Rocks of different mineralogy show different temperature characteristics: implications for biodiversity on rocky seashores

Boulder selection

The geologically-diverse Fleurieu Peninsula of South Australia is comprised of a variety of rock types. Six of these, in the form of small boulders (n = 6 boulders per rock type, maximum dimension ≤30 cm), were collected from four seashores (Appendix Fig. A1). From Southport (35°10′S, 138°27′E) boulders of either white fossiliferous limestone or orange fossiliferous limestone were collected, while fossiliferous sandstone that was yellowish brown in colour was collected from Seaford (35°11′S, 138°28′E) (Appendix Figs. A1 & A2). The two limestones and the fossiliferous sandstone had coarse surface textures and complex surfaces that were interspersed by cracks and depressions. From Marino Rocks (35°02′S, 138°30′E) boulders of both purple siltstone and grey siltstone were collected, while quartzite that was greyish to yellow-brown in colour, was collected from O’Sullivan Beach (35°07′S, 138°28′E) (Appendix Figs. A1 & A2). The two siltstones had smooth surface textures and featureless surfaces that generally lacked cracks or depressions, while quartzite had coarse, angular surfaces that also lacked cracks and depressions.

Six boulders of each rock type were collected (total N = 36), with boulders specifically selected to span the range of shapes and thicknesses that occurred for each type on each seashore (Appendix Fig. A2). Measurements of thickness were collected for each boulder to be used as a co-variate in statistical analyses. However, as the boulders investigated had only a naturally narrow range of thicknesses (6–14 cm), boulder thickness was never a significant co-variate, and was hence removed from all analyses to improve their statistical power.

Experimental location

Boulders were transplanted into a paddock on a farm at Kangarilla, which was located approximately 20 km inland from the coast (Appendix Fig. A1). This secure setting was selected to be independent of any confounding variables. For example, had experimentation been conducted at one of the seashores where boulders were sourced, variables such as wave splash, tidal movement, sediment or wrack deposition, shading by cliffs, under-boulder substratum or angle of repose may have affected the temperature of some boulders and not others. Setting boulders into a sandy beach was not attempted due to the boulder losses sustained during a translocation experiment in the same region (Liversage, Janetzki & Benkendorff, 2014; Janetzki, Fairweather & Benkendorff, 2018). Issues with sand scour or burial of boulders by sand were also likely on a sandy beach. Moreover, given the large population of the Adelaide region (1.3 million people in 2016), interference with boulders left lying on a beach was considered likely.

Design of the common-garden experiment

A square plot measuring 3 × 3 metres was excavated to a depth of approximately 10 centimetres (Appendix Fig. A2). The ironstone and soil matrix unearthed was replaced by washed yellow beach sand to simulate substrate matrices where experimental boulders were sourced (Appendix Fig. A2). The location of this plot was selected to ensure it had an east–west orientation (i.e., to follow the movement of the sun) free from any physical obstructions (e.g., buildings, trees) that might shade the plot. Boulders were arranged on the sand matrix in four groups that each contained nine boulders (Appendix Fig. A2). Boulders were arranged such that each group could be sampled without accidently disturbing or shading other groups (Appendix Fig. A2). Boulders were randomly assigned to each group and their location was re-randomised on four occasions. The boulder plot was covered with a tarpaulin to shelter boulders from insolation prior to sampling, although small differences in surface temperature among rock types were identified at time zero on several days sampled (Appendix Figs. A3 & A4).

Rock temperatures were measured on 17 days spread over an 18-month period (Table 1). Each day was targeted for its forecast cloud cover and air temperature to investigate generally how rock temperature may be influenced by seasonality and daily weather conditions. Sampling was completed on days where no rainfall was forecast, as wet surfaces were likely to confound our ability to accurately measure rock temperature (Lathlean & Seuront, 2014; Seuront, Ng & Lathlean, 2018). Approximately 20 min before the commencement of sampling, the tarpaulin was removed from the boulder plot and each boulder was submersed (<1 min) in a tub filled with seawater. Submersion wetted the boulder surfaces to simulate conditions on the seashore, where boulder surfaces are wet when first emersed by the receding tide. Each boulder was returned to its specific spot in the plot where it was allowed to drain and dry (this took no longer than 5–10 min as water rarely permeated boulder surfaces). There was no evidence of differential patterns of drying over this 5–10 min timeframe among the six rock types investigated.

