MAPPING STREAMFLOW CHARACTERISTICS IN THE MOST UPSTREAM BASINS THROUGHOUT JAPAN USING ARTIFICIAL NEURAL NETWORKS

Abstract

We developed and validated artificial neural networks (ANNs) to map the streamflow characteristics in the most upstream basins throughout Japan. The ANNs output mean annual runoff height (QMEAN) and percentiles of daily streamflow, including nine different groups, by inputting basin characteristics, including climate, land use, soils, geology, and topography. The generalization performances of the ANNs showed R 2 = 0.70 in the QMEAN and R 2 = 0.20 – 0.74 in the streamflow percentiles. We succeeded in mapping the streamflow characteristics in the most upstream basins throughout Japan, which reflected the rainfall and snowfall characteristics in the country. The streamflow characteristic maps revealed that devel-oping run-of-river hydropower stations in heavy snowfall areas, such as the Tohoku and Hokuriku regions facing the Sea of Japan, is suitable.