160 likes | 275 Views
Robust Dual Frequency Radar Profiling Algorithm. Mircea Grecu 1 , Lian Tian 1 , and Simone Tanelli 2 GEST, UMBC and NASA GSFC Jet Propulsion Laboratory. Outline. Motivation General considerations Solution Generalized HB approach Ensemble methodology Results Conclusions. Motivation.
E N D
Robust Dual Frequency Radar Profiling Algorithm Mircea Grecu1, Lian Tian1, and Simone Tanelli2 GEST, UMBC and NASA GSFC Jet Propulsion Laboratory
Outline • Motivation • General considerations • Solution • Generalized HB approach • Ensemble methodology • Results • Conclusions
Motivation • Drop size distribution variability is the main source of uncertainties in reflectivity precipitation relationships • Even when multiple observations are available, radar-only retrievals are uncertain • For small drops the equations are identical • Reflectivity factors may be subject to severe attenuation
GPM challenges • Single frequency radar retrievals have to be consistent with dual frequency radar retrievals • Forward attenuation correction tends to be the method of choice in single frequency retrievals, while reverse attenuation correction is popular in dual-frequency retrievals • Radar retrievals have to be consistent with combined radar/radiometer retrievals • Radar retrievals algorithms need to be computational fast to be included in combined methodologies
Outline • Motivation • General considerations • Solution
Candidate solution • Generalized Hitschfeld Bordan profiling methodology can be applied to derive generic Ku (13.8-GHz) radar retrievals as a function of various parameters (e.g. vertical N0 profiles) • These parameters are optimized as a function of Ka (35.5-GHz) radar observations where these are available.
Incorporation of Ka band information into the Ku solution • Variational based methodology is used • Observation vector Y=[ZmKa, PIAKa, PIAKu] is from X=[N0] using the HB algorithm and cost function F(X)=0.5(Y-Yobs)TR-1(Y-Yobs)+0.5(X-XB)TB-1(X-XB) is minimized using a gradient based procedure • Grad(F(X)) is calculated using the adjoint formulation
Outline • Motivation • General considerations • Solution • Generalized HB approach • Variational methodology • Results
Synthetic data experiment • Ka observations were simulated from TRMM Ku observations for randomly assumed N0 profiles • DSD shapes were assumed known • TPW and cloud water profiles were assumed known • Random errors (1dB standard deviations) were included into the Ka observations
ResultsLiquid Water Content Single Frequency Retrievals Dual Frequency Retrievals C=0.92 RRMS=0.52 RBIAS=-0.05 C=0.80 RRMS=0.73 RBIAS=-0.15
ResultsDrop Size Distribution Intercept Dual Frequency Retrievals Single Frequency Retrievals C=0.89 RRMS=0.73 RBIAS=-0.01 C=0.62 RRMS=0.87 RBIAS=-0.02
Outline • Motivation • General considerations • Solution • Generalized HB approach • Ensemble methodology • Results • Conclusions
Conclusions • Variational methodology appears to be robust and perform satisfactorily • The methodology facilitates the development of combined radar radiometer retrievals • It is currently incorporated in a variational combined radar/radiometer framework • The impact of cloud water, TPW, phase transition and DSD shape uncertainties needs to be rigorously assessed.