PakCat catalogue record

Entropy Concept in Feature Entraction and Hydrologic Time Series Synthesis 'Source Title' Modeling Hydrologic Processes, Proceedings Of 3rd International Hydrology Symposium

By Panu, U.S

Bibliographic details

Summary

It is reasonable to assume that sequences of daily, weekly or monthly flows occur in well defined groups that can be considered to possess what can be called streamflow patterns. A pattern is a description of an object. The objects of concern here are groups of data on flow observed at regular time intervals. Streamflow patterns consist of appropriate properties. Streamflow patterns consist of appropriate properties, n in number, describing each of these groups. Consider that a streamflow pattern can be assigned for flows in each season within a geophysical year. There are, then, as many streamflow patterns within the year as there are seasons. Further, each such streamflow pattern can be taken to belong to a pattern class. A part of this paper discusses a feature extraction algorithm for selecting features in a pattern class based on the entropy concept of Shannon (1948), whereby uncertainty in the selection of features is minimized. Towards this end the input data is transformed through proper consideration of the joint prohability distribution of the elements in the pattern vector. The remaining part of this paper discusses the development of a model for streamflow time series synthesis. This involved the diviswion of the time series into pattern ther, it is also required to determine the inter pattern as well as the intra pattern (inter feature) relationship.