Sensitivity of Streamflow Simulations to Temporal Variability and Estimation of $Z–R$ Relationships

This study focuses on the sensitivity of streamflow simulations to temporal variations in radar reflectivity–rainfall (i.e., $Z–R$ ) relationships. The physically based continuous-mode distributed hydrologic model—gridded surface subsurface hydrologic analysis—is used to predict runoff during three major rainfall-runoff periods observed in a $35{\mathrm{km}}^2$ experimental watershed in southern Louisiana. $Z–R$ relationships are derived at a series of temporal scales ranging from a climatological scale, where interstorm $Z–R$ variations are ignored, down to a subevent scale, where variations in rainfall type (convective versus stratiform) are taken into account. The analysis is first performed using $Z$ and $R$ data pairs derived directly from disdrometer drop size distribution measurements, and then repeated using WSR-88D radar reflectivity data. The degree of sensitivity in runoff simulations to temporal variations in $Z–R$ relationships depends largely on the method used to derive the parameters of these relationships. Using event-specific $Z–R$ relationships results in accurate hydrographs when the parameters are derived based on bias removal and minimization of random errors of rainfall estimates. Methods based on least-squares fitting require refining the derivation of $Z–R$ parameters down to a subevent scale, which is not practically feasible. A simple and practical method based on selection of a climatologically representative exponent of the $Z–R$ relationships and adjusting the multiplier coefficient through bias removal still results in reasonably accurate runoff simulations, but only when event-specific $Z–R$ relationships are used.