Development of a fuzzy logic X-band hydrometeor identification algorithm Brenda Dolan and Steven Rutledge Colorado State University September 18, 2006 European Radar Conference

Road Map  Motivation and Background  Methodology Simulations using T-matrix and Mueller matrixSimulations using T-matrix and Mueller matrix  Results  Future work

Introduction  Analyze the microphysical and kinematic structure of storms observed by the CASA IP1 network Dual-Doppler analysisDual-Doppler analysis Polarimetric analysisPolarimetric analysis  Hydrometeor ID (HID)  Rain rate /snowfall estimation  Collaborative Adaptive Sensing of the Atmosphere (CASA) Integrated project (IP1) Network of 4 polarimetric X- band radars located in OklahomaNetwork of 4 polarimetric X- band radars located in Oklahoma

Introduction  Why is radar identification of hydrometeors important? Identify regions of mixed phaseIdentify regions of mixed phase Determine possible regions of freezing rain and sleetDetermine possible regions of freezing rain and sleet Use to drive rainrate and snowfall algorithmsUse to drive rainrate and snowfall algorithms Identify locations of possible ice contaminationIdentify locations of possible ice contamination Distinguish between aggregates and ice crystalsDistinguish between aggregates and ice crystals Locate large hailLocate large hail Help validate numerical modelsHelp validate numerical models Dolan and Rutledge, 2006

S-band vs. X-band  Extensive research and algorithm development at S-band (Straka et al. 2000, Liu and Chandrasekar 2001, Tessendorf et al. 2005) Include observational dataInclude observational data Extending into C-bandExtending into C-band  Little work has been done on HID algorithms for X-band  Why do we expect the HID algorithm for X-band and S-band to be different? Mie scatteringMie scattering ResonanceResonance AttenuationAttenuation K dp proportional to 1/  Bringi and Chandrasekar 2001, Matrosov et al. 2006)K dp proportional to 1/  Bringi and Chandrasekar 2001, Matrosov et al. 2006)

Fuzzy Logic One approach to HID is using fuzzy logic One approach to HID is using fuzzy logic Combine information from polarimetric variables, as well as temperature Combine information from polarimetric variables, as well as temperature Z h, Z dr, K dp,  hv in this case (no L dr ) Z h, Z dr, K dp,  hv in this case (no L dr ) Soft boundaries and ranges for each of the variables Soft boundaries and ranges for each of the variables Variables can be weighted Variables can be weighted Each hydrometeor type is given a score based on how well the variables fit into the fuzzy set Each hydrometeor type is given a score based on how well the variables fit into the fuzzy set Hydrometeor with highest score is considered the hydrometeor type at that data point Hydrometeor with highest score is considered the hydrometeor type at that data point

Fuzzy Logic: Membership Beta Functions (MBFs) a m b ZhZh ZhZh Z dr K dp LDR ρ hv T These MBFs are from Tessendorf et al. (2005) for S-band

Hydrometeor Types  We are interested in distinguishing the following precipitation types: RainRain DrizzleDrizzle Wet (low-density) graupelWet (low-density) graupel Dry (high-density) graupelDry (high-density) graupel Snow (wet)Snow (wet) Snow (dry)Snow (dry) Pristine ice crystals (dendrites/columns)Pristine ice crystals (dendrites/columns) Vertically aligned ice crystalsVertically aligned ice crystals Hail (added complications)Hail (added complications)

Hydrometeor Types  We are interested in distinguishing the following precipitation types: RainRain DrizzleDrizzle Wet (low-density) graupelWet (low-density) graupel Dry (high-density) graupelDry (high-density) graupel Snow (wet)Snow (wet) Snow (dry)Snow (dry) Pristine ice crystals (dendrites/columns)Pristine ice crystals (dendrites/columns) Vertically aligned ice crystalsVertically aligned ice crystals Hail (added complications)Hail (added complications)

Methodology  Determine the “most probable” ranges of variables at X-band Thus determining m (midpoint) and a (width) of MBFThus determining m (midpoint) and a (width) of MBF Using T-matrix and Mueller Matrix (Barber and Yeh 1975, Vivekanandan et al. 1991) scattering models to simulate a variety of precipitation eventsUsing T-matrix and Mueller Matrix (Barber and Yeh 1975, Vivekanandan et al. 1991) scattering models to simulate a variety of precipitation events  Provides simulated radar data  Assign a “confidence” to the determined ranges => b (MBF slope) As the algorithm is tested, modify these to become “more confident” about data rangesAs the algorithm is tested, modify these to become “more confident” about data ranges  Run simulations for both X and S-band (for comparison with literature)

