Comparing Statistical and Semi-Distributed Rainfall–Runoff Models for a Large Subtropical Watershed: A Case Study of Jiulong River Catchment, China

Abstract

In this contribution, the authors present their preliminary investigations into modeling the rainfall–runoff generation relation in a large subtropical catchment (Jiulong River catchment) on the southeast coast of China. Previous studies have mostly focused on modeling the streamflow and water quality of its small rural subcatchments. However, daily runoff on the scale of the whole catchment has not been modeled before, and hourly runoff data are desirable for some oceanographic applications. Three methods are proposed for modeling streamflow using rainfall outputted by the Weather Research Forecast (WRF) model, calculated potential evaporation (PET), and land cover type: (i) a ridge regression model; (ii) NPRED-KNN: a nonparametric k-nearest neighbor model (KNN) employing a parameter selection method (NPRED) based on partial information coefficient; (iii) the Hydrological Simulation Program-Fortran (HSPF) model with an hourly time step. Results show that the NPRED-KNN approach is the most unsuitable method of those tested. The HSPF model was manually calibrated, and ridge regression performs no worse than HSPF based on daily verification, whilst HSPF can produce realist daily flow time series, of which ridge regression is incapable. The HSPF model seems less prone to systematic underprediction when replicating monthly-annual water balance, and it successfully replicates the baseflow index (the flow intensity) of the Jiulong River catchment system.

Integration of Advanced Soft Computing Techniques in Hydrological Predictions

“…This special edition of Atmosphere is tailored to fill the existing gap by including papers on advances in the contemporary use of soft computing techniques in hydrological modelling.This Atmosphere Special Issue collected five original papers focused on research associated with the integration of advanced soft computing techniques in hydrological predictions. Han et al [8] of Xiamen University and the University of New South Wales presented three models, including a nonparametric k-nearest neighbor model, which employs a parameter selection method based on partial information coefficient to simulate the rainfall-runoff generation relationship in the Jiulong River catchment, China. Tayyab et al [9] of China Three Gorges University and the Huazhong University of Science and Technology developed a novel hybrid artificial neural networks model based on ensemble empirical mode decomposition and discrete wavelet transform to predict riverflow, validating its efficiency at the Upper Indus Basin.…”