Resolution Requirements for the Spatial Sampling of Rainfall

Abstract

Spatial rainfall information is used for a wide variety of applications. The resolution at which this information is sampled can have a significant impact on its usefulness to these applications. This research is directed towards assessing the magnitude of this impact. Using a high resolution radar dataset of three months’ duration, rainfall is degraded in spatial and temporal resolution to a range of spatial and temporal scales so that— from a comparison with the original high resolution data—the error arising from lower resolution sampling can be quantified. On average, across all significant events in the dataset, this sampling error (expressed in RMSE normalized to the mean average rainfall accumulation) was observed to be approximately 0.2 for 2000 m/300 s sampling. A relationship was found between the mean characteristic length scale of an event and spatial sampling error magnitude, allowing prediction of sampling error using only low resolution data. By simulating a rain gauge network using the same dataset, the contribution that sampling error makes to the total radar/rain gauge bias was estimated to be, at a minimum, approximately 30%. Gauge correction of radar data in light of these sampling errors was also assessed and a correction scheme that would minimize the impact of sampling errors was proposed. Additionally, the impact of sampling error on a distributed rainfall runoff model (GSSHA) was examined. The magnitude of this error was fairly substantial (a variation of 14–30% in peak flow for conventional radar resolutions) and was found to be greater for less saturated initial soil moisture conditions.