Predicting the seed germination response of four rare Grevillea species (Proteaceae) to current and future temperatures

Study species

Grevillea iaspicula McGill, G. masonii Olde & Marriott, G. rivularis L.A.S.Johnson & MacGill., and G. wilkinsonii Makinson belong to two higher-order groups proposed by Makinson (2000), namely, Pteridifolia and Floribunda. These species are listed as either Endangered (G. iaspicula and G. masonii) or Critically Endangered (G. rivularis and G. wilkinsonii) under the EPBC Act (as of July 2025). All are threatened because of their highly restricted distributions in New South Wales, with recorded Extent of Occurrence and Area of Occupancy ranging between 4 and 270 km² and 4 and 56 km² respectively. Each species has fewer than 1000 known extant individuals, occurring across no more than eight populations. Regeneration can occur only from seed for all species except G. masonii, which has a lignotuber. Seed germination research, propagation and translocation have been recommended as future conservation management actions for each species. For these reasons, the species were selected for this study.

Seeds of Grevillea iaspicula, G. masonii, G. rivularis and G. wilkinsonii were collected from wild populations across the eastern coast of Australia, between Lawrence (29.49°S, 153.10°E) and Tumut (35.31°S, 148.23°E), NSW. This area experiences mean annual minimum and maximum temperatures between 8.3°C and 25.9°C respectively, and mean annual rainfall between 624 and 997 mm. The species were found in open, dry woodlands dominated by eucalypt species or in open shrubland. Mature seeds were collected post-dispersal by placing fine mesh bags over immature fruiting branches of representative plants from one population of each species, with collections made 4–6 weeks later. Seeds were processed and cleaned in a laboratory and placed in an environmentally controlled drying room (15°C and 15% RH) for at least 2 weeks. Sufficient Grevillea rivularis seeds could not be collected in 2024 because of the limited availability of seed from the only extant population. As a result, a historical seed collection from 2010 was used for this study. These seeds were stored at −20°C in vacuum-sealed foil packets and retrieved from storage for use. Prior to experimentation, seeds were equilibrated at ambient temperature and humidity for 24 h. A germination test was conducted on the stored seeds to assess potential viability loss after 14 years in cold storage; 100% viability was recorded (data not shown). Seeds of the remaining three species were collected in January 2024. For each collection, seeds were pooled representatively of seed production from multiple maternal plants from a single population, so as to give the required number of seeds for the experiment.

Experimental design

Experiments were conducted in May and June 2024. Because we were interested in the germination response to temperature, the seed coat was removed from all seeds with a scalpel and fine forceps to overcome physiological dormancy imposed by the seed coat. The bi-directional thermogradient plate (TGP) was set-up in a 36-cell half grid (eight cells along the x- and y-axis) so that only cells that received 12 h of light during the warm part of the diurnal cycle were used. Two species were run concurrently by using sealed 90 mm sterile polycarbonate bi-plates (Livingstone International, Australia), which have a built-in barrier to separate seeds of each species, and the plates were rotated a random amount every 2–3 days to overcome within-cell temperature differences. Seeds were sown onto a water agar medium (7 g L⁻¹, depth ~5 mm) and each cell contained 10 seeds of each species, resulting in a total of 360 seeds for each species. Germination was checked every 2–3 days for 5 weeks and was scored when the radicle emerged >2 mm from the seed. Post-experiment seed viability was not determined because visibly damaged or otherwise non-viable seeds were discarded when the seed coat was removed, and extreme temperatures on the TGP can affect seed viability differently from mild temperatures. TGP cell temperatures were recorded using an infra-red thermometer (ZyTemp TN408LC, Bacto Laboratories, Mount Pritchard, NSW, Australia) on the surface of the plate, during light and dark periods on the days that germination was checked. Final cell temperatures were calculated as an average of temperature readings over the experimental period.

Data analysis

All data analyses were performed in the R statistical platform (R Core Team 2021) and repeated for each species to determine how germination changes with future temperature predictions. Germination data were analysed following Collette et al. (2022), by using the open-access ThermoGradient R-notebook workflow (ver. 1.2) (Collette et al. 2024). The individual seed was the unit of replication for each TGP temperature cell. We interpolated and predicted final germination proportion and time to 50% germination (t₅₀) for all 36 temperatures by using generalised additive model (GAM) selection with a binomial (final proportion germination) or gaussian (t₅₀) distribution. The raw data and best model outputs were visualised and checked using quilt plots. Next, the decimal latitude and longitude coordinates of the local seed collection site for each species were utilised to obtain minimum and maximum monthly temperature data (t_min and t_max) under current and future climate scenarios. Current temperature predictions were modelled using WorldClim 2.1 and obtained from the WorldClim online database (https://www.worldclim.org/data/index.html). Future climate predictions for t_min and t_max were derived from the Coupled Model Intercomparison Project Phase 6 (CMIP6) models, also available from WorldClim (https://worldclim.org/data/cmip6/cmip6_clim2.5 m.html). We used the ‘middle of the road’ scenario (SSP2-4.5) and the ‘fossil-fuel development’ scenario (SSP5-8.5) to project germination data, representing both moderate and worst-case scenarios. Eight CMIP6 model variations for SSP2-4.5 and SSP5-8.5 were included for the years 2081–2100, with a spatial resolution of 2.5 min (approximately 4.5 km²). The monthly temperature data were averaged to generate the required datasets. Final germination proportions and t₅₀ were predicted for both current and future temperature scenarios by using the selected GAM.

