A comparative analysis of precipitation estimationmethods for streamflow prediction

Abstract

Surface hydrologic models are widely used for streamflow prediction, forecasting and for understanding hydrologic processes. They are also an important tool for contributing to the resolution of wider resource and environmental issues, providing information to support policies and decisions for water resource management. Precipitation is a key input to hydrologic models and is however also the major source of predictive uncertainty. Whilst station-based observed precipitation data can be adequate for hydrologic modelling in small catchments, they may not be sufficient for large catchments, in particular for large catchments with a mountainous terrain. Areal estimation of precipitation is a potential option to provide more precise precipitation input to models for large catchments. Conventionally, for areal precipitation estimation, station-based precipitation data are interpolated across the model domain using various methods, including Spline fitting, Inverse Distance Weighting (IDW) and the classical Thiessen Polygon, which are among the more popular and commonly used methods. Different precipitation interpolation methods will affect the spatial and temporal variability of areal precipitation inputs, resulting in different uncertainties when used to help calibrate a surface hydrologic model. This paper investigates the effect of the above three types of precipitation interpolation methods (ANUSPLIN surface, IDW surface and Thiessen polygon) on streamflow predictions. The Chaohe basin located in northern China is selected as the study area. It is an important headwater of the Miyun Reservoir which provides drinking water to Beijing and surrounding townships. Three lumped, surface hydrologic models (GR4J, IHACRES and Sacramento) are selected to study the accuracy and predictive uncertainty of these three types of precipitation interpolation on daily streamflow. The models were calibrated separately using discharge observations from three gauges in the basin. The results show that the ANUSPLIN surface interpolation performs the best overall under various combinations of conditions. The IDW surface also performs well in the upper and middle basin but the Thiessen polygon is inferior to the other two methods. The comparison of the three hydrologic models shows that IHACRES and Sacramento perform better than GR4J. The best combination is areal rainfall estimated using the ANUSPLIN derived surface with the IHACRES model in the case study catchments, though the Sacramento model is a close second. Show more