Mistletoes are hemiparasitic plants and keystone species in many ecosystems globally. Given predicted increases in drought frequency and intensity, mistletoes may be crucial for moderating drought impacts on community structure. Dependent on host vascular flows, mistletoes can succumb to stress when water availability falls, making them susceptible to mortality during drought. We counted mistletoe across >350,000 km² of south-eastern Australia and conducted standardised bird surveys between 2016 and 2021, spanning a major drought event in 2018-19. We aimed to identify predictors of mistletoe abundance and mortality and determine whether mistletoes might moderate drought impacts on woodland birds. Live mistletoe abundance varied with tree species composition, land use and presence of mistletoebirds. Mistletoe mortality was widespread, consistent with high 2018/19 summer temperatures, low 2019/20 summer rainfall and the interaction between summer temperatures and rainfall in 2019/20. The positive association between surviving mistletoes and woodland birds was greatest in the peak drought breeding seasons of 2018/19 and 2019/20, particularly for small residents and insectivores. Paradoxically, mistletoes could moderate drought impacts on birds, but are themselves vulnerable to drought-induced mortality. An improved understanding of the drivers and dynamics of mistletoe mortality is needed to address potential cascading trophic impacts associated with mistletoe die-off.

During 2019/20, we conducted habitat assessments at 2,111 monitoring sites spanning over 300,000 km² of south-eastern Australia (Figure 1). We selected monitoring sites in areas of woodland habitat deemed suitable for two Critically Endangered bird species: the regent honeyeater and swift parrot Lathamus discolor. We used a combination of MaxEnt habitat suitability models (Figures S1 and S2), expert field searches and the location of previous sightings to inform site locations. Since both bird species are habitat specialists²⁹, our sampling encompasses the highest quality remaining woodlands in south-eastern Australia.

Each monitoring site was a 50m radius (0.79 hectares) around a fixed GPS location to two metres accuracy. During one visit to each site in 2019/20, we recorded the habitat fixed effects detailed in Table 1. Our mistletoe counts focussed on the three most abundant mistletoe species (Family Loranthaceae) in the study range: box, needle-leaf and long-flowered. We conducted a 360° search of the canopy from each site centroid, deviating from this point where necessary to count accurately the number of live and dead mistletoe clumps present.

Table 1: Site-level and visit-level fixed effects obtained for identifying predictors of mistletoe abundance and health and to model the effect of mistletoe abundance on woodland bird abundance. For further information on the fixed effects, see Table S1.

Bird surveys

We conducted 9,012 point-count surveys at a total of 1,218 monitoring sites in the Austral spring/summer breeding season (August to January) between 2016 and 2021 (Table S2). Each survey was conducted by one of 15 professional ornithologists, with 86% of surveys completed by seven observers. Our rapid (5-minute) census, involving one minute of regent honeyeater song broadcast, was designed to maximise the detectability of such rare, nomadic habitat specialists by increasing the spatial extent of surveys without compromising detectability³¹. We recorded the maximum count of all bird species detected visually or aurally within a 50m radius of the fixed-point location during each site visit, along with a blossom score for each site. Observers remained at the site centroid as much as possible, but deviated where necessary to identify individual birds to species level or to obtain accurate counts of birds occupying heavily flowering trees near site boundaries. We did not include transient birds flying through or over study sites in the counts. The blossom score was a five-level factor (Table 1); a simple way of modelling variation in blossom abundance on nomadic species occupancy patterns³². To account for intra-seasonal variation in flowering phenology and associated changes in woodland bird distribution/abundance³³, we surveyed as many sites as possible (77%) twice; once in spring between August and October and again in early summer between November and January.

Climate data

We sourced climate data from the Australian National University Climate surface database (ANUCLIM v6.1³⁴). We obtained national monthly maximum temperature and rainfall measures between 2017 and 2020, and derived these measures for each of our monitoring site locations from a 250m national raster. See Supplementary file S1 for further information on derivation of the climate data. For mistletoe analysis, we calculated annual mean maximum rainfall and temperature measures averaged across the summer months of November to February, when mistletoes are most susceptible to drought impacts³⁵.

