台灣為海島型國家,雨量充沛,但因特殊地理條件常造成缺水問題,伴隨水庫區內非點源污染嚴重,對庫區之水資源運用造成嚴重的影響。因此若能建立預測水庫水體的模式,提供我們模擬預測水庫水質之變化,便能提供地方單位做決策。本研究使用倒遞類神經網路探討降雨與水庫水體水質之相關性,進而以降雨資料預測水庫水體水質,所選擇預測之水質項目為卡爾森指標中的總磷、透明度、葉綠素,並以當日降雨量,前五日平均降雨量,月平均降雨量進行水庫水體水質之預測。本研究結果顯示,月平均降雨與水庫水體水質存在較高之相關性,換言之,採用月平均降雨量來預測水庫水體水質能力佳,其中又以總磷、葉綠素a之預測結果較好。
Reservoir water quality protection is an important work in water resource management. Meteorological and hydrological properties can influence the water quality in a reservoir. This study applied the Back-Propagation Neural Network (BPNN) to predict reservoir water quality. According to rainfall data, the BPNN can learn and describe the relationship between rainfall data and reservoir water quality. The results show that the prediction efficiency of reservoir water quality is high according to average monthly precipitation. However, the prediction efficiency of reservoir water quality is poor according to the average precipitation for the five days prior to the water quality monitoring date or the precipitation on the water quality monitoring date.