การคาดการณ์พื้นที่เสี่ยงต่อการเกิดดินถล่มในจังหวัดอุตรดิตถ์โดยประยุกต์ใช้เทคโนโลยีภูมิสารสนเทศร่วมกับโครงข่ายประสาทเทียม

Abstract

The study had 2 main objectives: to develop the demonstrative model of Artificial Neural Network (ANN) which suitable for the predictions landslide susceptibility areas in Uttaradit Province and to apply Geo-Informatics technology and ANN in predicting the susceptibility areas. The research methodology included 1) The classification of selected landslide areas by satellite imagery 2) The analysis of affective factors of landslides 3) The development of the fundamental ANN model by comparison learning algorithms between LM and BR and the testing of the number of hidden nodes from 1-2n+1 4) The further development of ANN model, which will be expanded in 5 factors; (a) the comparison between real number and rang number of the factors (slope, elevation, rainfall, land use, watershed classification, distance from water and distance from faults), (b) the prescription of the break-point of slope factor 16.70 degrees, (c) the comparison of input factor between 5 factors and 7 factors , (d) the comparisons of the testing results both pre and post corrections of grid size from the satellite data and (e) the comparison of the learning algorithms between LM and BR and number of hidden nodes 5) Use the best results of studied model to predict the landslides susceptibility areas. The study has shown that the most affective factors to landslides consist of slope, elevation, rainfall, distance from water and distance from faults. The development of the fundamental ANN has found that learning algorithm LM regarding to the 2 hidden nodes, is the most appropriate method to predict the landslides susceptibility areas with the least error of statistic value. Moreover, the development of ANN models has also found that the most effective factors are included of, the testing process of real number, the experimental of designated the break-point of landslide 16.70 degrees, the testing process by using 5 factors and the testing after resolving grid size from the satellite data. However, the comparison learning algorithms between LM and BR has shown that the LM is more effective than BR and the most effective number of hidden nodes is equal to the number of input variables. The results of landslide prediction has shown that the most susceptibility areas cover the north of Lap Lae and Muang district, and some parts of Thapla district where the devastated landslide had occurred in 2006.