Flood disturbance affects morphology and reproduction of woody riparian plants

Study site

Our study was undertaken in the Yellingbo Nature Conservation Reserve located around 45 km east of Melbourne, Victoria, Australia (Fig. 2). The reserve is embedded in an agricultural landscape and is around 640 ha comprising narrow riparian zones bordering local creeks. Low-lying floodplains along the Cockatoo and Macclesfield Creeks, which were focus of our surveys, are dominated by ‘Sedge-rich Eucalyptus camphora Swamp’ community²⁶. These forests naturally experience seasonal to near-permanent inundation and vary in structure from open forest to woodland. The highly flood-tolerant mountain swamp gum Eucalyptus camphora is the sole overstorey species. The midstorey is dominated by thickets of woolly tea tree Leptospermum lanigerum and scented paperbark Melaleuca squarrosa, both of which are flood tolerant small trees or shrubs²⁷. The largest remnants of this forest type are found within the Yellingbo Nature Conservation Reserve where they suffer dieback as a result of past human alterations of local watercourses²⁸. The long-term survival of this threatened riparian forest likely depends on management interventions. Despite thorough documentation of declining tree and shrub condition, the ecology of the three major woody species is not well understood^29,30.

Survey design

We confined the survey area to elevations lower than 120 m above sea level as Eucalyptus camphora swamp does not occur above this elevation within the reserve²⁷. Only areas mapped as vegetation communities containing the studied species were included. The survey area was further limited to match the extent of a hydrological model (see below) and was in total 1.69 km².

In order to ensure that survey points were distributed across the hydrological gradient, we simulated different sized flooding events using a hydrologic model (described below). The spatial extents of these events were then used to classify the study area into four broad flooding categories. Flooding categories one, two and three comprised areas which were flooded by low, medium and high flow events, respectively. Flooding category one represents the wettest parts of the floodplain whereas categories two and three are less frequently flooded. Flooding category four contained the rarely inundated parts of the survey area that remained unflooded in the modelled flow events.

To equally represent all flooding categories, we used a stratified random sampling approach. We generated 40 random coordinates within the area represented by each category and four additional points per point were generated as spares in case some positions were unsuitable for sampling.

During field surveys, we visited locations by navigating with a handheld GPS device (Garmin etrx30) to the predefined points. From there, we surveyed the nearest individual of each of the species E. camphora, M. squarrosa and L. lanigerum and mapped their actual geographic position. If no tree was found within a radius of 10 m of a given sample point, we visited the closest point from the spare dataset instead. If no individual was present near any of the four closest spare points no tree was recorded at the location.

After visiting all of the original points (including the four extra points) additional points were generated randomly in the areas where the two shrub species were found during the course of previous sampling. We thereby increased sample size for each of these less widespread species to approximately 20 individuals per flooding category. We conducted all surveying and tree and shrub measurements from March to April 2018 to take advantage of low water levels and therefore best accessibility. In total, we sampled 292 trees comprising 133 E. camphora, 78 L. lanigerum and 84 M. squarrosa.

Tree surveys

Elongated stems are a major feature of woody plants defining their overall architecture. To characterise and compare growth habits, we measured diameter at breast height (DBH) and height (to highest live foliage) of each tree and shrub. In some cases (for 21/133 E. camphora and 1/84 L. lanigerum) visibility impairment precluded height measurement via clinometer. For multi-stemmed individuals, we counted all stems, measured their DBH, and determined height of the tallest one. To yield crown width we measured maximum crown diameter and perpendicular crown diameter for every sampled plant and calculated the mean.

Flooding and associated unstable, boggy substrates might deter trees from the usual vertical growth and force them into leaning positions. Thus, we recorded inclination angle of the main stem (at DBH level) relative to vertical using a protractor.

The emergence of epicormic sprouts, be it a symptom of stress or sign of recovery³¹, is a common reaction to disturbance and reflects a tree’s ability to regenerate vegetatively. We estimated epicormic growth using a scale from 0 to 3 indicating absent, scarce, common or abundant expression of epicormic growth³².

Using the same scale, we assessed sexual reproduction by estimating the combined relative abundance of reproductive structures, namely buds, flowers and capsules. Flowers indicate only current reproductive activity and not all species were flowering during the fieldwork campaign. Owing to serotiny and the long timespan for bud crop development, different developmental stages of reproductive structures (current and past reproduction) can appear simultaneously on a single tree.

Growth and reproduction may both be affected by plant condition, for which crown vigour has been proven as a suitable and rapid measure^22,31,33. For each sampled plant, we assessed crown vigour by visually estimating the proportion of the potential crown supporting live foliage to the nearest 5%.

Moreover, growth rates and tree shapes can be significantly influenced by competition. For each sampled tree or shrub, we therefore measured the distances to its nearest four neighbours, one in each compass quadrant and calculated the average nearest-neighbour distance. For E. camphora, only neighbouring trees were included, whereas for L. leptospermum and M. squarrosa, both trees and shrubs were considered neighbours.

