Comparing diameter and order approaches with simulation modeling of root longevity and turnover: Estimation accuracy and error partition

Background/Question/Methods The question of whether order or diameter approach should be preferred in estimating fine root longevity and turnover is still inadequately resolved. This is due partly to weak analysis methods (e.g., Cox regression) and fuzzy data, and partly to lack of explicit definitions of the criteria to judge the preference quantitatively. One solution to the problem is to use a root turnover simulator (RTS) to generate complete root population data suitable for applications of powerful methods and for significance ranking based on estimation accuracy and error partition, two criteria that can help clarify the issues. Two simulation experiments were used. First, a factor-removal experiment was conducted with four submodels in RTS: (1) maximum heterogeneity that considers both factors, (2) minimum heterogeneity that considers neither factor, (3) order only, and (4) diameter only. Relative errors of output variables and relative importance of captured heterogeneity were calculated and contrasted between the two approaches. Second, to determine error in root estimates associated with each approach, uncertainty analysis of Monte Carlo simulations was used to partition errors by running RTS with diameter-related parameters and order-related parameters separately. Error contributions to root estimates by the factors and the model parameters were quantified.

Results/Conclusions

With estimation accuracy defined by “true values” from the maximum heterogeneity submodel, the factor-removal experiment suggested that the diameter approach tended to cause more error than the order approach. The simulation results also indicated that the order approach seemed to capture higher % of variability in root populations. However, the results of the uncertainty analysis showed little difference in the relative error contribution by each approach. The essence of the order and diameter approaches is to incorporate heterogeneity in a root population into estimation by stratifying the data into homogeneous groups to reduce estimation error. The approach that can capture more heterogeneity should be preferred, even though the implications of the results presented here depend on the realism of the RTS model that still needs further testing.