分野別セミナー

Climate change has a significant impact on crop yields, especially in an agricultural country like Vietnam. In this study, climate change is measured by changes in the distributions of maximum and minimum daily temperatures during 30 years, from 1987 to 2016. We address the impact of these weather covariates on rice yield per year in each province through a compositional scalar-on-function regression approach. A total of 1890 samples, i.e. province's daily temperature per year, are expressed as density functions in the Bayes space $B^2$. The functional centered log-ratio transformation, $clr$, converts the density function from $B^2$ space to $L^2$ space. We discretize the observed temperature densities and then smooth them using splines in $L^2$ with a zero integral constraint, which are adapted to our problem. Smoothing splines of the $clr$ temperature function and a scalar dependent variable are treated as a functional linear regression model in $L^2$. The estimated function, represented in a $ZB-$spline basis, is obtained by minimizing the sum of squared errors and then transferred back to $B^2$ with the functional inverse $clr$ transformation. The results in $B^2$ directly provide insight on the impact of climate on rice yield.