EcoSta 2022: Start Registration
View Submission - EcoSta2022
A0659
Title: Climate change and rice yield: Compositional scalar-on-function regression approach Authors:  Thi Huong Trinh - Thuongmai University (Vietnam) [presenting]
Abstract: Climate change has a significant impact on crop yields, especially in an agricultural country like Vietnam. Climate change is measured by changes in the maximum and minimum daily temperature for 30 years, from 1987 to 2016. We address the impact of weather, here the maximum (and minimum) daily temperature on rice yield per year in each province through a compositional scalar-on-function regression. A total of 3780 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 $clr$ inverse transformation. The results in $B^2$ directly provide insight into the impact of climate on rice yield.