Flood forecasting using a fully distributed model: application of the TOPKAPI model to the Upper Xixian Catchment
- 1Bureau of Hydrology, Ministry of Water Resources, 2 Lane 2, Baiguang Road, Beijing 100053, P.R. China
- 2Department of Geo-Environmental Sciences, University of Bologna, Via Zamboni 67, 40126, Bologna, Italy
- Email for corresponding author: firstname.lastname@example.org
Abstract. TOPKAPI is a physically-based, fully distributed hydrological model with a simple and parsimonious parameterisation. The original TOPKAPI is structured around five modules that represent evapotranspiration, snowmelt, soil water, surface water and channel water, respectively. Percolation to deep soil layers was ignored in the old version of the TOPKAPI model since it was not important in the basins to which the model was originally applied. Based on published literature, this study developed a new version of the TOPKAPI model, in which the new modules of interception, infiltration, percolation, groundwater flow and lake/reservoir routing are included. This paper presents an application study that makes a first attempt to derive information from public domains through the internet on the topography, soil and land use types for a case study Chinese catchment - the Upper Xixian catchment in Huaihe River with an area of about 10000 km2, and apply a new version of TOPKAPI to the catchment for flood simulation. A model parameter value adjustment was performed using six months of the 1998 dataset. Calibration did not use a curve fitting process, but was chiefly based upon moderate variations of parameter values from those estimated on physical grounds, as is common in traditional calibration. The hydrometeorological dataset of 2002 was then used to validate the model, both against the outlet discharge as well as at an internal gauging station. Finally, to complete the model performance analysis, parameter uncertainty and its effects on predictive uncertainty were also assessed by estimating a posterior parameter probability density via Bayesian inference.