Runoff analysis in humid forest catchment with artificial neural network

Hydrometeorological data, i.e. meteorological, water discharge and moisture content data have been collected over the past 10 years in the Tone area o f central Japan. By analyzing soil moisture data and by making inferences f rom field studies, possible factors influencing stream discharge are explor ed. The soil moisture data obtained from 40-cm depth carry the integrated e ffect of the upstream catchment area and are important for estimating strea m discharge. Vertical infiltration is important in the upper 2D-cm, due to the high hydraulic conductivity of this part of forested soil. However, lat eral flow through this layer becomes dominant during very high rainfall and /or following a long succession of rainfall events, resulting in rapid thro ughflow. A new type of artificial neural network (ANN) model based on a bac k propagation algorithm is formulated using the analyses. The formulated AN N model makes use of soil moisture data in estimating stream runoff and may be considered useful as an aid to catchment monitoring. (C) 2000 Elsevier Science B.V. All rights reserved.