|
Activity Number:
|
35
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Sunday, August 2, 2009 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Statistics and the Environment
|
| Abstract - #303592 |
|
Title:
|
Prediction Implications of Nonlinear Mixed-Effects Forest Biometric Models Estimated with a Generalized Error Structure
|
|
Author(s):
|
Shongming Huang*+ and Shawn X. Meng and Yuqing Yang
|
|
Companies:
|
Alberta Government and Alberta Government and Alberta Government
|
|
Address:
|
Biometrics Unit, Forest Management Branch, 8th Floor, Edmonton, AB, T5K 2M4, Canada
|
|
Keywords:
|
forest growth prediction ; longitudinal data ; nonlinear mixed model ; correlated and heterogeneous errors
|
|
Abstract:
|
Three forest growth models were estimated using the first-order methods of nonlinear mixed model technique, based on the longitudinal data collected from permanent sample plots and sectioned trees. To account for the correlated and unequally varied nature of the data, a generalized error structure was selected after evaluating more than 20 error structures. Model predictions with and without using the generalized error structure were compared on model fitting and independent model validation data sets. Results showed that the models estimated with the generalized error structure produced larger biases and variations on all data sets. This suggests that, while the successful accounting of the generalized error structure may be useful in hypothesis testing and interval estimation, it will likely result in larger prediction biases for forest models developed as predictive tools.
|
