An Evaluation of Radar Rainfall Estimates from Specific Differential Phase

Specific differential propagation phase (KDP) is examined for estimating convective rainfall in Colorado and Kansas. Estimates are made at S band with KDP alone and in combination with radar reflectivity (ZH). Results are compared to gauge observations by computing bias factors, defined as the sum of gauge-measured rainfalls divided by the sum of radar estimates at gauges reporting rainfall, and the correlation coefficient between the gauge and radar-estimated amounts. Rainfall accumulations computed from positive-only values of KDP (provided ZH ≥ 25 dBZ) yield bias factors that vary from 0.76 to 2.42 for 3 storms in Colorado and from 0.78 to 1.46 for 10 storms in Kansas. Correlation coefficients between gauge-observed and radar-estimated rainfalls are 0.76 to 0.95. When negative KDP’s are included as negative rainfall rates, bias factors range from 0.81 to 3.00 in Colorado and from 0.84 to 2.31 in Kansas. In most storms, the correlation between gauge and radar rainfalls decreases slightly. In an experiment with the KDP/ZH combination, rainfall rates are computed from KDP when KDP is ≥0.4° km−1 and from ZH for KDP < 0.4° km−1 and ZH ≥ 25 dBZ. Neglect of the negative KDP’s and substitution of the always positive ZH rainfall rates result in a tendency to overestimate rainfall. Bias factors are 0.63–1.46 for Colorado storms and 0.68–0.97 for Kansas storms, and correlation coefficients between gauge and radar amounts are 0.80–0.95. In yet another test with the KDP/ZH pair, rainfall estimates are computed from KDP when ZH ≥ 40 dBZ and from ZH when 25 ⩽ ZH < 40 dBZ. For this experiment, bias factors range from 0.90 to 1.91 in Colorado and from 0.88 to 1.46 in Kansas. Correlation coefficients are 0.80–0.96. Since bias factors and correlation coefficients between estimated rainfalls and gauge observations for KDP are similar to those for radar reflectivity, there was no obvious benefit with KDP rainfalls for a well-calibrated radar. Large underestimates with KDP in two storms were attributed to rainfalls dominated by small drops. In one storm, the problem was aggravated by widespread negative KDP’s thought related to vertical gradients of precipitation. An advantage of KDP-derived rainfall estimates confirmed here is an insensitivity to anomalous propagation.