WATERSHED SIMILARITY ANALYSIS USING UNSUPERVISED-SUPERVISED NEURAL NETWORKS

摘要

This is the second phase of watershed similarity analysis with a much larger knowledge base, more represented parameters, and the completion of proof of concept. Incorporation of Geographic Information Systems (GIS) into Unsupervised-Supervised Artificial Neural Networks (ANNs) was applied to quantify the similarity of watershed characteristics. The goal of this approach is to find the best match watershed from a large knowledge base of over one thousand watersheds and to determine the reliability of "transplant" watershed information during the clustering and classification stages. The prediction stage of the study compares the hydrographs between this unknown watershed and the best-selected watershed to verify the similarity performance. The correlation coefficient for hydrograph prediction (2 years daily flow as an example) could reach 0.92 from this demonstration case. It is shown that the basin area ratio provides a reasonable conversion factor to make the magnitude adjustment for hydrograph prediction.