- 期刊
應用適應性模糊推論系統與自組性演算法評估未來降雨之不確定性
APPLYING ADAPTIVE NETWORK-BASED FUZZY INFERENCE SYSTEMS AND GROUP METHODS OF DATA HANDLING TO ESTIMATE FUTURE RAINFALL UNCERTAINTY
摘要
本研究比較適應性網路模糊推論系統（Adaptive Network-Based Fuzzy Inference System, ANFIS）與自組性演算法（Group Method of Data Handling, GMDH）啟發式自組織架構（Heuristic Self-Organization）之特性，應用於統計降尺度之效能。研究案例選用大致同緯度分別位於中央山脈兩側之西側臺中與東側花蓮測站，探討中央山脈兩側臺中與花蓮未來降雨趨勢與不確定性。以ANFIS與GMDH為架構，利用四種IPCC所提供之AR4全球環流模式（General Circulation Model, GCM）歷史與未來情境（A1B）資料、臺中與花蓮測站歷史氣象紀錄，發展ASDM（ANFIS Statistical Downscaling Model）與GSDM（GMDH Statistical Downscaling Model），評估未來情境中期（2050年∼2069年）與長期（2080年∼2099年）之降雨趨勢與不確定性。本研究利用四種GCM模式歷史情境與測站月雨量，以均方根誤差（Root Mean Square Error, RMSE）做為ASDM與GSDM模式建立評估指標，使用四種GCM模式未來A1B情境資料推估未來月降雨量。模擬結果顯示：1.臺中與花蓮測站用於建立模式之測試資料模擬結果均以GSDM模式較佳。2.臺中與花蓮之平均降雨變化率以冬季較為明顯。臺中以ASDM較GSDM降雨變化幅度明顯，花蓮則反之。3.預測中、長期降雨而言，臺中測站以GSDM模式預測效能較佳，花蓮測站則以ASDM模式為佳。4.未來中、長期臺中與花蓮枯水期雨量增加率相對高於豐水期，而豐水期降雨量可能面臨缺水之風險相對提高。5.臺中及花蓮測站中、長期均以ASDM之降雨變異性較高，顯示GSDM模式具有自我組織與優化模式參數之演算法，表現優於ASDM模式。
並列摘要
This study compares the performance of downscaling models developed based on adaptive network-based fuzzy inference systems (ANFIS) and group method of data handling (GMDH) with heuristic self-organization. The Taichung and Hualien stations, which are located on the two sides of the Central Mountains in Taiwan, are chosen as case studies to demonstrate the approach and investigate the future trends and uncertainty of rainfall of the two stations. The ASDM (ANFIS-based Statistical Downscaling Model) and GSDM (GMDH-based Statistical Downscaling Model) are developed based on ANFIS and GMDH. Four GCM historical and A1B scenarios provided by IPCC and historical monthly rainfall of Taichung and Hualien are employed to build the two models. The A1B scenarios of four GCM models are applied to estimate the trends and uncertainty of monthly rainfall of the two stations in the future mid-term (2050-2069) and long-term (2080-2099). The ASDM and GSDM models are built based on the root mean square (RMSE) in model training and testing phases. The simulated results show: 1. The GSDM model exhibits better results both in the Taichung and Hualien stations. 2. The average rainfall variation rate between Taichung and Hualien is more obvious in winter. The ASDM in Taichung has a more pronounced rainfall variation than the GSDM, while the opposite is true for Hualien. 3. Regarding the prediction of medium and long-term rainfall, the GSDM model at the Taichung station has better prediction performance, while the ASDM model at the Hualien station has better prediction results. 4. The rate of increase in rainfall in the dry period is relatively higher than in the wet period both in Taichung and Hualien; the risk of shortage of rainfall in the wet period is probably to be higher. Finally, the analysis of rainfall variability between the Taichung and Hualien stations shows that the variability of rainfall in the ASDM is higher than the GSDM in the mid and long-term. This demonstrates that the GSDM with the ability of heuristic self-organization is better than the ASDM.