In this study, a series of experiments were performed in a laboratory flume
with a medium-packed bed. Surface runoff was controlled to pass at various
flow rates, velocities, and runoff depths over the medium bed. Runoff samp
les were taken at the end of the flume, and the concentration of potassium
chloride was analyzed. The relationships between the controlled input varia
bles and the affected output variables was modeled using artificial neural
networks (ANN). Many different ANN-architectures were investigated and this
work shows that an optimum architecture with minimum RMS error at five hid
den nodes was observed. (C) 1998 IAWQ Published by Elsevier Science Ltd. Al
l rights reserved.