Quantile regression for plant growth and climate relationships

BRUNO, C.; BALZARINI, M.

Florencia

Conferencia; XXVIIth International Biometric Conference; 2014

International Biometric Society

This study compared the growth responses to water deficit and temperatures of different varieties of sugarcane using quantile regression analyses (QRA) to establish how these climate-related covariables can explain growth. The method was used as a way to estimate the conditional quantiles of the growth rate distribution in a linear model including the observed covariables in order to provide a more complete view of possible causal relationships between growth and climate. Growth is not uniquely driven by a given covariable included in the model, other potential limiting factors could be at not permissive levels, increasing the heterogeneity of growth rates with respect to the measured covariable(s). The heterogeneous variance may result in non-significant effects for these variables when the focus is exclusively on changes in the mean. Fitting a range of QRA lines with increasing quantil (τ) from 0.05 to 0.95 allowed to find significant effects of these covariables, that are not captured by a classical linear regression. In our experimental data, plots were established for 3 months and measured at weekly intervals in outdoor cultures. Consecutive height data were used to estimate daily growth rates for each variety. Growth data were analyzed as a function to absolute water deficit and growing-degree day through a QRA for longitudinal data to account for the repeated measure scheme. The effects of water deficit and temperature on growth rate were different between varieties at the extreme (lower and upper, respectively) quantile levels. QRA proved to be powerful tools in distinguishing growth responses to climatic variables from variety differences.