Neural network predictors of average score per taxon and number of families at unpolluted river sites in Great Britain

摘要: Abstract Biological monitoring of river water quality in the United Kingdom and several other European Commonwealth countries is based on Monitoring Working Party (BMWP) system. Central to present day application this system prediction “unpolluted” average score per taxon (ASPT) number families (NFAM). The paper outlines need for such predictions proceeds develop predictors ASPT NFAM using neural networks. basic principles networks are outlined a brief introduction their structure function given via typical example. Important preliminary considerations fully discussed, as model selection, training testing procedures selection relevant input variables. results impact analyses, designed optimise structures networks, reported discussed. In-depth analyses performance independent test data also relative industry's current model, RIVPACS III, presented. investigations into bias error predicted values discussed related some possible inadequacies database. It concluded that: significantly more reliable than those NFAM; performed marginally better III; can be directly, without reference site type or biological community, from few key environmental variables; there scope improved if additional collected.