We experimentally tested our hypotheses on five grass species commonly found in riparian zones of south-eastern Australia, particularly northern Victoria. This region encompasses the southernmost river systems of the Murray-Darling Basin. Major waterways include the Goulburn, Campaspe and Loddon Rivers, which all flow into the Murray River. Rainfall is variable across the region, broadly ranging from 200 to 600 mm per annum (Bureau of Meteorology 2020). Grasses are a dominant structural component of many riparian communities of the waterways in this region, particularly on mid to upper banks (Roberts and Ludwig 2006; Cottingham et al. 2010).
Five grass species were selected for this study: Bromus catharticus (prairie grass), Dactylis glomerata (cocksfoot), Lolium perenne (perennial rye grass), Rytidosperma caespitosum (common wallaby grass) and Poa labillardierei (common tussock grass). These species are frequently recorded in riparian surveys along major rivers in the region, particularly downstream of dams (Greet et al. 2012, 2013a; Jones and Mole 2018). The first three species are undesirable exotic tufted pasture grasses widespread throughout south-eastern Australia, particularly in disturbed areas, while the latter two are desirable native species. Bromus catharticus is an annual or short-lived perennial and the remaining four species are perennial. Although all species occur both within and outside of riparian areas, P. labillardierei is particularly common along streams and alluvial flats, as well as being a dominant species of lowland temperate grasslands and grassy woodlands, and subalpine grasslands (Birch et al. 2015; Clarke 2015; VicFlora 2019).
The five species were chosen as a representative set of common grasses with tufted growth forms that includes both exotic and native species and which are also likely to show a gradation of tolerance to submergence duration. Three of the species, B. catharticus, D. glomerata and R. caespitosum, are considered to be terrestrial dry species, indicating that they occur more widely in the landscape but can invade or persist in riparian zones (Casanova 2011). However, experimental studies have shown that these species can exhibit a degree of tolerance to complete submergence. For example, Striker and Ploschuk (2018) found B. catharticus and D. glomerata survived five days of complete submergence, but growth declined during both the submergence and subsequent recovery periods. Similarly, Kitanovic (2019) found D. glomerata and R. caespitosum survived submergence for 35 days during late winter and early spring with 100% survival, while B. catharticus had a 50% survival rate, although plant growth declined with increasing submergence duration. Lolium perenne has been previously classified as a terrestrial damp species in northern Victorian wetlands (Reid and Quinn 2004) and can grow and survive under waterlogged conditions (McFarlane et al. 2003). In addition, Banach et al. (2009) found it exhibited a 100% survival rate and increased leaf biomass after 3 weeks of submergence, with survival rates dropping by 20–30% after 6 weeks of submergence. Poa labillardierei has been previously classified as a terrestrial damp species (Gehrig and Nicol 2010), indicating it can germinate and establish on saturated or damp ground but cannot tolerate flooding in a vegetative state (Casanova 2011). Recent experiments have shown that it can survive 35 days of submergence in experimental conditions (Kitanovic 2019).
Seeds for the experiment were either sourced from a commercial seed supplier (R. caespitosum, L. perenne and D. glomerata) or collected in the field (P. labillardierei and B. catharticus) next to the Campaspe River in northern Victoria, near the township of Rochester (− 36.380974, 144.708964). Soil texture analysis conducted within two separate studies at riparian sites within the southern Murray-Darling Basin showed that soils were dominated by sand and sandy loam to a depth of 0.9 m (Hao et al. 2017; Hu et al. 2017). In April 2018, seeds were germinated in trays containing seed raising mix in a heated glasshouse with a temperature range of 18–25 °C and were kept moist via mist irrigation. After 1 month, 80 seedlings of each species were planted into individual 1.9 L pots containing a 7:1 sand to topsoil mix with a 3–5 cm base layer of pine bark. The pots were placed in an unheated polytunnel for 7 months to allow the plants to establish. Plants were watered for four minutes three times a day and fertilised with 0.5 g/L of fertiliser (N/P/K = 20:8.7:16.6) once a month for the first 3 months and then once every 2 months for the remaining period.
The experiment was conducted in a covered outdoor area in Burnley, Melbourne, Victoria (− 37.828299, 145.020861). The experimental design comprised 16 water tanks measuring 102 cm tall and 97.5 cm diameter. Each tank was randomly allocated to one of eight experimental treatments, resulting in two tanks per treatment. These treatments consisted of four levels of submergence: (1) no submergence (0 weeks), simulating a low summer flow more typical of pre-regulation conditions; (2) 2-week submergence pulses interspersed with 2-week dry periods; (3) 4 weeks of continual submergence; and (4) 8 weeks of continual submergence (Fig. 1). Each treatment also had two levels of shading (unshaded/shaded) to simulate lower light availability that may be experienced in turbid water. Full descriptions of each treatment are provided in Table 1. Tanks in the shaded treatments were covered with a shade cloth that reduced light penetration by 80%.
For each species, 54 plants of a similar size were selected for the experiment, with six randomly chosen for pre-treatment biomass harvesting (see below) and the remaining 48 allocated to one of the eight experimental treatments. Each treatment therefore comprised six replicate plants per species, split across two tanks, with each tank containing fifteen plants (three of each species). The exception was L. perenne, where only 50 plants were available; for this species, four plants (rather than six) were selected for pre-treatment biomass harvesting while two treatments (the unshaded 0-week and shaded 2-week treatments) received one fewer plant each (i.e. five rather than six replicate plants). Plants were placed on a platform inside each tank, which was positioned approximately halfway up the tank.
