Once the ANN topology was determined, we tested
our hypothesis of extension from [4] that, if monthly
rainfall (P9) could be included in the factors, augmenting
the model with it would increase the predicting
accuracy. As references, MLP and RBF model
were trained with P1 - P8 data from 2011, 2010 and
2006. Except in 2006 where both models did equally
well, the MLP could pre-dicted the flood better than
the RBF, with the accuracies of 68.8, 77.8 and 81.1
percents, respectively compared to those by using
RBF of 63.4, 73.9 and 81.1 percents (Fig 10). When
including P9 into the learning of both models, the
predictability increased as anticipated for 2011 and
2010, yet slightly dropped in 2006.