Key features of the seed germination response to high temperatures


Author for correspondence: tel +64 3 3591703; email


The responses of seed germination rates (GRs) and final germination percentages to soil temperature follow trends similar to those for many plant processes. There is an optimum temperature for seed germination (To), which is defined as the soil temperature at which the highest germination percentage is achieved by a seed population in the shortest possible period of time (Mayer & Poljakoff-Mayber, 1975). At both sub-optimal (< To) and supra-optimal (> To) temperatures, the reductions in the GR and germination percentage scale with the amount by which the soil temperature is less than or greater than To, respectively, until threshold temperatures (base temperature Tb and ceiling temperature Tc) are reached at which germination is completely prevented (Fig. 1).

Figure 1.

Predicted relationship between temperature and germination rate (GR–T) using the hydrothermal time (HTT) model for Daucus carota (carrot) assuming the rate of thermal time accumulation is (a) constant or (b) increases above the optimum germination temperature (To). On both panels, predictions are shown for the fifth (dotted line), 50th (solid line) and 95th (dashed line) percentiles. Parameter values for the HTT models are taken from Watt et al. (2011). Base temperature (Tb) and ceiling temperature (Tc) occur where the plotted lines for GR–T intersect the x axis at the left-hand and right-hand sides, respectively.

These responses of seed germination to temperature and their underlying physiology have been thoroughly described in the literature and are very familiar to seed scientists. Specific aspects, such as thermoinhibition, when seeds fail to germinate at high soil temperatures, are of great importance for commercial crops and have been investigated for many crop species. However, there are two key features of germination commonly observed within seed germination studies that are nonetheless seldom discussed in the literature. First, To is often not a specific temperature, but rather a range of temperatures seen as a broad, curvilinear peak in the plot of GR vs temperature (GR–T). Second, the GR–T response varies between the seed percentiles in the population, where the order in which seeds germinate is specified by the seed percentile (G), such that G = 1 is the first percentile of seeds to germinate, G = 2 is the second percentile to germinate, and so on, up to G = 100, which represents complete germination of a seed population. Fast-germinating seeds (the lowest percentiles) often have higher values for To and Tc, compared with slower germinating seeds (the higher percentiles), which reach their To and Tc at lower temperatures.

From an extensive literature review of GR–T responses, we identify the extent to which these two key features are shown by the data. We discuss two common mathematical models that have been used to describe the GR response to temperature, and which model more accurately describes the important features of GR–T. Lastly, we discuss how this more accurate model aligns with our physiological and ecological understanding of the seed germination process.

Germination data

The data used for this study were assembled from the published literature. Studies were selected that fulfilled all of the following criteria: (1) germination took place under constant controlled conditions of temperature and water potential; (2) graphs showed regularly spaced measurements of GR for a wide range of temperatures that clearly straddled To; and (3) measurements were at sufficiently close temperature intervals that the shape of the inflexion of GR–T at To could be determined. It is worth noting that we reviewed a large number of studies that did not fulfil all of these criteria.

In total, data from 37 species were used (Table 1 and Supporting Information Table S1). Species were defined as crop, wild or weed, depending on the seed source used in the published research. In some cases, species were partially domesticated and were classified as wild/crop. Species were also classified with regard to growth habit according to the USDA & NRCS (2012) classification and their botanical family. As might be expected, the majority of germination studies were of crop species, and this resulted in a high representation of forbs in the growth habit categories, and Fabaceae, Poaceae and Amaranthaceae in the botanical family categories.

Table 1. Categorization of the 37 species into dataset (D1, D2), type (crop/wild/weed), growth habit and family
SpeciesTypeGrowth habitFamilyReferenceD1 Shape of GR–TD2 To change with G
  1. For each species, the shape of the GR–T response is categorized as curvilinear (Curvi.), broken or indeterminate (Indet.), and the scaling of To with percentile (G) is categorized as static, negative (Neg.) or positive (Pos.).

