Scaling Local Tree Mortality Models to a Regional Scale Using Combined Estimators and a Meta-Regression Approach

• Activity Number: 183
• Type: Contributed
• Date/Time: Monday, August 2, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Suborna Ahmed* and Valerie LeMay+
• Companies: The University of British Columbia and The University of British Columbia
• Keywords: generalized logistic model ; meta-modeling ; hierarchical related regression ; scaling models ; James-Stein estimators

Abstract

Information needs at larger spatial scales often involve the use of regional scale models. One naïve approach is to fit a regional scale model using all available data. However, this does not make efficient use of existing local scale models. We examined a number of combined estimators (e.g, James-Stein) in a meta-modelling approach to scale models to a regional level. The application is tree-level mortality models for the Boreal Forest of Canada representing 10% of the World's forest resources. Since a generalized logistic model was used because of unequally-spaced repeated-measures data, modifications to combined estimators for linear models are proposed. Data from Alberta were used to create artificial populations to compare the accuracy of proposed meta-regression approaches to the naïve approach. Further work using hierarchical related regression is planned.