Provider: Ingenta Connect
Database: Ingenta Connect
Content: application/x-research-info-systems
TY - ABST
AU - Guan, T.
TI - Effects of Correlation among Parameters on Prediction Quality of a Process-Based Forest Growth Model
JO - Forest Science
PY - 2000-05-01T00:00:00///
VL - 46
IS - 2
SP - 269
EP - 276
KW - rank correlation
KW - uncertainty analysis
KW - Ecological model
KW - Cholesky decomposition
N2 - A nonparametric method was introduced as a technique for evaluating the effects of parameter correlation on prediction quality of process-based forest growth models. The method was based on a rank correlation and Cholesky decomposition. For a given data matrix, by reordering the observations, the method would produce a rearranged matrix with the desired correlation structure. The method was computationally simple and efficient. Using a process-based model calibrated for red pine (Pinus resinosa Ait.) as an example, small-scale Monte Carlo simulations revealed that parameter correlation had different effects on the model's prediction means and variances, depending on the importance of the parameters involved. In general, given the same level of correlation, correlation between two important parameters would have greater influences on prediction quality than correlation between two less important parameters. Parameter correlation slightly affected the prediction means but could significantly change prediction variances. The relationship between prediction quality and the degrees of correlation, however, was not necessarily a linear one. The same parameter correlation might also have different effects on different state variables. For. Sci. 46(2):269-276.
UR - http://www.ingentaconnect.com/content/saf/fs/2000/00000046/00000002/art00014
ER -