River network based characterization of errors in remotely sensed rainfall products in hydrological applications

ABSTRACT

The authors propose a hydrologic evaluation framework for gridded rainfall products. This framework makes use of the Spatial Stream Network (SSN) statistical method to provide spatial characterization of the discrepancies between two gridded rainfall products. The SSN method relies on using stream network length rather than the traditionally used Euclidean distances.It also accounts for the flow connectivity information between the network segments. This concept is relevant in hydrological modeling since rivers transport accumulated precipitation that occurred over different parts of the basins, and stream networks do not represent Euclidean space. To demonstrate, we used this framework to compare the satellite rainfall product called Integrated Multi-satellitE Retrievals for GPM (IMERG) with the ground-based Multi-Radar/Multi-Sensor (MRMS) rainfall product. The results show that the magnitudes of the rainfall discrepancies tend to decrease as rainfall accumulates in the downstream direction. However, the covariance range between these discrepancies is much larger along flow-connected stream network segments than in flow-unconnected stream segments. This in turn could have an effect on the error correlation of the predicted discharges. In addition, the spatial linear models of rainfall errors improved significantly with SSN based models in comparison to pure Euclidean separation distance models.

Acknowledgments

We gratefully acknowledge the financial support from the Iowa Flood Center (IFC) towards the completion of this study. The second author was also supported by the Rose & Joseph Summers endowment.