Investigation Of Health Assessment For Urban Lakes System Based On Probabilistic Neural Networks(PNN)

In recent years, with the improvements of social, economy and lifestyle, urbanlakes have been paid attention due to its unique ecology and landscape features. Urbanlakes are mostly artificial and the ecosystem complexity is low, or suffered seriouspollution. The lake’s ecosystem integrity was destroyed, leading to weak the capabilityto resist outside interference. According to the State of The Environment In China’sBulletin in2010, lakes with eutrophication status accounted for42.3%in twenty-sixlakes. Meanwhile, almost urban lakes were heavy-or abnormal-nutritionaleutrophication state. Thus, revealing the ecosystem health state of lake is veryimportant for government to develop effective control programs and policy.The methods and indicators of lakes ecosystem health evaluation varied withdifferent regions and subjects. So, it was difficult for them to be generalized because ofits subjectivity. In this study, probabilistic neural network model (PNN) with itsobjectivity in determining classification weight was established in order to solve aboveproblems. According to the data of water quality in Baiyun lake in Guangzhou duringtwo years, eight indicators involving biochemical oxygen demand (five days), totalnitrogen, total phosphorus, dissolved oxygen, biological diversity, trophic state index,energy quality and structure were chosen to assess the health state of Baiyun lakeecosystem. Meanwhile, these indexes were classified into five levels based on theenvironmental standards and previous research. The results of PNN were obtained bytraining the samples of indictors. It was showed that the ecosystem had taken a turnpositive trend, but still fragile, and the ability to purify water was limited. All ofmonitoring sites showed seasonal variation, that was, wet period of the ecosystemhealth was significantly better than that in the dry season, while the annual variationwas not significant. According to the results, three suggestions were proposed asfollowing:(1) increasing the submerged plants, floating plants and others ecologicalmeasures;(2) the sewage must be prevented around the lake;(3) increasing theamounts of projects for improving water quality. In conclusion, as compared to the BP neural network model, the PNN for lakesecosystem health evaluation can be faster and more accurate, and therefore the modelhas a strong generalization.