Observed data

Complete germination (1.0) was achieved across a range of temperatures for all four species (Fig. 1). Nearly all G. masonii seeds (0.96) germinated within 30 days. Seeds of G. wilkinsonii and G. rivularis exhibited slightly higher final germination proportions at milder temperatures (10–15°C). G. wilkinsonii preferred constant and low-amplitude temperature regimes, whereas G. rivularis showed a slight preference for high-amplitude temperatures (Fig. 1). In contrast, G. iaspicula displayed a narrow germination response, favouring cooler temperatures (≤15–20°C) and, to a lesser extent, warmer temperatures including 33/17°C and 30/22°C.

Fig. 1.

Modelled data of final proportion germinated for (a, e, i) Grevillea iaspicula, (b, f, j) G. masonii, (c, g, k) G. rivularis and (d, h, l) G. wilkinsonii under multiple temperature regimes. Data were generated using a TGP, and modelled with generalised additive modelling. The first row (a–d) shows rasterised quilt plots of the modelled data, with day temperature (°C) on the x-axis, night temperature (°C) on the y-axis and final germination proportion on the z-axis. These plots show the best fitting models from Table 1. The middle row (e–h) shows the final germination proportion predicted into real world current and future predicted temperature for each month at the location of the source population for each species under the IPPC shared socio-economic pathway SSP2-4.5, for the years 2081–2100. The final row (i–l) shows the same data but for the SSP5-8.5 climate scenario for the years 2081–2100.

The t₅₀ was less than 5 days for temperatures above 15°C for G. masonii and G. rivularis (Fig. 2). G. wilkinsonii and G. iaspicula germinated over a longer period, with most temperatures resulting in a t₅₀ greater than 15 days (Fig. 2).

Fig. 2.

Modelled data of time to 50% germination (t₅₀) for (a, e, i) Grevillea iaspicula, (b, f, j) G. masonii, (c, g, k) G. rivularis and (d, h, l) G. wilkinsonii under multiple temperature regimes. Data were generated using a TGP, and modelled with generalised additive modelling. The first row (a–d) shows rasterised quilt plots of the modelled data, with day temperature (°C) on the x-axis, night temperature (°C) on the y-axis and time to 50% germination (t₅₀) on the z-axis. These plots show the best fitting models from Table 1. The middle row (e–h) shows the t₅₀ predicted into real world current and future predicted temperature for each month at the location of the source population for each species under the IPPC shared socio-economic pathway SSP2-4.5, for the years 2081–2100. The final row (i–l) shows the same data but for the SSP5-8.5 climate scenario for the years 2081–2100.

Model predictions

The models with the best fit are outlined in Table 1. Model fit for final germination proportion was good, with mean error and RMSE values ranging from 0.06 to 0.18 and from 0.01 to 0.14 respectively, and correlations between the observed and predicted data ranging from 0.76 to 0.99. G. rivularis had the poorest fit because of its highly variable but broad germination response across all temperatures.

When the data were modelled under current temperatures at the local site, final germination was predicted to be consistent year-round for G. masonii, whereas G. iaspicula showed higher germination predicted from late autumn to early spring (May to September; 0.89 ± 0.07 to 0.92 ± 0.07). G. rivularis was predicted to have its highest germination between spring and autumn (October to April; 0.82 ± 0.05 to 0.83 ± 0.05), and G. wilkinsonii showed a slight preference for winter temperatures.

Under future climate scenarios SSP2-4.5 and SSP5-8.5, germination was predicted to remain consistent with current climate predictions, except for increases in predicted germination at summer temperatures for G. iaspicula (December to February, SSP2-4.5: 0.50 ± 0.10 to 0.63 ± 0.09; SSP5-8.5: 0.45 ± 0.10 to 0.69 ± 0.09), and winter temperatures for G. rivularis (June to August, SSP2-4.5: 0.74 ± 0.06 to 0.78 ± 0.06; SSP5-8.5: 0.79 ± 0.05 to 0.81 ± 0.05).

Model fit for t₅₀ was good for G. rivularis and G. masonii, with mean error and RMSE values of 2.74–2.81 and 1.03–1.97 respectively, and correlations of 0.91–0.98. By contrast, poor model fit for t₅₀ was observed for G. wilkinsonii and G. iaspicula, with mean error and RMSE values of 4.59–5.64 and 4.16–4.74 respectively, and correlations of 0.58 and 0.55.

Under current climate, t₅₀ was predicted to be lowest during summer months and highest during winter months for all species. The t₅₀ predictions for G. masonii ranged across the year from 4.60 ± 0.81 to 6.83 ± 0.76 days under the current climate, and were faster by an average of 0.54 ± 0.5 and 0.43 ± 0.08 days under the SSP2-4.5 and SSP5-8.5 future scenarios respectively.

Similarly for G. iaspicula and G. wilkinsonii, t₅₀ was consistent throughout the year (18.29 ± 1.31 to 22.03 ± 5.13 days and 15.79 ± 1.41 to 21.69 ± 4.13 days, respectively) and was predicted to have a negligible change under both future climate scenarios. For G. rivularis, t₅₀ was >10 days for winter months and <10 days for every other month. Under the SSP2-4.5 and SSP5-8.5 future climate scenarios, t₅₀ was predicted to be faster every month by an average of 1.70 ± 0.11 d and 2.87 ± 0.18 days, respectively.