Suchergebnisse

The revised Bathing Water Directive (rBWD) (2006/7/EC) of the European Parliament requires monitoring of bathing water quality and, if early-warnings are provided to the public, it is permissible to discount a percentage of exceedance events from the monitoring process. This paper describes the development and implementation of both Decision Tree (DT) and Artificial Neural Network (ANN) based machine learning models for 8 beaches in south-west England, UK, as bases for early warning systems (EWS) and compares their performance for one beach. Weekly bacteria-count samples were gathered by the Environment Agency of England (EA) over a 12-year period from 2000-2011 during the 20-week bathing season and this data is used to calibrate and test the models. Daily sampling data were also collected at 5 of the beaches during the 2012 season to provide more robust validation of the models. As a benchmark, models are also compared with use of simple thresholds of antecedent rainfall to classify water quality exceedances. Evolutionary Algorithm-based optimisation of the ANN models is employed using single-objective approach using area under the Receiver Operating Characteristic (ROC) curve as fitness function. The optimum operating point is established using a weighting factor for the relative importance placed on false positives (passes) and false negatives (exceedances). The models use a number of input factors, including antecedent rainfall for the catchment adjacent to each bathing beach. A possible technique for automating selection of inputs is also discussed. ; Environment Agency (SW)