The experiment commenced on February 15, 2019, when tanks were filled to the appropriate levels with tap water and ran for 8 weeks. Water levels were approximately 45 cm above the level of the substrate within the pots, resulting in plants being submerged by an average of approximately 18 cm, depending on initial height. At approximately 2-week intervals, each plant was measured for height, defined as the longest green section of leaf. Once green leaf material was no longer evident, plants were recorded as dead. At the end of the experiment, plants recorded as dead were placed back in the polytunnel, watered regularly and monitored for recovery (none recovered). The remaining live plants were harvested for biomass assessment. For each plant, leaf and root material was separated, gently washed of soil, dried for at least 72 h at 60 °C and weighed.
The mean daily maximum air temperature during the 8 weeks of the experiment was 24.4 °C (range 15.6–38.1 °C) while the mean daily minimum was 14.6 °C (range 7.4–25.1 °C), measured at the Melbourne (Olympic Park) weather station (ID086338) located 3.5 km to the west (Bureau of Meteorology 2020). Water temperature, pH, electrical conductivity and dissolved oxygen were measured in each tank on two occasions during the fourth and fifth weeks of the experiment. Dissolved oxygen ranged between 2.98 and 8.04 mg/L, electrical conductivity ranged between 0.065 and 0.010 mS/cm and pH ranged between 5.91 and 6.95. Dissolved oxygen was lower in the shaded compared to unshaded 4-week and 8-week submergence treatments. Mean water temperatures across all tanks measured on the two occasions were 17.1 °C and 23.1 °C. As a comparison, in rivers of south-eastern Australia water temperatures generally show a marked difference between winter (generally less than 10 °C) and summer (22—26 °C) (Online Resource 1).
Survival rates were assessed as the percentage of plants showing evidence of green leaf material. The log-rank test was used to test the null hypothesis that survival curves differed between treatments. Log-rank tests were constructed using the Kaplan–Meier method in R version 3.6.0 (R Core Team 2019), with the survival package version 2.38 (Therneau 2015). This approach takes into account uncensored data, where an experiment ends prior to the death of all individuals, so as to avoid underestimating the lifespans of these individuals (Pyke and Thompson 1986). However, due to the small number of replicate plants for each species in each treatment combination, comparisons were made separately between levels of shade and levels of submergence.
Growth was assessed using two variables: (1) height, defined as the longest green section of leaf (i.e. maximum green leaf height), measured at approximately 2-week intervals (days 0, 14, 29, 43 and 56 of the experiment), and (2) change in total biomass (roots and leaves combined), assessed at the end of the experiment on living plants. Trends in growth responses were also examined using relative height growth and rates of growth (cm per week) with both metrics showing very similar responses as height. Change in total biomass for each plant was calculated as the final biomass less the species mean biomass of the initially harvested plants (as described above). The height and biomass data were zero-inflated due to the high numbers of plants that died during the experiment, resulting in insufficient numbers of replicates for most species and treatments. As such, the analyses of growth (height and biomass) were performed on living plants only. Firstly, maximum green leaf height was compared, for living plants, between treatments at day 14. Secondly, maximum green leaf height was compared between treatments at day 29 for the three species that survived the first 4 weeks of inundation: L. perenne, R. caespitosum and P. labillardierei. Thirdly, maximum green leaf height and the change in total biomass at 8 weeks was analysed for P. labillardierei only, as most individuals of the other species did not survive in the submergence treatments.
Linear mixed effects models were used to evaluate the relationship of plant height and biomass with treatment. The data were explored visually to check for assumptions of normality with subsequent transformations applied where necessary (outlined below). All model covariates were binary or factor variables and were therefore not scaled prior to analyses. Treatments (shaded and submerged) and their interactions were included as fixed effects. We constructed linear mixed-effects models in R using the lmer function in package lme4 version 1.1–21 (Bates et al. 2015). The plant height models were constructed at different survey periods (2, 4 and 8 weeks) to compare heights at different stages during the experiment for those species with the majority of plants assessed as still alive. For comparisons at two weeks, the submergence treatment was treated as a two-level factor (i.e. 0 weeks vs. 2 weeks of submergence), with the 2-week, 4-week and 8-week treatments pooled because they were identical after 2 weeks. Similarly, for comparisons at 4 weeks, the submergence treatment was treated as a three-level factor (i.e. 0 weeks vs. 2 weeks vs. 4 weeks of submergence), with the 4-week and 8-week treatments pooled because they were identical after 4 weeks. Treatment tank identity was included as a random effect to account for variation between tanks. Fixed effect coefficients for the models indicated effect sizes for each of the treatment levels compared to controls, i.e. 0-week submergence and unshaded. Interactions were included for submergence and shade treatments. For the biomass model, the response variable data were log transformed to allow for model assumptions of normality. For this log-linear model, the exponentiated fixed effect coefficients provided the percentage increase in the response for that treatment compared to the control (percentage calculated as 10coeff − 1*100). Confidence intervals for each model were calculated using the confint function within the stats package. Post hoc contrasts to assess effects and significance between treatment factors were conducted on models using the emmeans function in the emmeans package version 1.4 (Lenth 2019), with significance level of 0.05.
All graphs were produced in R version 3.6.0 (R Core Team 2019), using either the base package or the ggplot2 package version 3.2.1 (Wickham 2016).