Fitting photosynthesis irradiance response curves with nonlinear mixed-effects models

摘要

Studying photosynthesis-irradiance (P-I) relationships is fundamental to plant ecophysiology and ecological research. Important ecological questions are often inferred, based on findings from P-I studies. A P-I relationship is intrinsically nonlinear, and measurements typically contain random variations and other complicated data structures. Although rarely adopted, the best approach to analyze a P-I relationship is to use nonlinear mixed-effects (NLME) modeling. We believe that such a failing is mainly because ecophysiologists are unfamiliar with and uneasy in using the approach. Consequently, the main objective of this study is to help ecophysiologists better understand how to use NLME modeling to fit P-I and other response curves. Using data from an artificial shading experiment as an example, we outline a 'backward elimination' strategy for fitting P-I curves by NLME modeling. We also summarize our model development process that led to the final model. Compared with nonlinear models estimated by ordinary and generalized nonlinear least squares estimations, the final NLME model had the smallest estimated error variance, enabling it to detect the presence of photoinhibition where the other 2 models failed. Our study shows that an inadequate estimation approach might not only be statistically inefficient and less powerful, but may also lead to incorrect inferences.