Performance of tropical forest seedlings under shade and drought: an interspecific trade-off in demographic responses

Seedlings in moist tropical forests must cope with deep shade and seasonal drought. However, the interspecific relationship between seedling performance in shade and drought remains unsettled. We quantified spatiotemporal variation in shade and drought in the seasonal moist tropical forest on Barro Colorado Island (BCI), Panama, and estimated responses of naturally regenerating seedlings as the slope of the relationship between performance and shade or drought intensity. Our performance metrics were relative height growth and first-year survival. We investigated the relationship between shade and drought responses for up to 63 species. There was an interspecific trade-off in species responses to shade versus species responses to dry season intensity; species that performed worse in the shade did not suffer during severe dry seasons and vice versa. This trade-off emerged in part from the absence of species that performed particularly well or poorly in both drought and shade. If drought stress in tropical forests increases with climate change and as solar radiation is higher during droughts, the trade-off may reinforce a shift towards species that resist drought but perform poorly in the shade by releasing them from deep shade.


Supplementary figures and tables
Tables S1-S2 and Figure S1 present the responses of all 91 species in the study to shade and drought, and are included in this file from page 19 onwards.

Figure S1.1 Relationships between species responses to shade and inter-annual drought (i.e. dry season severity) for growth (a), survival (d), or growth versus survival (b,c) when individuals that resprouted, were visually damaged or infected by pathogens were included in the analysis. Solid lines indicate significant relationships (p < 0.05). Negative relationships indicate a trade-off between shade and drought responses. Correlations are weighted by the uncertainty in species tolerances (smaller dots have higher uncertainty and lower weight, see equation (5) in the main text). Colours identify species with insignificant (grey) or significant responses to shade (orange), inter-annual drought (blue) or both (red).
Figure S1.2 Relationships between species responses to shade and spatial drought (i.e. the inverse of soil water potential) for growth (a), survival (d), or growth versus survival (b,c). None of the relationships was significant. Correlations are weighted by the uncertainty in species tolerances (smaller dots have higher uncertainty and lower weight, see equation (5) in the main text). Colours identify species with insignificant (grey) or significant responses to shade (orange), spatial drought (green) or both (red). Figure S1.3 Relationships between the fast-slow continuum and responses to spatial drought for (a) growth and (b) survival. The position of species along the continuum was quantified based on a weighted PCA of demographic rates (growth, survival, number of sapling recruits) of trees ≥ 1 cm dbh recorded in the BCI 50-ha plot 1 . Low and high scores correspond to species with fast and slow demographic strategies, respectively. Colours identify species with insignificant (grey) or significant responses to shade (orange) or spatial drought (green). Relationships were consistent when the fast-slow continuum was calculated using seedling performance and/or seed number additionally (see Table S1.2).

Figure S1.4 Width of the 95% credible interval of growth responses (upper panels) and survival responses (lower panels) of species to shade (a,d), spatial drought (b,e) and inter-annual drought (c,f) versus sample size (i.e. the number of seedling observations of species in the respective models). Filled and unfilled circles represent species with significant and nonsignificant responses, respectively.
Figure S1.5 Pearson correlation between shade index (means over all years) and spatial drought index at the 200 seedling census sites. Figure S1.6 Pearson correlations of soil water potentials (MPa) at 15, 40 and 100 cm depth. Samples were taken at 36 seedling census sites and 66 sites along on the border of the 50-ha plot and in a 10-ha plot bordering the full northern side of the 50-ha plot. *** All correlations are significant at p < 0.001. Figure S1.7 Pearson correlations of soil water potentials (MPa) measured at the 200 seedling census sites among the four soil moisture sampling periods. *** All correlations are significant at p < 0.001. Table S1.1 Numbers and percentages of species (in parentheses) with significant growth or survival responses to shade, drought and ln(height).

Soil water retention curves
To identify outliers in soil water potential (SWP) measurements, we constructed soil water retention curves for 25 of the 200 seedling sites. After measuring SWP of each sample, we determined soil water content (SWC) from fresh mass (f) and dry mass (d) determined after 72 hours at 105°C (SWC = (f-d)/d). We then selected 25 samples from different seedling sites that covered all soil types (cf. Baillie et al. 2 ) and topographic habitats (cf. Harms et al. 3 ) of the 50-ha plot to construct soil water retention curves. By selecting sites with different edaphic characteristics, we ensured that the curves represented different possible combinations of SWP and SWC that can be expected at the sites. Thus, substantial deviation from the curves likely indicate measurement error.
To construct the retention curves, we first added distilled water to the soil sample until saturation (0 MPa).
The soil was then gradually dried for approximately 30 minutes, weighted and measured for SWP with a WP4C Dewpoint PotentiaMeter (Decagon Devices, Inc, Pullman WA, USA). This was repeated 6-13 times until SWP was lower than -7 MPa. After this, gravimetric SWC was again determined as stated above.
Finally, we fitted a third-order polynomial line through the observed SWP vs. SWC to construct each curve, correcting the line to 0 MPa when SWP was predicted to be positive.
To determine SWP outliers, we calculated standard deviation (SD) of the SWC (horizontally) and SWP (vertically) that were measured across all sampling rounds (excluding the observations used to construct the curves). Six SWP samples deviated by more than 1 SD from the most extreme retention curves ( Fig.   S2.1). These samples were considered outliers and excluded from the analysis. With the remaining samples, we calculated the median SWP per site and multiplied these by -1 to obtain our spatial drought index (Ds).

Model implementation and diagnostics
Posterior distributions and error components were modelled using the Bayesian inference software package RStan version 2.16.2 4 in R version 3.4.1 5 . Convergence was monitored by running four chains with different starting values. We also used the potential scale reduction factor (Rhat) to check for convergence of the model. Rhat did not exceed 1.1 for any of the parameters in any model, indicating that the models converged 6 . To prevent divergent transitions, we adapted the initial step size, target acceptance rate and maximum treedepth where necessary. We centred and scaled all independent variables to mean 0 and standard deviation 1 to speed up convergence.
Chains of all models that we ran mixed well and converged in less than 100 iterations. For the main models, we used a burn-in period of 1000 iterations and an additional 2,500 iterations after burn-in per chain, giving a total of 10,000 iterations (2500 per chain) for the analyses.
For each model, the proportion of explained variance (R 2 ) was calculated following Gelman and Hill 6 : where ε are the model residuals (including all samples after warmup) and y are the observed growth or survival of all observations.

All species responses
Here we provide Figure S1, which visualises all species responses to shade and drought. We also provide Tables S1-S2 with the parameter estimates and 95% credible intervals for all species. Inter−annual drought index (−mm) (f) Table S1 Mean and 95% CI of responses in terms of relative growth rates to shade (β1), spatial drought (β2) and inter-annual drought (β3) for all species. Further shown are the number of observations of a species in years with and without shade estimates, mean RGR (β0) and the response to ln(height) (β4). Bold and underlined values indicate significant responses to shade or drought, i.e. the 95% credible interval (CI) excluded zero. Underlined values indicate that the CI of β0 or β4 excluded zero.  Table S2 Mean and 95% CI of responses in terms of first-year survival to shade (β1), spatial drought (β2) and inter-annual drought (β3) for all species. Further shown are the number of observations of a species in years with and without shade estimates, the survival constant (β0) and the response to ln(height) (β4). Bold and underlined values indicate significant responses to shade or drought, i.e. the 95% credible interval (CI) excluded zero. Underlined values indicate that the CI of β0 or β4 excluded zero.