Statistical analysis

We used R v3.4.3³⁶ for all statistical analyses. We first checked for spatial autocorrelation in mistletoe abundance and mortality data using correlograms of Moran’s I via package ncf v1.2-5³⁷. To check for cross-correlation between covariates, we used GGally v1.4.0³⁸, but no covariates showed consistent strong positive or negative correlation with others.

To account for interspecific variation in the suitability of tree species as mistletoe hosts³⁹, we ran a centred and scaled Principal Component Analysis (PCA) of the proportional contribution of each tree species to canopy cover across sites using stats v3.6.2 in base R. The first two Principal Component axes, which we included in subsequent mistletoe and bird models (Tree species PC1 & PC2), together explained 9% of the total variation in tree species composition. Because the proportion of total variation explained by the PCA was relatively low, we also conducted a separate analysis of the association between individual tree species and live mistletoe abundance (Supplementary file S2).

To account for spatial autocorrelation in the mistletoe and bird data, we fitted a series of Integrated Nested Laplace Approximation (INLA) Generalised Linear Mixed Models (GLMMs) via package INLA v21.02-23⁴⁰. The INLA models included a Stochastic Partial Differentiation Equation (SPDE) random term that calculates the distances between the spatial location of monitoring sites using Matern covariance^41,42. We selected the best models as those with the lowest Deviance Information Criterion (DIC) value and assessed their goodness of fit based on conditional R2 values43.

To answer question 1 What are the predictors of mistletoe abundance and the drivers of mistletoe mortality? we first used live mistletoe counts as the response in a GLMM with a negative binomial error structure. The model included as fixed effects: Land use, canopy cover, tree species composition, tree health, tree age and distance to standing water, with region included as a random term (Table 1). To assess the association between mistletoebird presence (the key disperser of mistletoe fruits¹⁵) and noisy miner abundance (a key driver of mistletoebird distribution⁴⁴) on mistletoe abundance, we re-ran the model on the subset of sites where we conducted bird surveys (Figure 1) and included mistletoebird presence and noisy miner abundance as fixed effects in the saturated model.

We replaced live mistletoe abundance with dead mistletoe abundance as the response measure to identify the predictors of mistletoe mortality. To the fixed effects included in the live mistletoe model described above, we added live mistletoe abundance, mean maximum monthly summer (Nov-Feb) temperature and summer rainfall for 2017/18, 2018/19 and 2019/20, as well as the annual interaction between these temperature and rainfall measures.

To answer question two What is the relationship between live mistletoe abundance and woodland bird abundance, and how does this relationship change during drought? we calculated four woodland bird abundance response measures, based on overall bird abundance, body size, residency status and feeding guild (Tables 2 and S2). We excluded noisy miners from bird abundance measures due to their impacts on woodland bird abundance⁴⁴, and instead included noisy miner abundance as a fixed effect in the woodland bird models (Table 1). The response measures were counts of birds described in Table 2. Fixed effects included overall blossom score, breeding season, noisy miner abundance, canopy and shrub cover extent, tree species composition, land use, distance to standing water, survey time, and live mistletoe abundance (log + 1 transformed). We included observer and region as random terms. To examine how the relationship between woodland birds and mistletoes changed during drought, we included in the woodland bird models the interaction term live mistletoe abundance × breeding season. We also conducted supplementary analyses to further explore the relative importance of mistletoe and eucalypt blossom during the drought for woodland birds (Supplementary file S3). For all bird models we used a Poisson error structure.

Table 2: Bird response measures used in models to answer question 2: What is the relationship between live mistletoe abundance and woodland bird abundance, and how does this relationship change during drought? Note the species composition of bird response measures are not mutually exclusive. See table S3 and the raw dataset available via the Dryad digital repository.

The files contained in this sumbmission are excel.csv files and can be opened and run with the accompanying R-script