Cross-regional characterisation and prediction of reproductive phenology in Prunus cerasoides Buch. -Ham. Ex D. Don using sequential learning

Abstract

Reproductive phenology provides insights into plant adaptation strategies under changing climates, and thus requires extensive studies on intraspecific variations across climate gradients. In this study, we examined the reproductive phenology of the Himalayan wild cherry (Prunus cerasoides Buch. -Ham. Ex D. Don) at two climatically contrasting sites – the tropical Mizoram, part of the Indo-Burma region, and the temperate Uttarakhand, part of the western Himalayas, between 2019 and 2023. Monthly climatic variations in temperature, rainfall, humidity, and wind speed, as well as the reproductive phenological observations of the flowering and the fruiting phases, were recorded. Comparative analyses revealed an earlier and shorter flowering period in Uttarakhand compared to Mizoram, suggesting site-specific adaptive responses of the Himalayan wild cherry. We developed and implemented a staggered machine learning pipeline using regularised regression models (Lasso, Elastic Net and Ridge) to predict four key events: first flowering day, peak flowering day, last flowering day and the fruit drop day, using site-specific monthly climatic data. Temperature, rainfall, and their interaction were the major determinants of reproductive timing, contributing to nearly 90% prediction accuracy accounting for MAE of around 5 days across all events on average. Residual analyses further showed higher accuracy for Uttarakhand predictions, consistent with stronger climatic correlations at this site. These findings highlight substantial cross-regional variation in phenological sensitivity and underscore the potential of machine learning to integrate climate–phenology relationships. Our approach provides a framework for forecasting reproductive events in tree species under climate variability, with implications for conservation, pollination ecology, and climate adaptation strategies.