A Two-Stage Approach Integrating SOM- and MOGA-SVM-Based Algorithms to Forecast Spatial-temporal Groundwater Level with Meteorological Factors

To obtain accurate and effective forecasts of groundwater level, a two-stage approach integrating Self-Organizing Maps (SOM-), Multi-Objective Genetic Algorithm and Support Vector Machine (MOGA-SVM-based) algorithms is developed herein using the optimal input combinations of meteorological factors in a complex spatial-temporal groundwater system. In the first stage, an SOM-based clustering method is used to separate distinct and meaningful spatial groundwater zones. In the second stage, a temporal analysis model integrating MOGA with SVM is developed to identify the optimal input combinations. An actual application is conducted using the Choushui River Alluvial Fan in Taiwan as the case study; it currently has over-pumping and land subsidence problems. The MOGA-SVM model is compared with an existing model based on the SVM to demonstrate the superiority of the proposed approach. Moreover, the effective meteorological factors in different spatial zones can be determined by using the proposed approach to show the spatial characteristics, and these factors can significantly improve the forecasting performance, especially for long lead-time forecasting. In conclusion, the proposed spatial-temporal approach is applicable to a huge and complex groundwater system; it provides an alternative to the existing model for water resources management problems.