Influence Factors The Performance Of Artificial Neural Network Model For Flood Forecast: Case Study Chiangmai Municipality

Abstract

This research is about the use of rainfall grid data from the WRF-ECHAM 5 model with grid size 20 x 20 kilometers for flood forecasting at Chiang Mai Municipality by artificial neural network model. If the rainfall grid could be reduced to 10x10 kilometers, the potential of model for flood forecasting would be increased, so to reduce the rainfall grid, two interpolation techniques were used.- IDW (Inverse Distance Weight) and Kriging,. Moreover, the data of water level at Mae-Ngad Somboon Chon Dam is included as input variables. Also, finding the suitable architecture structure of model comparing the number of 1 and 2 hidden layers (learning algorithm is Levenberg-Marquardt (LM) with number of hidden nodes 1 Node, 50% and 100% of input variables. Flood events between 1994-2006 are used in this study. There are three experiments; A (input variable is rainfall grid data 10x10 Km from IDW and Kriging), B (input variables are rainfall grid data 10x10 Km from IDW and Kriging and water level data from Dam) and C (input variables are rainfall grid data 20x20 Km and water level data from Dam). The result found that experiment A with IDW rainfall grid 10x10 Km, 1 hidden layer with number of hidden nodes same with number of input variables (100%) has the most effect with artificial neural model with 0 - (-0.6) meters error.

Keywords

• interpolation
• artificial neural network
• flood forecast
• wrf-echam5
• chiang mai