedPhysics is varied Total number of matched objects= 1465 Total number of matched objects= 1447 Total number of matched objects= 1066 Total number of matched objects= 1172 Total number of matched objects= 494 Total number of matched objects= 641 Avg C error per object= 100.3385 Avg C error per object= 102.0084 Avg C error per object= 112.5541 Avg C error per object= 115.5445 Avg C error per object= 109.5923 Avg C error per object= 109.9551 final_FAR= 0.034278final_FAR= 0.045515final_FAR= 0.098139final_FAR= 0.19117final_FAR= 0 final_POD= 0.50517final_POD= 0.49897final_POD= 0.49581final_POD= 0.54512final_POD= 0.27444final_POD= 0.35611 final_BIAS= 0.5231final_BIAS= 0.52276final_BIAS= 0.54977final_BIAS= 0.67395final_BIAS= 0.27444final_BIAS= 0.35611 final_TS= 0.49627final_TS= 0.48737final_TS= 0.47043final_TS= 0.4829final_TS= 0.27444final_TS= 0.35611 Underproduction (low bias) Slight late timing bias

11. Initial results and future Little to no improvement in spread when varying physics on parent domains (particularly low levels where underdispersion is most notable) Tendency for too few convective rainfall objects (underdispersion?) Future -Generate more statistics (significance) -Continue running varied physics on inner domains

12. Spread in terms of convection In theory, a sufficiently large, well calibrated ensemble should encapsulate (time-space) observed convective objects r=200 km t=+/- 2 hr