An evaluation of weather radar adjustment algorithms using synthetic data

Citation:

Silver M, Karnieli A, Marra F, Fredj E. An evaluation of weather radar adjustment algorithms using synthetic data. Journal of Hydrology [Internet]. 2019;576 (February) :408–421.

Abstract:

Adjustment of weather radar estimates using observed precipitation has been an accepted procedure for decades. Ground observations of precipitation typically come from rain gauges, but can also include data from diverse networks of sensors, with different levels of reliability. This study presents a standardized framework for evaluating adjustment algorithms using synthetically constructed, but realistic, rain grids and weather radar rainfall. Ground observation points are randomly placed throughout the synthetic storm domain and the precipitation for each sensor is extracted from the true rain. Then a subset of the sensors are defined as unreliable, and a log-normal error factor is applied at those locations. This double network of rain sensors could be applicable, for example, when rainfall is derived from signal attenuation between commercial microwave link (CML) antennas. Past research has tested CML observations as a source of precipitation data and validated various radar adjustment algorithms. However, a comprehensive evaluation of adjustment algorithms using accurate gauge data mixed with CML observations at different densities is lacking. Five adjustment algorithms are applied to the synthetic radar grid: Mean Field Bias (MFB), a Multiplicative algorithm, Mixed (additive and multiplicative), Conditional Merge (CondMerge) and Kriging with External Drift (KED). Generation of the synthetic framework, and application of the adjustment algorithms is repeated for 150 realizations. Comparison of coefficient of determination (R2), root mean square error and linear regression for all adjustment procedures over all realizations indicates the following results. Only MFB and KED adjustments performed well when using accurate gauges. The kriging based KED was able to achieve good adjustment also with the addition of error-prone sensors. CondMerge and the Mixed and Multiplicative, however, resulted in poorer adjustments.

Website