BCI tree death fractions had excess variation across 20 m cells under local species and size shuffles in 1985–2015

We tested spatial variation in tree deaths after control for species, starting stem size and local area. The data covered six census intervals from 1985 to 2015 at Barro Colorado Island (BCI), Panama. The study used one 50 ha forest plot. The data had 1,344,231 eligible tree-interval records and 154,989 deaths. We kept death totals fixed within each species, diameter class and 100 m area. We then assigned these deaths to random trees in each group. We tested 1,250 cells of 20 by 20 m. Observed variation was 2.38 to 2.66 times mean variation from 999 shuffles in each interval. Variation was measured as the mean squared residual of the cell death fraction. No shuffled value reached its observed value. Each six-test Holm p value was 0.006. Excess root mean square variation was 2.35 to 2.81 percentage points. All nine sensitivity checks gave ratios above one in all six intervals. For trees with diameter at least 100 mm, the ratios were smaller, at 1.16 to 1.38. Cell residual correlations across adjacent intervals were 0.23 to 0.30 after local control. These correlations are descriptive values. The tests rejected random assignment of deaths within the stated groups. It does not identify a cause of death, show a climate effect, or test another forest. Fine habitat, tree condition and local disturbance remain possible explanations. The package has input column extracts, source hashes, all results and simulation draws, figures, code and offline repeat instructions.

tree mortality · spatial variation · Barro Colorado Island · conditional randomization · forest dynamics · open data