The surface temperature of each boulder was measured at one-hour intervals, commencing at 0900 h and concluding at 1400 h daily. This five-hour time course was selected to simulate the average length of time boulders on local mid-lower seashores are emersed during a single low tide (NJ unpublished data). The 0900–1400 start and finish times were selected to simulate the timing of summer daytime low tides for seashores on the Fleurieu Peninsula. Sunrise occurred between 0555 h during the height of summer and 0724 h during the depths of winter.

Surface temperature was measured with a Fluke Ti20 thermal imaging camera (Fluke Corporation, Everett). The thermal resolution of this camera was ≤0.2 °C at 30 °C, with accuracy to 2% or 2 °C, whichever was greater. Default camera settings were employed, including emissivity, which was set at 0.95. This default emissivity was applied, even though emissivity may have differed within and among the rock types sampled, as previous thermography studies in the long-infrared range (9–14 µm) have shown that the emissivity of dry rock generally ranges between 0.95 and 1 (Rivard, Thomas & Giroux, 1995; Danov, Vitchko & Stoyanov, 2007; Cox & Smith, 2011; Lathlean, Ayre & Minchinton, 2012). Surface temperature was captured in situ, with archived thermal images processed later. Upper and lower boulder surfaces were imaged separately. In this study, lower surfaces are defined as the underside of the boulder that was in contact with the substratum and thus sheltered from insolation. Upper surfaces are defined as all remaining surfaces that were not in contact with the substratum and were potentially exposed to insolation.

Thermal images were recorded for all upper surfaces first without touching boulders. For lower surfaces, each boulder was briefly flipped upside down, and a thermal image recorded, before the boulder was returned to its original position. Overall, a total 7,344 thermal images were collected. Air temperature and cloud cover were also recorded at one-hour intervals when taking images. Air temperature was measured in the shade to the nearest degree Celsius with a glass thermometer. Cloud cover (i.e., sky condition) was estimated by how many eighths of sky were covered by cloud, which ranged from zero oktas (sunny, no clouds) to eight oktas (sky completely cloudy, no sunshine) (Li & Lam, 2001). Each day sampled was assigned to one of two weather condition categories based on their cloud cover. Days where the cloud cover never exceeded 3 oktas were assigned the ‘sunny’ category, while days where the cloud cover exceeded 3 oktas during sampling were assigned the ‘cloudy’ category (Table 1).

Archived thermal images were subsequently processed using the InsideIR version 4.0 software (Fluke Corporation, see Appendix for more information). The maximum and minimum temperature of substrata were determined for each replicate surface, and a temperature range (maximum–minimum) for each surface was calculated. The orientation, relative to the sun, for the maximum temperature, was categorised as either occurring on the boulder side nearest the sun, or on any other boulder side. Transects were drawn on images of boulder surfaces from the centre of the boulder side nearest the sun to the centre of the side opposite to quantify millimetre-to-centimetre scale patterns of temperature difference. Analysis of the temperature patterns along transects was undertaken on three sunny and three cloudy days, spanning the range of maximum daily air temperatures sampled. Images for the zero-hour and four-hour exposure times were investigated to look at changes across the day.

XRF analyses

The mineralogy of each rock type was determined through X-ray fluorescence (XRF) dispersion analysis, with separate tests completed for major mineral and trace elemental composition for three samples of each rock type. XRF analysis of each rock sample tested for 11 major minerals and 40 trace elements, with concentrations returned as % and parts per thousand, respectively. For major minerals, approximately one gram of each oven-dried sample (at 105 °C) was accurately weighed with four grams of 12–22 lithium borate flux (Norrish & Hutton, 1969). The mixtures were heated to 1,050 °C in a platinum/gold crucible for 20 min to completely dissolve the sample, and then poured into a 32 mm platinum/gold mould heated to a similar temperature (Norrish & Hutton, 1969). The melt was cooled rapidly over a compressed air stream and the resulting glass disks were analysed on a PANalyticalAxios Advanced wavelength dispersive XRF system using the CSIRO in-house silicates calibration program. For trace elements, approximately four grams of each oven-dried sample (at 105 °C) was accurately weighed with one gram of Licowax binder and mixed well (Norrish & Hutton, 1969). The mixtures were pressed in a 32 mm die at 12 tons pressure and the resulting pellets were analysed on a PANalyticalAxios Advanced wavelength dispersive XRF system using the CSIRO in-house powders program (Norrish & Hutton, 1969).