Aeschynomene americana CropPerennial sub-shrubFabaceaeMcDonald (2002)x  x  
Allium cepa CropAnnual forbAmaryllidaceaeRowse & Finch-Savage (2003)x     
Beta vulgaris CropAnnual forbAmaranthaceaeGummerson (1986)x   x 
Chamaecrista rotundifolia CropAnnual forbFabaceaeMcDonald (2002)x  x  
Cicer arietinum CropAnnual forbFabaceaeCovell et al. (1986)  xx  
Clausena excavata WeedTreeRutaceaeVieira et al. (2010)x     
Clitoria ternatea CropPerennial forbFabaceaeMcDonald (2002)x    x
Daucus carota CropAnnual forbApiaceaeRowse & Finch-Savage (2003)x     
Desmanthus virgatus CropPerennial forbFabaceaeMcDonald (2002)x    x
Dolichos biflorus CropAnnual forbFabaceaeLaboriau & Pacheco (1978)x     
Elymus elymoides Crop/wildGraminoidPoaceaeHardegree et al. (1999)x   x 
Elymus lanceolatus Crop/wildGraminoidPoaceaeHardegree et al. (1999)x   x 
Enterolobium contortisiliquum Crop/wildTreeFabaceaede Lima et al. (1997)x     
Eurotia lanata Crop/wildPerennial sub-shrubAmaranthaceaeWang et al. (2005)x     
Glycine latifolia CropPerennial forbFabaceaeMcDonald (2002)x    x
Glycine max CropAnnual forbFabaceaeCovell et al. (1986)  xx  
Indigo feraschimperi CropPerennial sub-shrubFabaceaeMcDonald (2002)x   x 
Lablab purpureus CropAnnual–perennial forbFabaceaeMcDonald (2002)x    x
Lens culinaris CropAnnual forbFabaceaeCovell et al. (1986)  xx  
Leucaena leucocephala CropTreeFabaceaeMcDonald (2002)x   x 
Lotononis bainesii CropPerennial forbFabaceaeMcDonald (2002)x    x
Macroptilium atropurpureum CropPerennial forbFabaceaeMcDonald (2002)x    x
Macroptilium bracteatum CropPerennial forbFabaceaeMcDonald (2002) x   x
Manfreda brachystachya WildPerennial forbAgavaceaeOrozco-Segovia et al. (1996)x   x 
Orobanche aegyptica WeedAnnual forbOrobancheaceaeKebreab & Murdoch (1999) x  x 
Papaver somniferum CropAnnual forbPapaveraceaeKamkar et al. (2012)x   x 
Pennisetum typhoides CropGraminoidPoaceaeGarcia-Huidobro et al. (1982) x    
Pinus densiflora WildTreePinaceaeWashitani & Saeki (1986)x   x 
Pinus radiata CropTreePinaceaeBloomberg et al. (2009)x   x 
Poa sandbergii Crop/wildGraminoidPoaceaeHardegree et al. (1999)x   x 
Pseudoroegneria spicata Crop/wildGraminoidPoaceaeHardegree et al. (1999)x   x 
Solanum tuberosum CropPerennial forbSolanaceaeAlvarado & Bradford (2002)x   x 
Stellaria media WeedAnnual forbCaryophyllaceaeGrundy et al. (2000)x     
Stylosanthes scabra CropPerennial sub-shrubFabaceaeMcDonald (2002)x  x  
Stylosanthes seabrana CropPerennial sub-shrubFabaceaeMcDonald (2002)x    x
Taraxacum officinale WeedPerennial forbAsteraceaeWashitani (1984)x   x 
Vigna unguiculata CropAnnual forbFabaceaeCovell et al. (1986)  x x 

Two datasets were assembled. The first dataset (D1) included data from studies in which graphs depicted the shape of GR–T for the 50th percentile (GR (50)). The second dataset (D2), which was a subset of the first, included data from studies in which graphs depicted GR–T for a range of percentiles. In almost all cases within the two datasets, germination was undertaken under moist conditions (water potential ≈ 0 MPa).

