多元迴歸式帶入馬可夫鍊（Markov Chain）模型求得金瓜寮溪流域未來之土地利用變化情形，則2052年河川水質之總磷、氨氮與硝酸鹽氮濃度，將較2004年分別增加11.1、35.4與11.0 ％。多元迴歸式帶入BAU（Business As Usual）求得之土地利用變化情形，則2052年河川水質之總磷、氨氮與硝酸鹽氮濃度，將較2004年分別增加46.1、132.0與17.4 ％。若自2010年起農地面積發展速率增加10倍，則2052年河川中之總磷、氨氮與消酸鹽氮濃度，較農地面積正常發展分別提高69.6、108.9與105.7 ％；其中總磷與氨氮濃度分別提高25.65 μg/L 與0.158 mg/L，並超過甲類水體標準。
This study demonstrated the effects of land change and streamflow on water quality in Feitsui Reservoir Watershed and aimed to quantify the relationships rather than determining the qualitative interactions. Pearson product-moment correlation was used to identify the water quality those significantly affected by land change; furthermore, multiple regression analysis was applied to link the relationship. The derived regression equations were finally served as decision-making tools to predict the water quality in several future land use scenarios.
The results showed that the most significant correlation between flow and water quality occurred in suspended solid (SS), which coefficient of determination (R2) is 0.573. The correlations in other water qualities were much low, i.e. 0.287 for total phosphorous (TP) and 0.149 for ammonia-nitrogen (NH4-N). Although the concentrations of all water quality increase with higher streamflow, the correlations are not confident except SS. In the contrary, the relations between land use and water quality were distinct, especially in TP, NH4-N and nitrate-nitrogen (NO3-N). The effect levels from various land use on water quality were different. Urban and agriculture land affected TP and NH4-N clearly, in which the R2 is more than 0.7 and 0.6, respectively. The influence of NO3-N was dominated by forest and the R2 was 0.727.
Three land use scenarios were assumed. They are business as usual (BAU), forecast from Markov Chain model, and fastened development of agriculture. In the third scenario, the developing rate of agriculture is assumed 10 times than the past and the exceeded land is converted from forest land. Water quality was predicted under the land change scenario with multiple regression equations. The end year of prediction is set in 2052 and the base year to compare is 2004. The predicted water qualities are increasing in the calculation period under each scenario. The concentrations of TP, NH4-N, and NO3-N are increased 46.1%, 132.0%, and 17.4% in BAU scenario. While applying Markov Chain scenario, the increasing rate are less than those in BAU. If fastening the agriculture development in the study area, the concentrations of TP, NH4-N, and NO3-N will be higher 69.6%, 108.9%, and 105.7% than that in 2004. The concentration of TP and NH4-N will be 25.65 μg/L and 0.158 mg/L, both violate the water quality standard.