The purpose of this study is to estimate the impacts of climate variables on rice productivity in Korea. In order to allow for a flexible functional relationship, the penalized spline regression which is a semi-parametric model is used. A panel data set of 49 cities and towns for the period from 1996 to 2011 is used for estimation. The spline regression model allows random intercepts incorporating the panel nature of the data set. Moreover, we identify the observations with extreme weather conditions such as flood in order to incorporate the increasing variability of climate factors. The study finds that rice productivity is significantly affected by technical change, sunshine duration and temperature in the grain-filling months. It is found that the variables of sunshine duration must be incorporated as non-parametric elements of the semi-parametric model. There is a positive relationship between sunshine duration and productivity. An inverted U-shape relationship between temperature and productivity is found. The study shows that the peak temperature comes earlier if the panel-nature of the data is taken into account.