Statistical analyses and data presentation

Frequencies of occurrence (%) for three patterns of temperature difference were tallied for the upper and lower surfaces of boulders for all six rock types after four hours exposure to insolation. Only data for three cloudy versus sunny days are presented, because each day is not strictly independent of each other.

Line charts plotting temperature dependent variables (maxima, range) versus exposure time were used to rank rock types in descending order from 6 (largest) to 1 (smallest) for their mean temperature range and maxima, for each surface on each day after four hours. The sum of ranks allocated to each rock type was then used to assign an overall rank to each rock type from 6 (largest) to 1 (smallest) for their temperature range and maxima, with the highest ranked rock types having the highest sum of ranks. Upper and lower surfaces were ranked separately. To quantify changes over time exposed, dependent variables for the upper and lower surface of each replicate boulder after four hours were subtracted from the same dependent variables for the same replicate surface at zero hours. This gave values for the change in maxima and temperature range over four hours, for the upper and lower surfaces of each boulder, on each day sampled. Comparing the magnitude of change over four hours was used in place of formal statistical tests, as measurements were made continuously for the same boulders and thus were not independent. To establish whether dependent variables differed between surfaces, upper-surface dependent variables were subtracted from lower-surface dependent variables, for each replicate boulder, for each exposure time on each day sampled. The resulting difference data were plotted as line charts to visually investigate surface differences over the exposure period.

Analyses were completed using PRIMER v7/PERMANOVA+ (PRIMER-e, Plymouth, UK), with significance set at α = 0.05. To test for mineralogical differences between rock types, separate multivariate analyses were completed for major mineral composition and trace element composition, with separate univariate analyses completed for the total content of each major mineral. Untransformed major-mineral data (measured as % composition) were used, while fourth-root-transformed (necessary to down-weight several dominant elements) trace-element data (measured as parts per million) were used. Euclidean-distance resemblance matrices were prepared, and PERMutational Analyses Of VAriance (PERMANOVAs) were run to test for mineralogical differences between rock types. For multivariate data, constrained ordination Canonical Analysis of Principal coordinate (CAP) plots (Anderson, Gorley & Clarke, 2008) were used to visualise rock-related mineralogical differences. A leave-one-out procedure was used to test the allocation success of the discriminant function for rock groupings in CAP, with permutation tests used to test the significance of the trace test statistic and first canonical eigenvalue. Vector overlays of Spearman Rank correlations (for Rho values > 0.8) were used to identify the major minerals and trace elements that best characterised the mineralogy of each rock type.

Patterns of temperature difference on upper or lower surfaces of different rock types

Boulder upper and lower surfaces, for all six rock types, had generally a heterogeneous surface temperature differing in maxima and minima after four hours exposure to insolation on all days sampled (Fig. 1). Moving in any direction across boulder surfaces (a representative thermal image showing each rock type’s surface temperature is provided in Fig. 2), temperature consisted of warmer and cooler areas with patch sizes <10 cm. When transects were drawn on thermal images, three qualitative patterns of temperature difference were identified. The first pattern was a temperature mosaic, which consisted of heterogeneous temperatures across the entire boulder surface (Fig. 3A). The temperature difference between the warmest and coolest mosaic areas was ≥5 °C. The second pattern was a temperature gradient, where temperature appeared to gradually decrease from the side nearest the sun to the side opposite (Fig. 3B). The temperature difference between the warmest and coolest gradient areas was ≥5 °C. The third pattern was limited temperature heterogeneity, which consisted of only small temperature differences <5 °C between the warmest and coolest areas (Fig. 3C).

The frequency of occurrence for these three patterns of temperature difference was strongly influenced by daily weather conditions (Table 2). On cloudy days, almost all (>94%) boulder surfaces for all six rock types were categorised as having limited temperature heterogeneity (Table 2). Temperature mosaics and gradients were seldom, if ever, observed on cloudy days (Table 2). On the two cooler sunny days, almost all (>94%) boulder surfaces for all six rock types were categorised as having either temperature gradients or mosaics (Table 2). Temperature gradients were more common than temperature mosaics, in a ratio of 3:1, on these sunny days (Table 2). On the hottest sunny day, all three patterns of temperature difference were observed (Table 2); however, boulder surfaces became generally hot, with temperature differences between the warmest and coolest areas mostly <5 °C (Table 2, Fig. 1). As a result, limited temperature heterogeneity was the most common (≥78%) temperature pattern identified on the hottest sunny day (Table 2), with gradients and mosaics observed at much lower frequencies than on the other sunny days.

