Refugia can facilitate the persistence of species through long-term environmental change, but it is not clear if Pleistocene refugia will remain functional under anthropogenic climate change. Dieback of species within refugia therefore raises concerns about their long-term persistence. Using repeat field surveys, we investigate dieback patterns of an isolated population of Eucalyptus macrorhyncha during two droughts and discuss prospects for its continued persistence in a Pleistocene refugium. We first confirm that the Clare Valley in South Australia has constituted a long-term refugium for the species, with the population being genetically highly distinct from other conspecific populations. However, the population lost >40% of individuals and biomass through the two droughts, with mortality being just below 20% after the Millennium Drought (2000–2009) and almost 25% after the Big Dry (2017-2019). The best predictors of mortality differed after each drought. While the north-facing aspect of a sampling location was a significant positive predictor after both droughts, biomass density and slope were significant negative predictors only after the Millennium Drought, and distance to the north-west corner of the park, which intercepts hot, dry winds, was significant after the Big Dry only. This suggests that more marginal sites with low biomass and located on exposed, flat plateaus were more vulnerable initially, but that heat-stress was an important driver of dieback during the Big Dry. Therefore, the causative drivers of dieback may change during population decline. Regeneration occurred predominantly on southern and eastern aspects, which would receive the least solar radiation. Occurrence in a refugium did not protect this population from dieback. However, gullies with lower solar radiation are continuing to support relatively healthy, regenerating stands of red stringybark, providing hope for persistence in small pockets. Monitoring and managing these pockets during future droughts will be essential to ensure the persistence of this isolated and genetically unique population.

The data contains three datasets derived from analysing data from multiple surveys of a red stringybark population (Eucalyptus macrorhyncha) in Spring Gully Conservation Park (SGCP), Clare Valley, Australia. These are the Tree Health Index (THI), Biomass and Drivers datasets, which are used in the analyses of the associated paper. Below I explain how each dataset was obtained.

The South Australian Department of Environment and Water (DEW) initiated a tree health monitoring program in 2009, during which four North-South oriented transects were established in SGCP. Each transect (between 1.2 and 1.8 km long) had sampling sites every 50 m. At each sampling site, the four closest canopy trees within a 10 m radius were marked with a permanent aluminum tag, their location recorded with a handheld GPS (brand and model unknown), and various measurements relating to their health status taken (see below). In total, 471 trees were surveyed, 30 of which were South Australian blue gums (Eucalyptus leucoxylon F.Muell.) and the remainder were red stringbark. Transects were surveyed in January and February 2009, March 2010, November 2011, August 2012, November 2013, and September 2014. Parameters recorded included tree status (dead/ alive; trees with dead stems but with living basal sprouts were scored as alive), crown extent (percentage area of assessable crown with live leaves), and crown density (percentage of skylight blocked by the leafy crown). Percentage values were recorded as eight categories: 0 (0%), 1 (1–10%), 2 (11–20%), 3 (21–40%), 4 (41–60%), 5 (61–80%), 6 (81–90%), and 7 (91–100%). The assessable crown was defined as consisting of all living and dead branches of the crown. In addition, epicormic growth and extent of reproductive activity (presence of flowers and/or fruits) were classified into four categories: 0 (absent, not visible), 1 (scarce, present but not readily visible), 2 (common, clearly visible throughout the assessable crown), 3 (abundant, dominates the appearance of the assessable crown). We calculated a summative index consisting of canopy extent, canopy density, and epicormic growth to indicate tree health – hereafter referred to as the tree health index (THI). Because crown extent and density are considered the most important indicators of tree health, we retained them at their larger scale (0–7, compared to 0–3 for epicormic growth), giving a maximum value of 17 for the THI. Trees that appeared dead at some surveys but that later resprouted (i.e., epicormic growth or basal sprouts), were retrospectively awarded a THI score of 1 (instead of zero). To get an indication of the health status of the red stringybark population, the proportion of dead trees and the average THI of all 441 stringybark trees surveyed repeatedly since 2009 were determined and this data is available in the THI dataset.

In September and December 2021, we revisited all trees that had been surveyed and tagged previously. Relocation of trees was achieved with high confidence because of the availability of GPS locations for each tree and because tags remained on, or had fallen directly beneath, at least two of the four trees at each site. Six sites consisting entirely of blue gum were not resurveyed (sites T1S02, T1S03, T1S04, T1S05, T1S30, T3S09). Methods for the resurvey focused on replicating the methods used in the earlier surveys (to facilitate comparisons) and on collecting additional information to provide area-based estimates of dieback. To achieve area-based estimates, we determined a center point for each site so that each of the four trees at a site was in a different quarter (delineated using the four cardinal directions). For sites with one or more blue gum trees among the four surveyed trees, blue gums were replaced by the nearest stringybark in the relevant quarter. In one instance, no nearest stringybark neighbour was present within 10 m and this site was excluded from analyses including biomass density. Additional trees were added to four sites that had less than four surveyed trees. This resulted in a total of 112 sites with 448 trees of red stringybark. This allowed estimating the tree density per hectare at each site by measuring, averaging, and squaring the distance of each tree to the center point. The inverse of this average distance was then multiplied by the value of the desired area (in this case 1 ha) to obtain an estimate of tree density, following the point-centered quarter method.

To estimate biomass, we recorded diameter at breast height (DBH) and tree height for each stem of a tree (trees regularly had multiple stems), living or dead. A tree with dead stems was considered alive if there was any epicormic or basal growth present. A stem was considered alive if epicormic growth was present above 1.3m in height. The height (Ht) was estimated to the nearest meter using a 1.5 m range pole that was held up vertically overhead to provide a reference of approximately 3.5 m height. The DBH was measured using a diameter tape 1.3 m above the ground. A wood density (WD) of 795 (± 19) kg.m^-3 was assumed for all trees. We used these values to calculated the above-ground biomass (AGB) as: AGB = 0.0673 × (WD × DBH² × Ht)^0.976. AGB was determined for every stem and then aggregated per tree, meaning a single individual could be composed of both living and dead biomass. We multiplied the estimate of number of individuals per hectare by the mean AGB per tree to obtain area-based estimates of biomass, i.e., biomass density. These calculations were done for each site (to obtain estimates of biomass density per site) and for all 112 sites combined (to obtain a parkwide estimate) and this data is available in the biomass dataset.

As an estimate of regeneration, the occurrence of seedlings (< 1 m tall, woody growth lacking) and saplings (< 1 m tall, woody growth present) within a 3 m radius of the center point was recorded. Seedling and sampling numbers for each site were combined to provide an indicator of recruitment. In addition, aspect (in degrees rounded to 10° intervals and determined with a compass) and slope (in degrees using a clinometer) were recorded for each site. We calculated ‘northness’ and ‘eastness’ as the cosine and sine of the aspect (in radians), respectively. Where trees within a site were located on different slopes in a valley, the aspect and slope were recorded for each slope and then averaged. Distance to the north-west corner of the park (the area most affected by hot, dry summer winds) was calculated as the planar distance between this location and the sampling locations using the “Near (Analysis)” geoprocessing tool in ArcGIS Pro. We calculated the proportion of dead trees per site in 2011 (Mortality 2011) and 2021 (Mortality 2021) and regeneration as indicators of dieback and persistence (response variables). These variables for the 112 sites of the 2021 survey, but including only trees that were surveyed in 2011 as well (a total of 441 trees) are presented in the Drivers dataset.