Autocorrelated growth of tropical forest trees: Unraveling patterns and quantifying consequences

Tropical trees show considerable variation in growth rates. Often this variation is not random, as some trees perform better than others and as growth may be temporally correlated. Using long-term growth data obtained from tree ring analysis, we studied the degree to which growth rates of four Bolivian rainforest tree species were autocorrelated and how this affected the output of growth simulations.Autocorrelated growth is commonly defined as the correlation between growth in one time interval with that in a subsequent interval calculated over all individuals of the population. We termed this total autocorrelated growth and identified its two components: temporally correlated growth rates of individual trees (within-tree autocorrelated growth) and persistent growth differences between trees (among-tree autocorrelated growth).Total autocorrelated growth was high (Pearson's r ∼ 0.75) between growth rates of subsequent years and decreased gradually at larger time lags. At time lags of 20 years growth rates were still positively autocorrelated in some species.Juvenile trees tend to have strong within-tree autocorrelated growth (Pearson's r ∼ 0.4–0.5), probably mainly caused by temporally correlated variation in light availability due to canopy dynamics. The within-tree autocorrelation was considerably lower in larger trees (Pearson's r < 0.2), and did – in contrast to juvenile trees – not contribute much to total autocorrelation. In larger trees total autocorrelation originated mostly from persistent growth differences among trees, caused by factors as site-specific differences or differences among trees in crown area or liana infestations. Among-tree autocorrelated growth was strong and long-lasting: differences between fast and slow growing trees were maintained for long periods.Incorporation of autocorrelated growth in bootstrap simulation models led to higher variation in age estimates compared to simulations without autocorrelation. Still, this variation was lower than that observed in tree rings. By using 5-year growth steps instead of the 1-year growth steps the observed variation increased and closely matched those in tree rings.Our findings emphasize the importance of incorporating autocorrelated growth in tree growth simulation models, for obtaining more realistic estimates of long-term growth and tree ages.