Predicting productivity and climatic resilience of coniferous–broadleaved mixed plantations with 3-PGmix and machine learning

The transition from monocultures to mixed plantations is a key initiative in China to enhance plantation quality. Predicting how such measures affect stand productivity and resilience under climate change is therefore crucial. We employed a process-based 3-PGmix model in conjunction with two machine learning methods (random forest and artificial neural network) to predict the spatial distribution of biomass and productivity over a 40-year period following the implementation of various transition measures under diverse climate scenarios. The findings revealed that an optimal tree species mixing proportion can significantly enhance stand productivity, with the most favorable mixing ratio observed at approximately 70% coniferous and 30% broadleaved species. In our study area, transitioning from pure Cunninghamia lanceolata stands to a mixed stand with Schima superba under this proportion resulted in a total stand productivity increase of 3336.74 t ∙ year-1 by 2060, averaging a yearly increment of 1.2 t ∙ ha-1. In the SSP245 scenario, biomass declined by 9.9%, 7.0%, and 7.7% for pure Cunninghamia lanceolata, pure Schima superba, and their mixed stands, respectively. Under the SSP585 scenario, the biomass of these stands decreased by 15.9%, 11.1%, and 13.4%, respectively. Sensitivity analysis showed that declining stand productivity primarily stems from temperatures exceeding optimal levels for tree growth. The negative impact of rising monthly maximum temperatures outweighs the positive effects of increasing monthly minimum temperatures and precipitation. Random forest outperforms artificial neural networks in simplifying 3-PGmix. The results of this study can offer valuable insights into transitioning pure coniferous plantations into coniferous-broadleaved mixed stands.