Experiments in Rainfall Estimation with a Polarimetric Radar

As part of a program to evaluate the utility of polarimetric radar for estimating rainfall, the National Center for Atmospheric Research's S-band dual-polarization radar was deployed in east central Florida during the summer of 1998. The field experiment, dubbed PRECIP98, coincided with a component of the National Aeronautics and Space Administration's Tropical Rainfall Measuring Mission. A unique dataset consisting of high-resolution polarimetric radar measurements, rain gauge observations, and raindrop disdrometer observations was collected.

Comparisons between radar observations and radar parameters derived from disdrometer observations (assuming a gamma drop-size distribution and equilibrium axis ratios) revealed that the radar estimates of differential reflectivity and specific differential phase were significantly less than those determined from disdrometer observations. Comparisons improved slightly when the calculations were based on a mean canting angle of 0^o and a standard deviation of 10^o and improved markedly for an empirical axis ratio relation representing more spherical drops.

Rainfall rates were estimated with the WSR-88D default radar reflectivity (Z_H) relation and with previously-published reflectivity-differential reflectivity (Z_H­Z_DR) and specific differential phase (K_DP) relations developed from simulations with equilibrium axis ratios and widely accepted ranges in the DSD governing parameters (N_o, L, and µ). Results showed small overall bias for radar reflectivity (~10%), but there was an overestimate of ~50% for the Z_H­Z_DR measurement pair and an underestimate of 30% for K_DP. The use of Florida-tuned relations for radar reflectivity and the reflectivity-differential reflectivity pair (assuming equilibrium axis ratios) reduced the biases to 3 and 8%, respectively. However, the bias for K_DP was unchanged, confirming an insensitivity to the drop-size distribution governing parameters.

Further fine tuning of the Florida relations for drop canting and more spherical drop shapes readily accounts for the residual bias with Z_H­Z_DR estimators. However, a significant underestimate of rainfall (14%) remained in the K_DP estimates even after allowing for more spherical drop shapes. This residual bias with K_DP is attributed to the loss of signal in storms dominated by small drops and the growth of errors at weak signals in light rainfalls.

In summary, rainfall estimates with Z_H­Z_DR estimators had the highest correlation with rain gauge observations (0.92), the smallest range in bias factors from storm to storm (1.73), and the smallest root mean square error for unbiased estimators (6.3 mm).