적설량 예측을 위한 신경망 및 다중회귀 모형의 적용성 평가

Abstract

Since snowfall is related to various meteorological variables such as temperature and precipitation, it is generated in nonlinear manner. Therefore this study constructs snowfall forecasting model using neural networks and multiple regression which can consider nonlinear process of snowfall. The study constructs the forecasting models for each station using temperature, precipitation, and snowfall depth observed from starting time of observation to 1999. And snowfalls are calculated for all stations by using temperature and precipitation in the period of 2000 to 2011. From the statistical analysis of the calculated snowfall, the proper model is selected. The selected models show the correlation coefficients of 0.700 to 0.949 and the adjusted determination coefficients of 41.7% to 89.8%. The applicability of neural network models is superior to other model at almost every station. But in some cases multiple regression models show better results than neural network models due to the lack of observational data during learning period and the extreme peak values which are not learned during forecasting period. According to the study, the results of the models confirm the predicting snowfall depth by using temperature and precipitation is possible and show neural network model is better than the existing statistical models.