Methodology: T-matrix  T-matrix: Simulates the scattering properties of a single particle Can change:Can change:  Radar wavelength (3.2 cm, 11.0 cm)  Particle input sizes (rain: 0.05 cm to 1.0 cm, drizzle: 0.1 cm- 0.6 cm)  Temperature (-10, 0, 10, 20 ºC)  Size-Axis ratio relationship (Goddard and Cherry, 1984)  Dielectric constant  Follow parameters for different hydrometeors outlined in Straka et al. (2000)  Run the model changing a single parameter at a time

Methodology: Mueller Matrix  Mueller Matrix: Generates the radar observables for a simulated “radar volume” Can changeCan change  Radar elevation angle (1.0, 30.0º)  Drop size distribution (DSD) (D o, N o )  Canting angles (distribution and std dev) (0-10º)  Range of particle diameters to use Base the parameters on Straka et al. (2000)Base the parameters on Straka et al. (2000)

Methodology: Mueller Matrix  Run the model for a wide variety of DSDs  Use Marshall Palmer for rain N(D)=N 0 exp - DN(D)=N 0 exp - D Fix N o at 80,000 m 3 cm -1Fix N o at 80,000 m 3 cm -1 Vary as a function of rain rateVary as a function of rain rate  =4.1 R R=1-200 mm hr -1 for rainR=1-200 mm hr -1 for rain  And assume: D o =3.67 /D o =3.67 /  Use monodisperse DSD for drizzle R= mm hr -1R= mm hr -1

Results: Rain dBzZ dr K dp  hv Blue : simulated X-band results Green : simulated S-band results Red : Straka et al (S00) S-band values (includes observational data)

Results: Drizzle dBzZ dr K dp  hv Blue : simulated X-band results Green : simulated S-band results Red : Straka et al (S00) S-band values (includes observational data)

Results: Key differences between S-band and X-band  For rain: Variables seem to be a bit higher at X-band over S-band Especially K dpEspecially K dp  We expect this with wavelength scaling  For drizzle: Z dr and K dp were higher for X- band over S-band

Future work  Resolve differences between S-band simulations and observations  Run simulations for Rain (using Gamma distribution)Rain (using Gamma distribution) GraupelGraupel SleetSleet Different ice crystals and snow typesDifferent ice crystals and snow types  Test new X-band HID against S-band using observations GPM Front Range Pilot ProjectGPM Front Range Pilot Project  Implement in CASA radar network Inter-compare with KOUNInter-compare with KOUN

Canting dist. thetamDSD type RRDoNoelev. Angle RAINGAUSSIAN0.1, , 10 MP1.0 to , 30.0 DRIZZLEGAUSSIAN0.1 to 4.0 MONO0.1 to , 30.0 SNOW (DRY) GAUSSIAN30MPS0.1 to , 30.0 SNOW (WET) GAUSSIAN30MPS0.1 to , 30.0 GRAUPEL (DRY) GAUSSIAN5.0, ???? 1.0, 30.0 GRAUPEL (WET) GAUSSIAN5.0, ???? 1.0, 30.0 PRISTINE ICEGAUSSIAN 0.5? 0.1 to , 30.0 VERTICLE ICEGAUSSIAN 0.5????1.0,30.0

λ (cm) T(ºC)ρd min d max Axis Ratio RUNSSOURCE RAIN to GC845S00 DRIZZLE to GC84105S00 SNOW (DRY) to to , LONG03 SNOW (WET) to to , 0.8 LONG03 GRAUPEL (DRY) to to BR84S00 GRAUPEL (WET) to to BR84S00 PRISTINE ICE <00.1 to LONG03 VERTICLE ICE <00.1 to S00

Results RAIN dbZZDR (dB)KDP (º/km)RhoT (ºC) X-band >-15 S-band >-15 SMBF22-62>00-6>0.93>-15 S00<60>0>0.95>0>-10 DRIZZLE X-band >0 S-band >0 SMBF<300-1~0>0.95>0 S >0.95>0 SNOW(DRY) X-band -35 to <0 S-band----<0 SMBF<40>00-1<0.93<0 S00< >0.95<0