台灣水庫的污染以集水區非點源為主，其中包含大量的營養鹽類及懸浮固體物，對水資源運用造成嚴重的影響。台灣氣候條件特殊，少數幾場豪大雨便包含大部分的非點源負荷輸出，污染集中再加上現場採樣不易，使得相關數據十分欠缺，造成暴雨非點源污染估計上極大的誤差。
本研究於八掌溪觸口橋處，採用離槽式自動連續監測，收集單場暴雨的逕流水質，包括濁度與硝酸鹽。建立完整的雨量、逕流及水質資料庫，再配合類神經網路(Artificial Neural Network, ANN)做模擬與預測。所模擬的結果可與ISCO Inc. 6200自動採樣器結合，將傳統經驗採樣方式提升為預測採樣方式，並評估在單場暴雨中溶解性與非溶解性污染物所產生的污染量與濃度變化。
結果顯示類神經網路，無論在逕流或是水質上能提供非常好的模擬結果。水位預測模式經過訓練與測試之後，預測未來3小時的變化，其平均判定係數R2為0.918，均方根誤差RMSE為0.319m；濁度預測模式，其測試結果在3小時後的R2為0.684，RMSE為222NTU。硝酸氮預測模式，其測試結果在3小時後的R2為0.885，RMSE為0.05ppm。SS負荷推估部分，3小時後的模擬值與實際觀測值做比較，在2006年4月27日的降雨場次上，僅低估1.34％，5月2日的降雨場次約低估2.42％。顯示以類神經模擬未來3小時的SS負荷，有良好的預測表現。
另外在應用類神經於預測採樣方法上，所推估的懸浮滓負荷，皆高於傳統採樣方式或Flow-stratified採樣法，與實際負荷量較為接近。因以類神經模式可模擬未來負荷變化趨勢，可事先安排採樣程序，故所推估出來的污染負荷與傳統採樣方式相比也會較接近實際值。

Non-point source pollution is the major pollution in Taiwan reservoirs’ catchment. It includes a large mount of nutrients and suspended solids which cause serious effects of water resources management. Because of special climate in Taiwan, most of non-point source pollution load occur in few rainstorm events. Monitoring data of the rainstorms will generate significantly error because of hard sampling and the related data are extremely deficient.
This study adopts off-site sampling and continuous monitoring in Bajhang River to collect runoff and water quality data in the rainstorms, include turbidity and nitrate, to establish integrated rainfall, runoff and water quality database. Then apply Artificial Neural Network (ANN) to simulate runoff and water quality information to forecast changing in the future. The results of simulation can combine with ISCO Inc. 6200 auto-sampler to promote traditionally experiential sampling approach to forecasted sampling approach. Therefore, we can accurately evaluate the change of dissolvable and undissolvable pollutants phenomena in the rainstorm events.
The first results show that ANN is capable for simulating runoff and water quality. After we training and testing water level forecasted model, the average coefficient of determination (R2) of water level at three hour later is 0.918, the root-mean square error (RMSE) is 0.319m. The average R2 of turbidity at three hour later is 0.684, RMSE is 222NTU. The average R2 of nitrate at three hour later is 0.885, RMSE is 0.05ppm. The comparison results of simulated and observed suspended sediment loads at three hour later is only underestimating 1.34% in 2006/4/27 rainfall event and 2.42% in 5/2 rainfall event. It shows that ANN can provide very well performance for simulating suspended sediment loads at three hour later.
The second results show that the simulated suspended sediment loads of ANN forecasted sampling method is higher then traditional sampling method or Flow-stratified method, is close to real suspended sediment loads. Because ANN model can simulate sediment loads variation in the future, we can make a sampling arrangement in advance so that estimated pollution loads will be close to real loads.

1. Abrahart, R. J., & White, S. M. (2001). Modelling sediment transfer in Malawi: Comparing backpropagation neural network solutions against a multiple linear regression benchmark using small data sets. Physics and Chemistry of the Earth Part B-Hydrology Oceans and Atmosphere, 26(1), 19-24.
2. Alp, M., & Cigizoglu, H. K. (2007). Suspended sediment load simulation by two artificial neural network methods using hydrometeorological data. Environmental Modelling & Software, 22(1), 2-13.
3. Anctil, F., Michel, C., Perrin, C., & Andreassian, V. (2004). A soil moisture index as an auxiliary ANN input for stream flow forecasting. Journal of Hydrology, 286(1-4), 155-167.
4. Anctil, F., Perrin, C., & Andreassian, V. (2003). Ann output updating of lumped conceptual rainfall/runoff forecasting models. Journal of the American Water Resources Association, 39(5), 1269-1279.
5. Brion, G. M., & Lingireddy, S. (1999). A neural network approach to identifying non-point sources of microbial contamination. Water Research, 33(14), 3099-3106.
6. Celikoglu, H. B., & Cigizoglu, H. K. (2007a). Modelling public transport trips by radial basis function neural networks. Mathematical and Computer Modelling, 45(3-4), 480-489.
7. Celikoglu, H. B., & Cigizoglu, H. K. (2007b). Public transportation trip flow modeling with generalized regression neural networks. Advances in Engineering Software, 38(2), 71-79.
8. Chang, F. J., Liang, J. M., & Chen, Y. C. (2001). Flood forecasting using radial basis function neural networks. Ieee Transactions on Systems Man and Cybernetics Part C-Applications and Reviews, 31(4), 530-535.
9. Cigizoglu, H. K. (2004). Estimation and forecasting of daily suspended sediment data by multi-layer perceptrons. Advances in Water Resources, 27(2), 185-195.
