Abstract:
Severe flood is one of major problems in Thailand. When it happens, damage to human life and their
properties can occur. One of the important measures for managing flood is the forecast of flood water
levels. At presents, a number of flood forecasting models have been developed. The Adaptive NeuroFuzzy
Inference System ANFlS is among some of the interesting models.
In this study, ANFIS model was used to forecast water levels in two different watersheds including I)
Yom watershed with small and steep basin and 2) ChaoPhraya watershed with large and flat basin. Water
levels in these watersheds were forecasted in advances, for I, 2, and 3 days. In addition, ANFIS was tested
in the case of incomplete data of the ChaoPhraya watershed. In the same day, the data from nearby stations
were filled-up and used for predicting the missing data of water levels. The results of water level computed
by ANFIS were compared with the results from an Artificial Neuron Networks (ANN).
It was found that the ANFIS has a capability for forecasting water levels in both areas even though these
basins have different characteristics in nature. When ANFIS was compared with ANN, it was found that
the results of water level predictions were quite the same. However, the results of highest and lowest water
levels computed by ANFIS were found agree we)) with the observed ones and were better than those
computed by ANN.