Using Neural Network

Pollution from diffuse sources in lakes and reservoirs has beeome a serious problem, The location of the diffuse source is still diMcult to idelltify and the load frorn it is also difficult to estimate because of its inherent properties. The rnechunism of the pollution from the diffuse sources is conceptually recognized, but the dctails, e.g., pathwtty to lakes and detention period in soil, are still obscure. Among the informative assets avai]able, most of which are uncertain, observed water qualities and climatic coiiditions at some stat,ions in lakes are certainly well quant,ified and reliable. This study aiiris to develop a method to estimate the relative pollutant loads of the diffuse sources nsing an artificial neural netwerk, in which the observed water quaLities aiid climate are used as keys to the solution. In addition, supplernentary tutoring procedure to aid the training of the neural network is introduced to reduce the error in the estimation. The method developed is applied to Lake Biwa under some hypothetical conditions. The results show that the method successfu11y estimates the relative pollutant loads and hence its significant contribution t・o help water quality rnaiiagement of lakes will be expected,