Validation of CIRA Tropical Cyclone Algorithms Julie Demuth, Mark DeMaria, John Knaff, Kotaro Bessho, Kimberly Mueller, and Ray Zehr CoRP Satellite Calibration and Validation Symposium 14 July 2005 Outline • AMSU intensity and wind radii estimation – M. DeMaria, J. Demuth, J. Knaff – new datasets – different methods – new estimation models • AMSU 2-D surface wind retrieval – K. Bessho, M. DeMaria, J. Knaff • IR wind structure estimation – M. DeMaria, K. Mueller, J. Knaff AMSU Intensity and Wind Radii • In general… – derive ~20 parameters from AMSU data – statistically relate them to dependent data (from extended best track) using MLR – develop algorithms to estimate TC intensity (MSW, MSLP) and axisymmetric 34-, 50-, 64-kt wind radii – use axisymmetric wind estimates with modified Rankine vortex model to estimate winds in NE, SE, SW, NW quadrants relative to TC center Int. & Winds - Data • n > 2600 cases … 5x more than before Data & – global dataset for intensity estimation 1999-2004 for AL, EP; 2002-04 for SH, WP; 2003- 45 cases at Cat-5 level 04 for CP, IO – for wind radii, used only cases with recon 12 hrs prior Atlantic 32.1% West Pacific 33.7% Central Pacific 0.2% Southern Hemisphere 7.2% Indian Ocean 1.4% East Pacific 25.4% 2x as many cases as before… 34: n=255 50: n=170 64: n=120 Int. & Winds - Methods • Added 4 variables to pool – tmax2, clwave2, tmax*clwave, p600 • Using “best subsets” MLR technique – tests all possible models with up to some N number of independent variables…we chose N=15 • Cross-validation – every model tested with 80/20 scheme run 1000 times • Model selection – minimize MAE of developmental and cross-validated datasets – = 0.01 for intensity models, = 0.05 for radii models Intensity - Results • MSW – NEW: R2=78.7%, MAE=10.8 kt – OLD: R2=76.4%, MAE=11.5 kt • MSLP: – NEW: R2 = 80.2%, MAE = 7.8 hPa – OLD: R2 = 76.4%, MAE = 8.9 hPa Intensity Results – Ivan Example 160 140 Old AMSU Est Hurricane Ivan Example (n=25) New AMSU Est MSW (kt) 120 Best Track 100 80 60 40 20 0 090317 090409 090505 090610 090707 090810 090923 091011 091200 091223 091409 091500 091600 Date and Time New MAE = 15.4 kt New RMSE = 18.0 kt Old MAE = 18.7 kt Old RMSE = 21.3 kt AMSU Wind Radii Results 35 34 MAE - new 30 NW 25 NE 34 MAE - old 20 50 MAE - new 15 10 50 MAE - old 5 64 MAE - new 0 64 MAE - old SW SE AMSU 2-D Surface Winds • Quick summary… – use nonlinear balance equation (Charney, 1955) to estimate 3-D wind field from AMSU data – compare AMSU-derived nonlinear balance winds at 850 hPa with QuikSCAT and H*Wind surface wind analyses AMSU wind speeds at 850 hPa linearly related to surface wind speeds characteristic biases of wind direction between AMSU and Quik SCAT or H*Wind – develop algorithm to convert 850 hPa to surface winds IR Wind Structure • Quick summary… – Use IR data to develop algorithms that estimate RMAX and V182 via MLR – Use these estimates with modified Rankine vortex model to estimate symmetric tangential wind profile – Add storm motion-derived wind asymmetry to reconstruct entire 2-D wind field Sources of More Info • Demuth et al. 2004 (JAM) • Demuth et al. (follow-up note submitted to JAM) • Bessho et al. (submitted to JAM) • Mueller et al. (submitted to Wea. Forecasting)