On sunny days patterns of temperature difference differed among rock types. The coarse, angular surfaces of quartzite had a mosaic of fine millimetre to centimetre scale patches of heterogeneous temperature (Fig. 2C, Appendix Table A1). The five remaining rock types all generally had temperature gradients, although the spatial arrangement of these gradients differed. The generally smooth and featureless surfaces of siltstone had simple gradients of warmer through cooler areas (Figs. 2A–2B, Appendix Table A1). In contrast, the two limestones and fossiliferous sandstone had complex surfaces intersected by shallow depressions and pits (<1 cm depth). Consequently, their temperature gradients were interrupted intermittently by these depressions and pits, which could be either warmer or cooler (by up to 2−3 °C) than the flatter surfaces immediately around them (Figs. 2D–2F). All three patterns of temperature difference were related to boulder orientation, with the hottest temperatures generally recorded for the side of boulders nearest the sun (>92%, Table 2). Each rock type generally had the same pattern of temperature difference on its upper and lower surfaces.

Maximum temperature differs between rock types and surfaces

After four hours exposure the mean maximum temperature was hotter than the air temperature for all six rock types on both surfaces, especially on sunny days (Figs. 4A–4B). The maximum temperature of upper and lower surfaces generally increased with time, with maxima often peaking around four hours and plateauing thereafter (Fig. 5A). The largest increases in maxima were generally recorded during the first two hours exposure to insolation, with smaller increases (and sometimes decreases) recorded thereafter (Appendix Figs. A3 & A4). After four hours, the hottest maximum recorded was 57.8 °C (sunny day, air temperature = 39 °C) for the upper surface of grey siltstone while the coolest maximum was 14.4 °C (sunny day, air temperature = 12 °C) for the lower surface of quartzite. Over four hours, increases in mean maximum surface temperature >20 °C were recorded for some rock types on several days, with the greatest increases recorded for upper surfaces on sunny days (Appendix Fig. A5).

When rock types were ranked from hottest to coolest mean maximum temperature after four hours, a consistent rank order across replicate days was identified, irrespective of weather conditions (Table 3). For both surfaces, the two siltstones consistently recorded the hottest maxima, followed by fossiliferous sandstone and orange limestone in descending rank order (Table 3, Fig. 5A). On upper surfaces, white limestone consistently recorded the second coolest maxima and quartzite the coolest, while on lower surfaces both rock types were equally ranked in terms of the coolest maxima (Table 3). After four hours exposure for the 17 replicate days, the smallest difference for mean maxima across the six rock types was 2.5 °C, while the largest difference was 10.2 °C. A similar rank order was identified for the change in maxima over four hours for both surfaces (Table 3, Appendix Fig. A5). The two siltstones generally had the largest increase in maximum temperature, while white limestone and quartzite had the smallest (Table 3, Appendix Fig. A5).

Generally, maxima behaved similarly on upper and lower surfaces (Fig. 4C). At the commencement of sampling on most days, small negative differences were detected between upper and lower surface maxima for all rock types, with lower surfaces having mean maxima that were slightly hotter (<2 °C) than upper surfaces (Appendix Fig. A6). Thereafter, small positive differences were detected between upper and lower surfaces for most rock types, with upper surfaces having hotter mean maxima than lower surfaces, although these differences never exceeded 5 °C (Fig. 5B, Appendix Fig. A6). The only notable exception to this trend was quartzite, which generally had small negative differences throughout, with lower surfaces sometimes having hotter mean maxima than upper surfaces (Fig. 5B). The difference in maxima between upper and lower surfaces was always smallest on cloudy days, with larger differences detected on sunny days (Appendix Fig. A6). Minima behaved similarly to maxima over four hours exposure to insolation with the same trends identified for rock type, surface and time exposed (refer to minima sub-section in Appendix for more details).

Temperature range does not consistently differ between rock types or surfaces

Generally, temperature range behaved similarly on upper and lower surfaces (Appendix Figs. A7 & A8). Temperature range was influenced by weather conditions (Fig. 1). A larger temperature range (5–15 °C) that was more variable between rock types and exposure times was recorded on sunny days for both surfaces (Fig. 5C, Appendix Figs. A7 & A8). In contrast, a smaller temperature range (generally <5 °C) that was less variable between rock types and exposure times was recorded on each cloudy day for both surfaces (Appendix Figs. A7 & A8). After four hours, the largest temperature range recorded was 16.3 °C (sunny day, air temperature = 15 °C) on the upper surface of grey siltstone while the smallest temperature range was 1.6 °C (cloudy day, air temperature = 38 °C) on the upper surface of purple siltstone.

