Current state of CI predictive capabilities: The NSSL perspective

Current state of CI predictive capabilities: The NSSL perspective Jack Kain, Stuart Miller, Patrick Marsh, Adam Clark, ValliappaLakshmanan, others NOAA/ARL/ATDD September 4, 2013

NOAA Hazardous Weather Testbed Experimental Forecast Program Experimental Warning Program Prediction of hazardous weather events from a few hours to a week in advance Detection and prediction of hazardous weather events up to several hours in advance Where practitioners and researchers work together… Local NWS forecast office: Regional responsibility Storm Prediction Center: Nationwide responsibility Forecasting Research Warning Research Satellite-based Research

The annual Spring Forecasting Experiment (SFE) • The culmination of year-long activities in the HWT…a 6-8 week experiment conducted each spring to evaluate emerging scientific concepts and tools in a simulated operational forecasting environment …Scenes from SFE2011-12

A key ingredient in recent SFEs has been ensemble prediction systems run by CAPS. WRF-ARW members were used to produce CI forecast guidance: - Core members with varying IC and Physics - Physics Members with varying PBL and/or MP schemes

CAPS 4 km ensemble forecast domain Focus on next-day (18-24 h) forecasts

Convective Initiation studies during the HWT 2011-12 SFEs • Objectives: • Establish a baseline skill level for CI prediction in current convection-allowing (~4 km grid spacing) models • Develop and test prototype numerical guidance and forecast products for CI prediction

Experimental Forecasts: Probability of Convection and CI 10% 70% 40% X 2230Z Forecast Domain Valid 20-00Z focus area for CI

Experimental Forecasts: Probability of Convection and CI Model Guidance

Experimental Forecasts: Probability of Convection and CI Human temporal probability forecast Ensemble Guidance Observed Initiation

CI studies during the HWT 2011-12 SFEs • Framing the Problem was a big challenge! • Do we want to study the CI process in 4-km models? No! • Do we want to write algorithms to detect ongoing/developing CI processes in the models? No! • Strategy: • Find a reasonable output proxy for deep convection in 4 km models • Identify convectively active points with high temporal frequency (5 mins) • Identify CI points as a subset of all convective points • Focus on POD? • Implications of False Alarms?

...a reasonable output proxy for deep convection in 4 km models: • Reflectivity ≥ 35 dBZ at the -10ºC level (MTR35): • - Reflectivity aloft, associated with graupel formation • - Good indicator of convection • Less contaminated by clutter, biological echoes than low-level reflectivity • Seemingly direct analogue between observed (88D) reflectivity and model-simulated reflectivity (needed for verification)

Time-Series of MTR35: Observed and Simulated NSSL NMQ observations NSSL 4km WRF (WSM6) simulations 11 May – 10 June 2011 OBS/SIM CA r2: 0.880 Simulated Observed Frequency Bias ~ 1

OK, so “MTR35” seems like a reasonable proxy for convection... What are the challenges in formulating an algorithm for detecting “new” convection – not associated with ongoing activity - Maximize Probability of Detection - Minimize False Alarm Rate

MYJ Forecast/Verification Domain YSU OBS

Another case...

0 ACM2 Forecast/Verification Domain MYNN OBS

Another case...

MYJ YSU OBS

Questions... >> Do we want to count every new cell as a CI event? >> Is real-time detection required (increases False-Alarm Rate)?

Real-time CI algorithm developed by ValliappaLakshmanan • Define convection as: • Reflectivity at -10C exceeds 35 dBZ • New convection: • Was below 35 dBZ in previous image • Images are 5 minutes apart • Done on a pixel-by-pixel basis • But allow for growth and movement of ongoing convection

ValliappaLakshmanan Real time: Image at t0

ValliappaLakshmanan Real time: Image at t1

ValliappaLakshmanan Real time: Observed CI

ValliappaLakshmanan Methodology • Take image at t0 and warp it to align it with the image at t1 • Warping limited to a 5 pixel movement • Determined by cross-correlation with a smoothness constraint imposed on it • 5 pixels in 5 min  60kmph maximum movement • Then, do a neighborhood search • Pixels above 35 dBZ with no pixel above 35 dBZwithin 15km of aligned image is “New Convection”

ValliappaLakshmanan Example: Image at t0

ValliappaLakshmanan Example: Image at t1

ValliappaLakshmanan Example: Image at t0 aligned to t1

ValliappaLakshmanan Classification

Adam Clark algorithm: >> Convective objects defined as groups of convectively-active grid points contiguous in space and time. >> CI points identified as time minima points within each object >> objects must span at least 30 mins Ongoing Convection 60 50 40 30 20 10 0 Minutes since point was convectively active CI Points

Adam Clark Example: CI over Oklahoma 24 May 2011 Observed CI: Many more storms than in ensemble members Microphysics members (MYJ PBL) Control member (Thompson/MYJ) PBL members (Thompson MP)

Adam Clark 2012 Spring Forecasting Experiment Graphics Pre-defined CI domain Cumulative sum of convective activity Timing of observed first CI Team-generated probabilities for first CI Cumulative sum of convective initiation SSEF system forecast of first CI

Verification Statistics over forecast-team defined time (+/- 5h) and space windows….

Stuart Miller Mean: -0.192 h σ: 1.523 h CI Timing CAPS “Core” ensemble

Stuart Miller Mean: -0.173 h σ: 1.287 h CI Timing (CI_2) PBL Set

Current state of CI predictive capabilities: The NSSL perspective