Premium
A single model ensemble versus a dynamic modeling platform: Semi‐distributed rainfall runoff modeling in a Hierarchical Mixtures of Experts framework
Author(s) -
Marshall Lucy,
Sharma Ashish,
Nott David
Publication year - 2007
Publication title -
geophysical research letters
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 2.007
H-Index - 273
eISSN - 1944-8007
pISSN - 0094-8276
DOI - 10.1029/2006gl028054
Subject(s) - probabilistic logic , surface runoff , context (archaeology) , representation (politics) , computer science , environmental science , drainage basin , statistical model , hydrology (agriculture) , geology , geography , machine learning , cartography , ecology , artificial intelligence , paleontology , geotechnical engineering , politics , political science , law , biology
One way to address the perceived biases in many hydrological model simulations is to incorporate the information from several models at once. In this study we present an alternative where the catchment can be modeled as existing under a number of discrete states, which are chosen in a probabilistic manner based on selected catchment indicators. We present an application of the approach in the context of a semi‐distributed catchment simulation model. We compare the predictive performance of several observed and simulated catchment indicators for aggregating the models via three catchments in different continental, geographic and climatic settings. Two main conclusions are drawn: (a) the multiple state representation of catchment behavior is an attractive alternative to single model parameterizations in terms of predictive power, and (b) each of the discrete probabilistic states is predictable using catchment indicators that reflect the lower order runoff producing mechanisms.