Understanding Precipitation Fidelity in Hydrological Modeling

Abstract

The objective of this research is to develop and demonstrate a methodology to specifically assess the interrelationships among estimated precipitation, observed streamflow, and hydrologic model performance. To satisfy this objective, this work introduces a new concept called “precipitation fidelity,” which is the correspondence of stream outflow to the estimated precipitation used as input into a hydrologic model. Simple annual and daily precipitation fidelity indexes are defined. The use of the precipitation fidelity indexes is then demonstrated for the Rivanna watershed in central Virginia modeled using an existing hydrologic model, the Chesapeake Bay Program Phase 5 watershed model, and the associated precipitation input data set. The precipitation fidelity results are used in conjunction with model output to identify the effect of precipitation estimation accuracy on model performance at both long and short time scales. Based on the daily precipitation fidelity measure, fidelity between the precipitation input and the observed streamflows was not observed for approximately one quarter of the days in the headwater watersheds. Days for which the estimated input precipitation has runoff-generating rainfall, but the observed stream discharge does not increase, have the highest average relative daily modeling errors and high area-weighted daily modeling errors. These results indicate that precipitation needs to be better represented in the headwater subwatersheds. Regression analysis using the analysis of covariance method was used to determine the statistical similarity between annual estimated precipitation and observed and modeled streamflows. Regression results suggested that direct hydrology calibration of the subwatershed of interest leads to both a higher level of correspondence between the estimated precipitation and modeled flows and an acceptable goodness-of-fit between the modeled and observed data.