A Comparative Study of Regression Models for Climatic Impact Factors on Seed Yield in Korean Red Pine Seed Orchards

Abstract

Korean red pine (Pinus densiflora Sieb. et Zucc.) is an important afforestation species in East Asia, known for its highly variable annual seed yield (masting), which complicates forest regeneration and sustainable management. Accurate prediction of seed production is essential, yet challenging due to multicollinearity, nonlinearity, and lag effects among climatic variables. This study compares multiple linear regression (MLR), elastic net regression (ENR), and partial least squares regression (PLSR) for modeling the influence of climate on seed yield in Korean red pine. Using 18 years of data (2003–2020) from a national seed orchard in Korea, models were evaluated via 10-fold cross-validation based on R² and root mean square error (RMSE). PLSR outperformed both MLR and ENR, achieving a test R² of 0.662, by effectively extracting latent variables to reduce multicollinearity—the first three components explained 74.3% of the variance. Key climatic drivers included July mean temperature (β = 0.42) and July precipitation (β = 0.38), aligning with species’ physiological needs during critical phenological stages. PLSR is particularly suitable for ecological modeling with limited samples and correlated predictors. Future studies should incorporate nonlinear and temporal dynamic models to improve predictive accuracy and mechanistic understanding.