In most large-scale distributed hydrological models for flood forecasting, river cross-sections are approximated with rectangles or trapezoids, due to the amount of manual works needed to introduce a large database of surveyed river cross-sections into the models. In this study, we firstly investigated the difficulties for introducing surveyed river cross-sections into the Rainfall-Runoff-Inundation model, a distributed hydrological model, and then we proposed an algorithm for introducing the database of river cross-sections surveyed by the Ministry of Land, Infrastructure, Transport and Tourism (MLIT), into the whole-Japan Rainfall-Runoff-Inundation model (JRRI model). With our proposed algorithm, we introduced in total 26032 cross-sections for the length of 7734.7 km, or 72.9 % of the rivers managed by MLIT, into JRRI model. Secondly, we investigated the effect of cross-section introduction on the accuracy of water level change predictions, through the comparative experiment of the heavy rainfall in the western part of Japan in 2018, with JRRI model with and without surveyed river cross-sections. The metrics of accuracy evaluation (NSE, KGE, RMSE, peak water level ratio, peak water level difference) are improved in the experiment with JRRI model with surveyed river cross-sections, suggesting that the negative baseline biases between observations and experiment outcomes have been eliminated. Especially, average and standard deviation of peak water level differences have been reduced from −2.04 m ± 1.70 m to 0.14 m ± 0.88 m. Overall, this study indicates that introduction of mass database of surveyed river cross-sections into large-scale distributed hydrological models can improve the accuracy of water level predictions for wide regions, and our proposed algorithm may serve it by allowing the introduction of mass database of cross-sections.
View full abstract