Multi-objective calibration of Xinanjiang model by using streamflow and evapotranspiration data

Abstract

Hydrological model calibration is normally carried out against a single set of observations, particularly streamflow, which often limit overall model performance. Compared to that, multi-objective model calibration can overcome these problems and result in better model performance. In this study, daily streamflow data and catchment evapotranspiration (ET) estimates in 210 catchments in southeastern Australia, were used for multi-objective calibration of Xinanjiang model. The daily ET was estimated from the state-of-the-art Penman-Monteith-Leuning (PML) model together with remotely sensed time-series vegetation data due to unavailability of actual ET data in most catchments. The results show that the PML ET estimates compare well with the flux measurements, and it is regarded as ‘ground truth’ and used for model calibration. Two calibration schemes are used to evaluate hydrological modelling performance, including single-objective calibration against streamflow data alone (Scheme 1) and multi-objective calibration against streamflow and evapotranspiration (Scheme 2). For model calibration, the median of NSE of daily streamflow for the 210 catchments are 0.78and 0.76 for Scheme 1and Scheme 2, respectively, while the median of R2between observed and simulated daily ET are 0.79and 0.88. Model regionalization results are overall consistent with the calibration results, with Scheme 2 showing a less degradation for streamflow and ET predictions compared to Scheme 1. The medians of NSE for Scheme 1 and Scheme 2 are 0.51and 0.52, respectively, while the median of R2 are 0.68and 0.79. The results show that the multi-objective calibration reduced the uncertainty of parameters and can improve runoff predictions in ungauged catchments. More researches should be carried out for further reducing the parameter uncertainty.