Numerical Prediction of paddy weight of Crop Cutting Survey using Generalized Geoadditive Linear Mixed Model

Rice production data is needed to support the information about achieving the second SDGs. Rice production data requires rice productivity data obtained from Crop Cutting Survey by BPS-Statistics Indonesia. The problem is that the measurement of unhulled rice weight in this survey is not always successful. This problem causes the unhulled rice weight data to be missing values. We proposed Geo-GLMM with covariate interaction to estimate the missing values. The proposed methods was compared by GLM, GLMM, and Geo-GLMM. The results showed that seed varieties, TSP/SP36 fertilizer, NPK / compound fertilizer, urea, organic fertilizer, the number of clumps per plot, pest attack, and climate impacts significantly affected rice productivity. Then, we selected the variables and got the best explanatory variables, namely seed varieties, fertilizer, interaction between urea and KCL fertilizer. Geo-GLMM with fertilizer interaction has better prediction performance than GLM, GLMM, and Geo-GLMM without interaction. Based on the results of the simulations, the Geo-GLMM with covariate interaction produces a smaller bias and RMSE. Therefore, it is recommended that the surveyors of Crop Cutting Survey continue to interview farmers when they fail to take sample plots, so we get covariate data can be used to estimate the unhulled rice weight.