A 10 m count of trees of the same species gave no clear gain in later small-tree death predictions at Barro Colorado Island, 1990–2015

We tested whether nearby trees of the same species improve predictions of death at the Barro Colorado Island 50-ha forest plot. We fit two logistic models on 195,405 tree intervals in 1985–1990. The starting diameter was 1–10 cm. The baseline used species, diameter, total nearby tree count, 100 m location blocks and measured follow-up time. The added term was the log of one plus the count of trees of the same species within 10 m. We held both models fixed and tested 883,994 later tree intervals, with 103,826 deaths, in 1990–2015. Mean loss changed from 0.335539236 to 0.335538565 nats. The relative loss reduction was 0.00020%, with a nominal 95% spatial-block range of -0.00852% to 0.00922%. The count gave no clear pooled gain. Fixed checks at 5 and 20 m gave small, mixed changes. An exploratory model with separate count slopes by species fit the training data better but worsened later loss by 0.155%. Source checks and a clean reproduction passed. The block ranges condition on the fitted models and observed periods. These results apply to the stated models and one forest. They do not show that biological density effects are absent.

forest census · tree death · neighbor counts · chronological prediction · Barro Colorado Island · reproducibility