Dormancy and stratification in the soil
To assess dormancy and dormancy loss in the soil (stratification), seeds from the two M. balbisiana accessions (bal106 and bal115) were incubated at alternating 35°C in the light for 6 hours and 20°C in the dark for 18 hours (based on Stotzky, Cox & Goos 1961; Kallow et al. 2021). Additionally, seeds were buried at two of the cooler Simulated-NHs (Spring/Exposed and Leuven/Exposed) in March 2019 and exhumed each month for a total of three months and placed in the incubator conditions described. Seeds were enclosed in small nylon meshed bags and were buried at 7 cm depth in March 2019. Two replicates of 30 seeds were used for each treatment and accession. For incubation, seeds were sown on moist sand (50 g fine sand, 14 ml deionized water) in Petri dishes (9 cm diameter), sealed in plastic bags, at Meise Botanic Garden. Germinated seedlings of these were recorded and removed weekly for two months. Germination was counted as radicle emergence to 2 mm.
Survival in the soil
At the end of the experiment, seeds planted at shaded Simulated-NHs (locations that showed no emergent seedlings in results) were extracted from pots and tested for viability using a tetrazolium chloride staining test or by incubation. Seeds sown in both 2019 and 2020 were used. Pots containing seeds were removed, seeds were separated from compost by sieving under running water. Extracted seeds from the first replicate of each treatment were tested for viability with tetrazolium chloride (TTC) with a maximum of 20 seeds. The TTC tests were carried out on embryos carefully extracted from seeds, using 0.5 % TTC solution buffered to pH 7 (method described by Kallow et al. 2020), incubated at 24°C for 24 hours in the dark. Extracted seeds from the second replicate were sowed in Petri dishes on top of moist sterilized potting compost (Peltracom, composition: 70 % white peat and 30 % black peat; pH: 5.5- 6.5; particle size: 0-10 mm) and placed in an incubator at a 24-hour cycled temperature pattern (based on temperature readings from Los Baños, the Philippines - sun exposed site, Fig. 1a). Seeds in the incubator were monitored every two weeks, for a maximum of five weeks. If very few seeds were extracted from the soil for a replicate, TTC tests were prioritized above incubator germination tests.
Data analysis
We calculated summary indices for germination tests using theGerminaR package in R (Lozano-Isla, Benites-Alfaro & Pompelli 2019). These included final germination percentage (GRP, %) (Labouriau & Valadares 1983), mean germination time (MGT, days) (Czabator 1962) and synchronization index (SYN, simultaneous germination within a replicate = 1, no overlap between seeds of a replicate = 0) (Primack 1980; Ranal & Santana 2006).
We summarized data logger readings to extract the mean, maximum, minimum, and range (maximum-minimum) during experimental ‘cue periods’. In semi-NH this was the whole experimental period (55 days). In simulated-NHs for each year we summarized data filtered to include readings during ‘cue periods’ - these were a period prior to the MGT of each replicate, set differently for shallow and deep planting based on the mean MGT for shallow and deep planting (shallow=32 days, deep=39 days). For locations with no germination, we used the first 39 days after sowing. We summarized logger readings for cue periods by calculating mean, maximum, minimum, range temperatures and light intensity readings (in 2020). We removed light intensity readings in the dark (<40 lux), to account for changes in day/night length during seasons and did not calculate minimum light intensity as it would be the same in any site.
We then carried out redundancy analysis (RDA) of summarized logger data in cue periods, scaled to unit variance, against a corresponding matrix of factorial variables (species, compartment, exposure, depth) using thevegan R package (Oksanen et al. 2019). The minimum adequate RDA was found by comparison of Akaike Information Criterion (AIC), by adding variables to the minimum model.
Final germination, for each experiment, was assessed using counts of germinated seeds at the end of experiments against seeds that did not germinate, thus accounting for sample size variance. Sample sizes were adjusted in the analysis to only include viable seeds, estimated from ER results of that year, using the formula:
\begin{equation} adjusted\ sample=seeds\ sown\ x\ viability\ of\ year\ sown\nonumber \\ \end{equation}
These data were used in generalized linear modelling (GLMs) of binomial data with logit link function. If overdispersion was present in binomial GLMs we used quasibinomial error structure. Minimum adequate models (MAM) were produced by removing variables from maximum models after comparisons with ANOVA and X 2 test. Estimated marginal means were made from models and used for post-hoc analysis using the emmeans R package (Lenth 2020). Additionally, we used GLMs for factorial variables on MGT and SYN using gamma error structure. We assessed the effect of environmental variables in cue periods on binomial germination outcomes using GLMs as described. Following this we tested variables for breakpoints to assess temperature thresholds or optimums. We did this on GLMs produced for each microclimate variable separately by using the algorithm with bootstrapping in thesegmented R package (Muggeo 2017). We used starting points estimated by plotting GRP against microclimate variables and trend lines using the non-linear regression method Loess.
We compared viability of seed survival of seeds exhumed after 1-2 years in the soil from the incubation test, TTC test, and maximum achieved from Simulated-NHs, against the original ER viability. We again used a binomial GLM for this, as described, and post-hoc contrasts against original ER with Dunnet tests. All analysis was performed in R (R Core Team 2020).

Results

  1. Effects of exposure to solar radiation and substrate on seed germination
In the semi-NH exposure to direct solar radiation (sun) rather than shade, significantly increased germination rates (z=20.963, p<0.001, Fig. 1a). There was no significant effect of substrate on germination outcome according to the GLM. Final germination percentage in the sun was 72% (soil), 71% (sand) whereas in the shade it was 19% (soil), 12% (sand).
The MGT in the shade was 22±2 days (mean, standard deviation, used hereafter), whereas in the sun it was slightly faster germinating in 18±2 days, albeit only few seeds germinated in the shade. Synchronization index in the sun was 0.25±0.10 and in the shade 0.27±0.07.