Hydrologic modelling

The surveyed floodplain area has a very low elevational gradient such that floods are low energy and geomorphology does not vary greatly across the system. As such, we did not explicitly examine geomorphology in this study and focused on hydrology. After completing tree surveys, we determined local flood regime history for each study tree using the output of a grid-based, 2-dimensional hydrological model built in TUFLOW classic (www.tuflow.com), which was calibrated with recent water-level data from four sites within the study area. The model generated historic-flow series (1998–present) of water levels across the study area with a 5-m grid-cell spatial resolution and a daily temporal resolution. See Greet et al.³⁴ for more details.

We extracted water-level time series for each surveyed tree and shrub from the model output. Using the recorded coordinates, individual water-level data were extracted for the grid cell in which the respective tree or shrub was located. Some individuals that were located next to the stream were allocated to a grid-cell that the model designated as the stream channel, resulting in them being erroneously characterised as permanently inundated. In these cases, water level data was extracted for the eight surrounding cells. We then excluded those that were also permanently inundated, and the average of the remaining cells was used to create a water level time series for that individual.

To characterise the flood regime history for each tree, we considered water levels of zero as dry and values greater than zero as inundated. Therefore, the first day with a water level greater zero marked the start of a flooding event and the reduction to zero the end of the respective event. Consequently, the number of consecutive days of flooding defined the length of a flooding event.

We calculated the following flood regime metrics for the modelled 20-year period (1998–2018): mean and maximum length of flooding periods, mean and maximum length of dry spells (not inundated periods), the mean length of flooding periods during the growing season (November–June), the average number of flooding events per year and mean flooding depth. All variables were skewed and thus log-transformed before we tested for correlation (Online Resource 1, Fig. SM1). Although flood magnitude has been found to affect herbaceous riparian vegetation in other systems³⁵, we assumed the observed flooding magnitude, i.e. mean water levels (mean = 0.06 cm, median = 0.01 cm, max = 0.92 cm), to be less important for the relatively tall trees and shrubs studied here (Fig. 3b). We further assumed maximum values to be less influential for tree and shrub growth over long periods. Hence, we selected two contrasting aspects characterizing long-term flood regime as predictors for further statistical analysis. These were the mean length of flooding events and the average number of flooding events per year representing duration and frequency of flooding. They were not strongly correlated with each other (Pearson correlation coefficient = 0.18). Both flooding duration and frequency have been shown to impact tree development in riparian ecosystems^36,37.

14 out of 292 sampled trees from across the study area were excluded from statistical analyses due to model outputs suggesting unrealistic high flood duration (i.e. mean inundation duration > 500 days) or frequency (i.e. > 300 events), likely owing to errors of local topography representation based on our field observations.

Statistical analysis

We performed multiple regression separately for each of the three species to:

1. 1.

   Assess the strength of relationships between flood regime (flood frequency, flood duration) and tree and shrub morphology (DBH of main stem, height, crown width, stem number, leaning and crown extent); and flood regime and reproductive strategy (the extent of sexual reproduction and epicormic growth), thereby testing hypothesis 1 and 2 (H1 and H2); and

2. 2.

   assess the relationships between morphology and both reproduction types, testing hypothesis 3 (H3).

For each analysis we used hierarchical partitioning to identify those variables which independently explained the most variance in morphology and reproduction, respectively.

First, we tested how much variation in morphology and reproductive strategy variables was explained by each of the two flood regime variables (H1 and H2). We fit 8 generalised linear models (response variables: main stream DBH, height and crown width, stem number, leaning, crown extent, sexual reproduction, and epicormic growth; predictor variables: flood frequency, flood duration). We chose the appropriate distribution used in the linear model for each variable (Table 1). Beta regression was undertaken using the betareg package³⁸ and ordinal regression using the MASS package³⁹.

We initially included the average nearest neighbour distance (a surrogate for competition) in models predicting morphology variables (H1). However, we later omitted this additional predictor as it generally did not increase the proportion of explained variance (Online Resource 2, Table SM1).

To assess how much variation in reproductive strategy variables was explained by morphology variables (H3), for each species, we calculated two additional linear models for the response variables of sexual reproduction and epicormic growth with each six predictor variables (main stem DBH, height, crown width, stem number, leaning and crown extent). Both of these models used a binomial distribution adapted for ordered factors.

For each model, we used hierarchical partitioning of log-likelihood values using the hier.part package⁴⁰ to determine the proportion of explained variance explained independently by each predictor variable⁴¹. This method allows identification of variables that have a strong independent correlation with the dependent variable, in contrast to variables that have little independent effect but have a high correlation with the dependent variable resulting from joint correlation with other predictor variables. Variables that independently explained a larger proportion of variance than could be explained by chance were identified by comparison of the observed value of independent contribution to explained variance (I) to a population of Is from 1000 randomizations of the data matrix. Significance was accepted at the upper 95% confidence limit (Z score > 1.65: Mac Nally⁴², Mac Nally and Walsh⁴⁰).

To assess the goodness of fit for each model, we present R² or pseudo-R² values (according to Nagelkerke using the DescTools package: Signorell et al.⁴³) for ordinal regression and Ferrari and Cribari-Neto⁶² for beta regression, respectively. We considered variables with a total contribution to explained variance (i.e. proportion explained × R²) > 0.05 to be influential variables and the direction of their effect important.

Lastly, we performed a PCA analyses and ordination to assess associations between different morphology attributes and reproduction variables across all species (H3).

All statistical analysis was performed in R version 3.5.0⁴⁴.

Source: Ecology - nature.com