In the present study, two artificial neural networks were developed to simulate outflow hydrograph
from earthen dam breach. The required data for the modelling were collected from literature,
laboratory experiments and a physically based model (i.e. BREACH). For the laboratory modelling, five
different materials were used for the construction of different dams of various sizes, and the process
of the breach was recorded by two video cameras to record the breach growth as well as the output
hydrograph. The genetic algorithm was also applied to divide the data into three statistically similar
sub-sets for training, validation and test purposes. The obtained results demonstrate that the results
of the artificial neural network (ANN) method are in good agreement with the observed values, and
this method produces better results than existing classical methods. Also, the experiments show
when cohesive strength is larger, the breach process becomes slower, and the peak outflow and the
final width and depth of breach become smaller. Moreover, when the friction angle is larger, the breach
process becomes slower, and the peak outflow and the final width and depth of breach become
smaller. However, the rate of breach formation is particularly dependent upon the soil properties.