Can long-term survival models improve inference in seed germination studies? evidence from tropical forest species
Sebastião Martins Filho, Marciel Lelis Duarte, Lausanne Soraya de Almeida, Glauciana da Mata Ataíde, Barbara França DantasAbstract
Background and Aims
Seed germination analysis often ignores temporal dynamics and non-germinating fractions, limiting biological inference. Conventional methods assume all seeds germinate, which is unrealistic for heterogeneous forest seed lots. Long-term survival (cure fraction) models offer a framework to address this limitation. This study uses seed germination of native Brazilian forest species to demonstrate the applicability and advantages of long-term modeling.
Methods
Five independent datasets from native Brazilian forest species were analyzed under varying environmental conditions, including temperature, osmotic stress, light, and seed traits. Germination was modelled as a time-to-event process using parametric (log-normal, Weibull, log-logistic) and semiparametric (Cox) approaches with long-term survival models. Kaplan–Meier estimators were used for initial exploration and model comparison. Model selection was based on graphical fit and Cox–Snell residuals.
Key Results
All datasets showed a fraction of non-germinated seeds, with germination curves stabilizing below 100%, which corroborates the use of cure models. Long-term models consistently outperformed traditional survival approaches, providing better fit and biological interpretability. Environmental factors differentially affected incidence and latency components: extreme temperatures and water stress significantly reduced the probability of germination, while having smaller effects on germination speed. The probability of germination varied widely between species and conditions, from 10% to 95%. The models also revealed complex interactions between environmental and seed characteristics that influence germination dynamics.
Conclusions
This study demonstrates that long-term survival models provide a robust, biologically realistic statistical framework for analyzing germination data from forest seeds. Partitioning the germination process into incidence (probability of germination) and latency (time to germination) components enabled key advances. First, it allowed precise quantification of the fractions of germinating and non-germinating seeds. Second, it provided more accurate and informative estimates than traditional survival methods. Finally, it enabled the isolation of the effects of environmental factors on germination capacity and speed.