Bias correction of radar rainfall estimates based on a geostatistical technique

Various types of rainfall characteristic have different rainfall drop size distributions (DSDs). DSDs that are measured by radar have a fundamental influence on parameters of the Z-R relationship; using a climatological Z-R relationship to estimate radar rainfall can lead to bias in radar rainfall estimates. This paper attempts to remove the source of bias in radar rainfall estimates due to an uncertain Z-R relationship by applying a local bias adjustment factor to a region that has the same climatological rainfall characteristic. Recorded historical daily rainfall data from 188 uniformly distributed rain gauges located under the radar umbrella and its vicinity were used for describing the climatological spatial pattern of rainfall in the study area based on kriging approaches. It was found that a simple kriging technique with the isotropic Bessel-J semivariogram model was the best method to classify climatological patterns of rainfall characteristic of the study area and therefore it has been used for identifying local bias correction areas of the proposed hourly local bias (HLB) correction method. Performances of different bias correction methods with various levels of complexity were evaluated from 500 of the calibrated and cross-validated gauges of the validated data set, selected randomly. These methods include mean field bias correction (MFB), hourly mean field bias correction (HMFB), hourly range dependent mean field bias correction (HRMFB), and HLB correction. Forty-four rainfall events recorded during 2003-2005 from the S-band Pimai radar located in Nakhon-Ratchasima Province, Thailand, and 50 automatic rain gauges were used in this study. The results of this study showed that, on average, the proposed HLB method could improve accuracy of radar rainfall estimates by 16.7%, 14.3%, 2.8%, 0.4% for the calibrated gauges, and by 11.8%, 10.2%, 9.4%, 4.1% for the cross-validated gauges when compared to non-bias corrected, MFB, HMFB, and HRMFB methods, respectively.