Abstract:
This research on Artificial Neural Network and Influence of Modeling on Flood Risk Management in Chiang Mai Municipality has 2 objectives. First, it is to study influence factors the performance of ANN model for predicting water level at P.1 station, using daily rainfall data from WRF-ECHAM 5 model with cell size 20*20 Kilometer and downscaled rainfall data of 10*10 Kilometer sized cells (Interpolation method downscaling: IDW and Kriging) and water level data from Mae Ngad Sombunchon dam, while setting preferred architecture structure of model as LM learning algorithm, 1 and 2 hidden layers and 1, 50% and node. And secondly, it is to analyze proposed product of the ANN models acceptance and usefulness in long- term flood risk management among selected responsible governmental agencies and communities. The questionnaires and interview guidelines were used together in in-depth interview, which is divided into 2 parts. The prior part contains questions that focus on designing flood management measures while observing usefulness of flood risk maps produced from the model compared with the risk map from historical data. The sample group can choose to use maps A, conventional map of flood occurrence, and/or B, predicted 30 years future flood risk map by the model for designing long-term flood coping measures. Then, the later part of the questions focuses on scoring recognition level of the benefits of the technology. All answers are analyzed together in revealing the consistency of the technology acceptance level and the factors behind it using Technology Acceptance Model or TAM as analytical framework. The result yields most appropriate input data to be IDW rainfall data grid 10*10 kilometer, 1 hidden layer with the performance to prediction of ANN model more than 2 hidden layer, and hidden node n node with the highest influence on the accuracy of the artificial neural network model, compared to the model that uses imported rainfall data for grid sizes of 20 * 20 kilometers and models that add water level data from dams together with grid size rainfall data. * 10 kilometers. The result though has an error of event with 0-(-0.6) when the error value is used to adjust the forecast value in the future for 30 years. The result of the prediction in the further reveals 4 floods events (on the year 2042, 2047, 2052 and 2059) during the 30 years span, confirming the influence of climate change on the number of rainy days and rain flood intensity. As for the second research objective, it was found that the acceptance level of most predicted flood risk users have the acceptance value of 3 ( of fairly high level.) Factors analysis points out that the acceptance at each level is related to the characteristics of work roles and work experience of the users in flood management more than other assumed factors under TAM. Furthermore, it is found that the administrative agencies that have a role in flood emergency management (working closely with communities) are not likely to easily accept the results of forecasted future risk from the model for flood management planning. On the other hand, from the point of view of the users that focus on legal planning and policies designing in the longer timeframe, the predicted future flood risk map is accepted and consequently chosen for implementation. Analysis from assessed levels of model acceptance in relation to flood coping measures further shows that the risk map B, forecasted by the model, are used in mostly policy type of flood coping measures than in structural measures.