When rock types were ranked from largest to smallest for the mean temperature range after four hours, there was little evidence of a consistent ranking across replicate days for both upper and lower surfaces (Table 4, Appendix Figs. A7 & A8). Rankings were similarly variable after both shorter and longer exposure times, with rank order often changing from one exposure time to the next (Appendix Figs. A7 & A8). No consistent ranking was identified either for the change in mean temperature range over four hours for either surface (Table 4, Appendix Fig. A9).

The temperature range difference between upper and lower surfaces (i.e., upper range–lower range) was always smallest on cloudy days, with larger differences detected on sunny days (Appendix Fig. A10). On cooler days with a maximum daily air temperature <30 °C, regardless of the day condition, small positive differences were generally detected, with the upper surfaces of most boulders having a larger temperature range than lower surfaces (Appendix Fig. A10). In contrast, on hotter days with a maximum daily air temperature ≥30 °C, small positive differences were detected only for the two siltstones. Small negative differences were often measured for the two limestones and quartzite, with lower surfaces having a larger temperature range than upper surfaces (Appendix Fig. A10). Overall, the two siltstones generally had the largest range difference between upper and lower surfaces, while white limestone and quartzite often had the smallest (Appendix Fig. A10).

Rock-related differences in maximum temperature are correlated with their mineral composition

Silicon dioxide (SiO₂) and calcium oxide (CaO) were the dominant major minerals detected, with orange limestone having a CaO-dominated mineralogy and all other rock types a SiO₂-dominated mineralogy (Appendix Table A3). Major mineral composition significantly differed among rock types (PERMANOVA permuted p-value =0.0001). For rock type differences, a CAP constrained-ordination plot used five axes to discriminate major-mineral differences, with the first two axes accounting for 99.54% (prop. G) of the total mineralogical variability (Fig. 6A). All samples were correctly classified using a leave-one-out procedure, and permutation tests for both the trace test statistic (p = 0.0001) and first canonical eigenvalue (p = 0.0001) were highly significant. The vector overlay of Spearman rank correlations (for rho values > 0.8) for major minerals associated with rock differences showed that each rock type had a specific major-mineral composition (Fig. 6A). Grey siltstone was characterised by higher aluminium oxide and potassium oxide contents, quartzite by the highest SiO₂ content, and orange limestone and fossiliferous sandstone by higher CaO contents (Fig. 6A). Rock type differences in the content of specific major minerals were also detected for 10 from 11 major minerals (largest significant permuted PERMANOVA p-value =0.0330 for magnesium oxide), with only sulfur trioxide not differing between rock types (PERMANOVA permuted p-value =0.1244).

Trace-element composition significantly differed among rock types (PERMANOVA permuted p-value =0.0001). For rock type differences, a CAP constrained-ordination plot used two axes to discriminate trace-element differences, with these two axes accounting for 79.8% (prop. G) of the total mineralogical variability (Fig. 6B). Some 88.9% of samples were correctly classified using a leave-one-out procedure, and permutation tests for both the trace test statistic (p = 0.0001) and first canonical eigenvalue (p = 0.0001) were highly significant. The vector overlay of Spearman rank correlations (for rho values > 0.8) for trace elements associated with rock type differences suggested that each rock type had a specific trace-element composition (Fig. 6B). The two siltstones were characterised by a higher trace-metal content (manganese and zirconium especially), quartzite by generally low trace-element quantities (although it had the highest ytterbium content), and the two limestones and fossiliferous sandstone by a higher chlorine content (Fig. 6B, Appendix Table A4).

The vector overlay of Spearman rank correlations (for rho values > 0.8) in CAP plots (Fig. 6) showed that the two siltstones, which consistently had the hottest maxima, had mineralogies that were characterised by a higher content of metallic oxides and trace metals versus all other rock types (Fig. 6, Appendix Tables A3 & A4). White limestone and quartzite, which consistently had the coolest maxima, were characterised by the highest content of SiO₂ and the lowest content of most metallic oxides and trace metals versus all other rock types (Fig. 6, Appendix Tables A3 & A4). Meanwhile, orange limestone and fossiliferous sandstone, which had intermediate maximum temperatures, were characterised by higher contents of CaO and chlorine, and metallic oxide and trace metal quantities that were generally lower than the two siltstones but greater than white limestone and quartzite (Fig. 6, Appendix Tables A3 & A4).