Visual inspection of the graphs in the datasets was undertaken to examine the GR–T response across the temperature range. For D1, the peak of GR–T at To was classified as either occurring at the intersection of two straight lines (broken stick) or as a broader curvilinear response. For D2, the inflexion of GR–T at To between the germination percentiles was classified as either occurring at the same temperature for all percentiles or occurring at varying temperatures that showed either a positive or negative relationship with germination percentile.

Data from 37 species were examined in D1. The shape of the GR–T response across the optimal temperature range was indeterminate in four of the 37 species (11%). For the remaining 33 species, a total of 30 species (90%) exhibited a curvilinear response across the optimal temperature range, whereas the remaining three species (10%) showed a broken stick response (Table 1).

In D2, there were 29 species that had sufficiently detailed GR–T responses to evaluate the extent to which To remained constant across germination percentiles (G). Of these 29 species, 15 species (51.7%) showed a decline in To with increasing G (Table 1). The relationship between To and G was positive in eight species (27.6%) and exhibited no discernible trend in the remaining six species (20.7%) (Table 1). All of the eight species that showed a positive relationship between To and G were tropical forage legumes with responses documented in a single study (McDonald, 2002).

Using mechanistic models to explain germination response to temperature

Early attempts to use mechanistic models to describe the germination response to temperature focused on the effects of temperature on reaction kinetics in the germination process. At sub-optimal temperatures, germination progress scales more or less linearly with thermal time, allowing GR to be plotted as a linear function of temperature. The decline in GR at supra-optimal temperatures can be plotted as an inversion of the sub-optimal relationship, resulting in a distinct ‘broken-stick’ plot of GR over the sub- and supra-optimal temperature range (Garcia-Huidobro et al., 1982). The implicit assumption in this model is that the reaction kinetics of germination are directly and linearly inhibited by increasing temperatures above To.

An advance on this approach was proposed by Alvarado & Bradford (2002), using a modification of the hydrothermal time (HTT) model (Gummerson, 1986). The HTT model is a threshold model that simultaneously accounts for germination percentages and GRs of a seed population (Gummerson, 1986). The GR for a specific seed percentile (GR(G)) at sub-optimal temperatures is specified by:

display math(Eqn 1)

where Ψ (MPa) is constant water potential of the soil or other surrounding medium; Ψb(G) (MPa), base soil water potential, a threshold value analogous to the base temperature below which the radicle does not emerge to complete germination; T (°C), constant temperature of the soil, Tb (°C), base temperature; θHT (MPa °C d), hydrothermal time constant for the seed population. The value of Ψb(G ) is defined by a unimodal frequency distribution, such that the low seed percentiles (G = 1, 2, 3..) have the most negative values for Ψb(G ), and the highest seed percentiles have the least negative values for Ψb(G ). For any value of Ψ, the term Ψ – Ψb(G ) is therefore largest for the lowest seed percentiles, resulting in the highest (fastest) values for GR. Conversely, Ψ – Ψb(G ) is least for the highest seed percentiles, resulting in the lowest (slowest) values for GR.

Modifications have been made to the HTT model to describe the observed decline in GR at temperatures above the optimum, as follows. A postulated increase in seed base soil water potential at supra-optimal temperatures can be modelled as a simple rightwards shift in the location of the frequency distribution for Ψb(G). The magnitude of the shift scales with the supra-optimal temperature and is specified by k(T – To), where k is a positive constant (Alvarado & Bradford, 2002). Addition of this term to Eqn 1 results in:

display math(Eqn 2)

so that both the term (Ψ – (Ψb(G) + k(– To))) and, consequently, GR(G) scale negatively with T.

The assumption made about thermal time accumulation at supra-optimal temperatures has a marked impact on the shape of the GR–T curves resulting from Eqn 2. If it is assumed that the rate of thermal time accumulation reaches a maximum at To, then T within the term (– Tb) in Eqn 2 must equal To throughout the supra-optimal temperature range. Under this assumption, a plot of GR–T will show To to occur at the convergence point of two straight lines (Fig. 1a, broken stick model) that will have an identical To for all germination percentiles (Alvarado & Bradford, 2002).

