Earlier species and size transitions reduced 2015 tree species share forecast error at Barro Colorado Island

We tested whether earlier tree transitions improve a ten-year forecast of species shares among trees with a measured main diameter of at least 10 cm. We used the 2019 census data release for the 50-hectare Barro Colorado Island plot in Panama. Species and diameter transition chances from 1985–1995 and 1990–2000 gave forecasts from the 2005 census to the 2015 census. Total variation error fell from 0.06347 for no change to 0.03774, a reduction of 40.5%. A paired resampling of ten spatial strips gave a central 95% range of 23.9–42.7%. This range describes spatial sensitivity within the plot, not forecast uncertainty across new decades or forests. Error was 16.8% lower than a reference based on earlier species count ratios. A forecast with size classes but no species-specific rates increased error by 30.6% relative to no change. The main model also reduced squared and squared Hellinger share error. The model predicted 21,450 large records against 20,829 observed; no change predicted 20,850. Better species shares therefore did not give a better total-count forecast. The package preserves selected source records, code, all scores, spatial resamples, figures, source checks, and repeat instructions. Missing diameters and record entry limit a biological reading of the transitions. The test applies to one plot and a retrospective later interval.

tropical forest · species composition · tree diameter · forest census · forecast test · Barro Colorado Island · demographic transitions