10. Cigizoglu, H. K., & Alp, M. (2006). Generalized regression neural network in modelling river sediment yield. Advances in Engineering Software, 37(2), 63-68.
11. Cover, T. M. (1965). Geometrical and statistical properties of systems of linear inequalities with applications in pattern recognition. Ieee Transactions on Electronic Computers, EC14(3), 326-334.
12. de Oliveira, K. A., Vannucci, A., & da Silva, E. C. (2000). Using artificial neural networks to forecast chaotic time series. Physica A, 284(1-4), 393-404.
13. Dibike, Y. B., & Solomatine, D. P. (2001). River flow forecasting using artificial neural networks. Physics and Chemistry of the Earth Part B-Hydrology Oceans and Atmosphere, 26(1), 1-7.
14. El-Din, A. G., & Smith, D. W. (2002). A neural network model to predict the wastewater inflow incorporating rainfall events. Water Research, 36(5), 1115-1126.
15. Jain, S. K. (2001). Development of integrated sediment rating curves using ANNs. Journal of Hydraulic Engineering-Asce, 127(1), 30-37.
16. Kaluli, J. W., Madramootoo, C. A., & Djebbar, Y. (1998). Modeling nitrate leaching using neural networks. Water Science and Technology, 38(7), 127-134.
17. Kerem Cigizoglu, H., & Kisi, O. (2006). Methods to improve the neural network performance in suspended sediment estimation. Journal of Hydrology, 317(3-4), 221-238.
18. Lawler, D. M., Petts, G. E., Foster, I. D. L., & Harper, S. (2006). Turbidity dynamics during spring storm events in an urban headwater river system: The Upper Tame, West Midlands, UK. Science of the Total Environment, 360(1-3), 109-126.
19. Leave, C. (1992). Neural networks - algorithms, applications and programming techniques - Freeman,JA, Skapura,DM. Journal of the Operational Research Society, 43(11), 1106-1106.
20. Lek, S., Guiresse, M., & Giraudel, J. L. (1999). Predicting stream nitrogen concentration from watershed features using neural networks. Water Research, 33(16), 3469-3478.
21. Lewis, J. (1996). Turbidity-controlled suspended sediment sampling for runoff-event load estimation. Water Resources Research, 32(7), 2299-2310.
22. Li, F. C., Angelier, J., Chen, R. F., Hsieh, H. M., Deffontaines, B., Luo, C. R., et al. (2005). Estimates of present-day erosion based on sediment transport in rivers: a case study in Taiwan. Comptes Rendus Geoscience, 337(13), 1131-1139.
23. Lin, G. F., & Chen, L. H. (2004). A non-linear rainfall-runoff model using radial basis function network. Journal of Hydrology, 289(1-4), 1-8.
24. Lischeid, G. (2001). Investigating trends of hydrochemical time series of small catchments by artificial neural networks. Physics and Chemistry of the Earth Part B-Hydrology Oceans and Atmosphere, 26(1), 15-18.
25. Lu, R. S., Lai, J. L., & Lo, S. L. (1998). Predicting solute transfer to surface runoff using neural networks. Water Science and Technology, 38(10), 173-180.
26. Minns, A. W., & Hall, M. J. (1996). Artificial neural networks as rainfall-runoff models. Hydrological Sciences Journal-Journal Des Sciences Hydrologiques, 41(3), 399-417.
27. Old, G. H., Leeks, G. J. L., Packman, J. C., Smith, B. P. G., Lewis, S., & Hewitt, E. J. (2006). River flow and associated transport of sediments and solutes through a highly urbanised catchment, Bradford, West Yorkshire. Science of the Total Environment, 360(1-3), 98-108.
28. Sharma, V., Negi, S. C., Rudra, R. P., & Yang, S. (2003). Neural networks for predicting nitrate-nitrogen in drainage water. Agricultural Water Management, 63(3), 169-183.
29. Sivakumar, B., Jayawardena, A. W., & Fernando, T. (2002). River flow forecasting: use of phase-space reconstruction and artificial neural networks approaches. Journal of Hydrology, 265(1-4), 225-245.
30. Thomas, R. B. (1985). Estimating total suspended sediment yield with probability-sampling. Water Resources Research, 21(9), 1381-1388.
31. Thomas, R. B., & Lewis, J. (1993). A comparison of selection at list time and time-stratified sampling for estimating suspended sediment loads. Water Resources Research, 29(4), 1247-1256.
32. Thomas, R. B., & Lewis, J. (1995). An evaluation of flow-stratified sampling for estimating suspended sediment loads. Journal of Hydrology, 170(1-4), 27-45.
33. Zhang, B., & Govindaraju, R. S. (2003). Geomorphology-based artificial neural networks (GANNs) for estimation of direct runoff over watersheds. Journal of Hydrology, 273(1-4), 18-34.
34. 張斐章，張麗秋，黃浩倫，(2003)。類神經網路－理論與實務。東華書局。
35. 張智華，(1999)。工廠及工業區非點源污染初期沖刷之研究。國立成功大學環境工程系碩士論文。
36. 梁晉銘，張斐章，陳彥璋，(2000)。複合演算類神經－模糊推論模式應用於洪水預測。中華水土保持學報 31(3)，183-191。
37. 黃家勤，(2001)。逕流污染定量之合理採樣頻率研究。國科會專題研究計畫成果報告，NSC 89-2211-E214-002。
38. 溫清光，(2001)。曾文水庫水質調查及改善計畫。經濟部水利署南區水資源局。
39. 溫清光，(2003)。曾文水庫水質調查宜改善後計畫。經濟部水利署南區水資源局。
40. 羅華強，(2005)。類神經網路－MATLAB的應用。高立圖書有限公司.