Inter-comparison of three models for δ13C of respiration with four regression approaches

Abstract

The carbon isotopic composition of ecosystem carbon pools and fluxes is a useful tracer in the carbon cycle. In this study, the carbon dioxide mixing ratio ([CO2]) and its δ13C were measured in-situ by wavelength-scanned cavity ring-down spectrometry in a subtropical coniferous plantation. We determined δ13C of respiration in the understory (δRs) using three models (i.e., Keeling plots (KP), Miller-Tans plots (MT), and flux ratio method (FR)) with four regressions (i.e., ordinary least squares (OLS), geometric mean regression (GMR), least-squares bisector regression (LSB), and orthogonal distance regression (ODR)). The selection of regression approaches affected KP and MT rather than FR. No significant differences were observed among the three models using OLS (p>0.05). The systematic bias and uncertainty of δRs would increase with the decrease of the [CO2] ranges (KP and MT) and the d13CO2/dz gradients (FR). The bias of δRs among the four regression approaches could be explained by the Pearson’s correlation coefficient of linear fit at small [CO2] ranges and low d13CO2/dz gradients. The uncertainty of δRs was primarily related to the δ13C measurement error based on a Gaussian noise simulation, and OLS was less sensitive to the δ13C measurement error. A threshold criterion for the [CO2] ranges or the d13CO2/dz gradients should be established to remove the δRs values with relatively large errors and improve the overall reliability of estimating δRs.