Sensitivity Analysis Of Chlorophyll-a Prediction Model Of Long-jing Lake

Long-jing lake is a closed deep lake with bad hydrodynamic condition and plentiful nutrient, it is easy to breed algae when other environment factors are in appropriate conditions, which will seriously impact the landscape function. This article aims at the research of algae growth influence mechanism, which will reveal the most influential environment factor to the algae growth, in order to provide guidance to prevent and control algae bloom. On the basis of water-quality monitoring, a BP neural network model for chlorophyll-a prediction was established specific to each kind of monitoring site, then analyzed the input parameters of each BP neural network model with local sensitivity analysis and global sensitivity analysis. The main results are as follows:① By way of correlation analysis, we calculated the relevance among the environmental factors between environmental factors and chlorophyll-a. The input parameters of BP neural network were selected according to the correlation analysis, selecting rules were: eliminate factors which are not relevant(R<0.3) with chlorophyll-a and merge those highly relevant(R>0.8) to each other. Finally, three group of input parameters were selected.② By comparing the prediction accuracy and decision coefficient of the chlorophyll-a prediction model, tansig was choose to be the transfer function of hidden layer, purelin was choose to be the transfer function of output layer. The hidden layer neurons number of prediction models for Open-air Theatre lake-bay, lake center and Qiuting Bridge lake-bay are 8, 13,13 respectively. The average forecast error of the model was 2.23%, 2.01% and 2.37%, which indicated the model reached a good forecast level.③With the applying of the local sensitivity analysis based on partial derivatives to three chlorophyll-a prediction model, we concluded that the chlorophyll-a on all three monitoring sites showed up high sensitivity to water temperature, ORP, TP and CODMn. Meanwhile, the sensitivity of every input parameter changed over time, they stayed at high sensitivity level and keep great difference with each other before mid-november, while stable and little difference after mid-november.④With the applying of Pa D2 global sensitivity analysis, the sensitivity of chlorophyll-a to every double factor has been worked out. It revealed that chlorophyll-a showed relatively high sensitivity to water temperature and TP, water temperature and CODMn, euphotic depth and CODMn. Otherwise, after compared with local sensitivity analysis, we obtained that even though some environment factors showed low sensitivity by means of local sensitivity, they might get higher sensitivity after combined with other environment factors because of the synergistic effect between them.⑤By comparing the sensitivity of TN with TP in local sensitivity and global sensitivity, we found that the sensitivity of TN was obviously lower than the sensitivity of TP, which indicated that L-J lake was a phosphorus restricted type lake.