Zero-covariance hypothesis in the error variance separation method of radar rainfall verification

Empirical test of the zero-covariance assumption in the error variance separation (EVS) method is presented. The EVS method is a way to filter out ground reference (GR) errors in radar rainfall verifications. It is based on a hypothesis that the errors of radar and gauge area-rainfall estimates are not significantly correlated. The test area within the Little Washita watershed in Oklahoma is covered by a relatively dense network of raingauges providing good approximations of the true area-rainfall used for this test. The investigation uses a large data sample of two 6-month periods and regards accumulation intervals from 15 min to 7 days. The test results are provided with bootstrap error bounds that confirm their statistical significance. The results show that, for this testing setup, the zero-covariance assumption in its previously postulated rigorous formulation is not fulfilled. However, despite the drawbacks, the EVS method can often provide better estimates of the radar error variances than the radar–raingauge comparisons that ignore the GR uncertainties.