When
Where
Special seminar talk by Dr. Yang Yang from the University of Hong Kong
Abstract
Are hydrological models with random parameter values useful? Yes, they are. For a given weather event, by comparing the actual response of a catchment with that of random models, we can gain valuable information about the similarities between the catchment and the specific model representations. Is it possible to create a profile for each catchment and model, so that the similarity between a catchment and a model representation can be predicted from the profiles alone (without running the model)? This study shows that by learning from the similarities between catchment-model representation pairs, we can assign each catchment and each model representation a numerical vector, and from which highly accurate predictions of catchment-model representation similarity can be derived. This result indicates that the hydrological functions of the catchments and model representations can be adequately characterized by a few latent factors. A new and effective method of model calibration is proposed, where the best fitting model representations of a new catchment (not in the database) can be easily retrieved from the database containing models with different conceptual model structures by estimating its latent factor values through evaluating a number of random models.
Bio
Dr. Yang Yang is a Postdoctoral Fellow at the Department of Civil Engineering, The University of Hong Kong, where he received his PhD degree. His research interests include machine learning applications in hydrological modeling, reliability assessment of modeling systems, and sustainable stormwater management.