There are two key differences from this predicted behaviour if the rate of thermal time accumulation is, instead, assumed to continue to increase linearly with respect to temperature throughout the supra-optimal range (Rowse & Finch-Savage, 2003). First, a plot of GR–T will show a broader curvilinear peak above To, as increases in GR caused by increases in the term (– Tb) are gradually offset by decreases in the term (Ψ – (Ψb(G) + k(– To))). Second, the value of To will vary across the seed population with To declining with higher germination percentiles (i.e. To scales negatively with G). Lower germination percentiles are predicted to have a higher To as the impact of increases in the term (– Tb) on GR compensates for decreases in the term (Ψ – (Ψb(G) + k(– To))) to a greater extent than at higher germination percentiles. Using parameter values for Eqn 2, described in Watt et al. (2011), the resulting GR–T responses for these two scenarios are shown for the germination of carrot seed in Fig. 1.

What conclusions may be drawn from the data?

The large majority of the data showed a curvilinear peak in GR–T, which is consistent with the HTT model that assumes an increasing rate of thermal time accumulation above To. This assumption implies that the inhibition of germination kinetics by the thermal denaturation of enzymes (a commonly offered explanation for declining GR at supra-optimal temperatures) does not occur in the range between To and Tc. A more realistic explanation for declining GR is the upwards shift in base seed water potential for germination (Ψb(G)) in the Alvarado and Bradford model (Eqn 2). This may have a mechanistic basis in that the value of Ψb(G) is posited to relate to physical constraints to embryo growth within the seed (Bradford, 2002; Finch-Savage & Leubner-Metzger, 2006). Both increased physical constraints to embryo growth and an upwards shift in Ψb(G) are associated with thermoinhibition at supra-optimal temperatures (Bradford & Somasco, 1994).

The second consequence of the HTT model with increasing thermal time accumulation above To is a predicted decline in To with increasing seed percentile. Although the evidence from D2 was less conclusive on this point, 52% of species showed negative scaling between To and G. Negative scaling between To and G may be a bet-hedging response to the asymmetry of risks for a seed if it germinates earlier or later than the optimum time for seedling establishment. This applies during summertime or dry season conditions when soil temperatures are frequently supra-optimal. After summer rainfall, seed populations may need to germinate rapidly in order to utilize rapidly evaporating moisture in the surface soil (Finch-Savage et al., 2010). However, early germination under summertime conditions carries a high risk of death by desiccation, if there is no further rainfall after the event that triggered germination. By contrast, delaying germination until safer, cooler conditions occur carries a lesser risk of germinating at the same time as many other seed populations and therefore succumbing to competition (Watt et al., 2011). Negative scaling between To and G allows part of the population to exploit opportunities for rapid early germination, while a large part of the population delays germination until cooler conditions with a low risk of desiccation.

McNamara et al. (2011) observed that, in relation to phenology, there is still ‘limited understanding of the relationship between cues and optimal timing, and especially about how this relationship will be affected by environmental changes’. An accurate plot of the GR–T response across percentiles for a seed population can provide a simple yet informative indicator of how seed populations respond to the unpredictable, interacting cues of soil temperature and soil moisture, in order to achieve optimal germination timing under summertime or dry season conditions.

In conclusion, the majority of studies examined here show that the GR–T response has a broad curvilinear inflexion occurring across a relatively wide temperature range. In many studies, To was found to vary with germination percentiles. Both of these features are widespread across taxa. Assuming an increase in the rate of thermal time accumulation above To, the HTT model can accurately simulate these important features of GR–T, whilst also providing insight into their physiological and ecological significance.


Professor Bill Finch-Savage (University of Warwick, UK) provided useful advice in the development of this paper. Johanna Bloomberg assisted with accessing and cataloguing the references used in this study.