Local microclimates can both amplify and mitigate extreme temperatures associated with climate change

Local microclimates can both amplify and mitigate extreme temperatures associated with climate change Esme Ashe-Jepson1,2 , Edgar C. Turner2,3 , Andrew J. Bladon2,4 1 Chair of Global Change Ecology, Biocenter, University of Würzburg, Campus Hubland Nord, 97074, Würzburg, Germany 2 Department of Zoology, University of Cambridge, Downing Place, Cambridge, CB2 3EJ, UK 3 University Museum of Zoology, University of Cambridge, Downing Place, Cambridge, CB2 3EJ, UK 4 Ecology and Evolutionary Biology Division, School of Biological Sciences, University of Reading, Whiteknights, Reading, RG6 6EX, UK Corresponding author: Esme Ashe-Jepson ([email protected]) Editor Dan Gavin Received 11 July 2025♦ Accepted 10 November 2025♦ Published 19 December 2025 Frontiers of Biogeography 18, 2025, e164843|DOI 10.21425/fob.18.164843 Copyright Esme Ashe-Jepson et al. This is an open access article distributed under the terms of the Creative Commons Attribution License (CC BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. RESEARCH ARTICLE FRONTIERS OF BIOGEOGRAPHY The scientific journal of The International Biogeography Society Abstract Climate change is a threat to global biodiversity, with changes to mean temperatures and increasing frequency and intensity of extreme weather events. Heatwaves in particular pose a threat to species’ persistence, as temperatures may rise above physiological tolerance. However, individuals rarely experience temperatures measured at the macroclimatic scale: topographic or vegetation differences result in microclimates that provide cool refugia (local temperatures below ambient) or even result in heat traps (local temperatures above ambient) during heatwaves. However, little is known about the stability of microclimates through a period of regional warming. In this study, we recorded microclimate temperatures across different microhabitats within a calcareous grassland nature reserve in Bedfordshire, UK, in 2018, 2019 and 2022. During this time, six heatwave events occurred, including the highest air temperatures ever recorded in the UK. We found that the ability of microhabitats to offset air temperatures varied with topographic aspect, slope, amount of bare ground, shelter, vegetation height, and vegetation type, with encroaching scrub and north-facing slopes showing the strongest abilities to maintain relatively stable microclimate temperatures with increasing air temperatures, in contrast to short vegetation on south-facing slopes which became heat traps. However, no combinations of environmental structures consistently maintained cool refugia during heatwaves. Microclimate temperatures were amplified close to the ground, whereas at 50 cm height temperatures were more stable and similar to the macroclimate temperature, therefore surface-dwelling species, such as many insects, may be particularly vulnerable to extreme heat. We identified a breakdown in the ability of microhabitats to maintain cool refugia above 7 °C, implying cool refugia become increasing rare and unpredictable with increasing temperatures. Our results indicate that many microhabitats will amplify the effects of climate change rather than mitigate them. Highlights • The innate unpredictability and scarcity of extreme temperature events makes them an evolutionary challenge; instead, microclimatic refugia are often suggested as a way for species to cope under climate change. • By comparing microclimate temperatures to macroclimate temperatures (from a nearby weather station) in central England, we highlight how variable local temperatures can be at fine scales relevant to small surface-dwelling organisms, with microclimate temperatures differing by as much as 20 °C. • Microclimate performance changed with increasing temperature, with more heat traps (microclimate temperatures above ambient) and fewer cool refugia (microclimate temperatures below ambient). • No environmental variables tested reliably maintained cool refugia at high temperatures. • Ground-level microclimate temperatures were amplified compared to temperatures at 50 cm height, implying that surface-dwelling organisms in grasslands, including many insects, may be particularly vulnerable compared to species that fly, climb, or live in tall vegetation. Frontiers of Biogeography 18, 2025, e164843 Esme Ashe-Jepson et al. 2 Keywords Central England, climate change, cool refugia, data logger, extreme weather, heat trap, heatwave, landscape structure, microhabitat Introduction Climate change is a growing threat to global biodiversity, through changes in mean temperatures and increasing frequency and intensity of extreme weather events (IPCC 2022). Temperature directly impacts individuals by altering physiological processes, many of which are relevant to fitness, such as metabolism, oxygen demand, immune functioning, and hormone production (Karl et al. 2011; Verberk et al. 2016; González-Tokman et al. 2020). By altering internal processes, these upscale to individual and population-level responses, such as changes to species’ behaviour (Hill et al. 2021), phenology (Inouye 2022), distribution (Pinsky et al. 2020; Soifer et al. 2025) and survival (Román-Palacios and Wiens 2020). Furthermore, temperature can indirectly impact interspecific interactions by causing a breakdown of close ecological relationships between species (Thackeray et al. 2010; Patterson et al. 2020), or through species’ range shifts creating novel communities and interactions (Lurgi et al. 2012; Alexander et al. 2015). Therefore, understanding how species experience temperature, particularly extremes, is of critical conservation importance. Most species, particularly small organisms such as insects, do not experience temperature at the macroclimate scale. Instead, small variations in topography and vegetation (microhabitats) can alter local climatic conditions across various scales in the form of microclimates, which can differ significantly from macroclimatic means (Bennie et al. 2008; Song et al. 2013). For example, temperatures can vary by 20 °C between north and south-facing mountainsides (Scherrer and Körner 2010). Landscape structure can therefore contribute to temperature differences within small areas, which can be similar to those found over large elevational or latitudinal gradients, and on one hand can act as climatic refugia for species to support persistence in otherwise adverse landscapes (cool refugia) (Suggitt et al. 2018). Alternatively, landscape structure can trap hot air or amplify temperatures experienced locally (heat trap), which would increase the prevalence of detrimental temperatures within landscapes. Cool microclimates may be particularly important for species with limited dispersal ability or physiological tolerance of high temperatures, or during extreme weather events such as heatwaves, where the direct temperature effects on many species will be most severe. Heatwaves, defined by the Met Office in the UK as extended periods of hot weather relative to the expected conditions of the area at that time of year, are becoming increasingly common. In the UK, 2022 was the hottest year on record, with the highest air temperatures since records began, reaching 40 °C for the first time (Yule et al. 2023). There is growing evidence that extreme weather events such as heatwaves, rather than gradual changes in mean temperatures, drive species’ responses to climate change (Bauerfeind and Fischer 2014; Ma et al. 2021). The evolutionary consequences of extreme temperature events are currently poorly understood (Grant et al. 2017), but can include strong selection in cases where specific phenotypes are affected (Van De Pol et al. 2010). This can be through truncation selection, whereby only individuals with a particular trait, such as thermal tolerance above or below a given temperature threshold survive or are fertile and contribute to the next generation. Such strong selection can increase the risk of population extinction (Vincenzi et al. 2012). Alternatively, should all individuals be equally affected by an extreme temperature event, or the impacts affect phenotypes randomly, there is little selective pressure to adapt. Adaptation to cope with temperature extremes can be inhibited by the innate unpredictability and scarcity of extreme events, particularly for species with short generation times. Therefore, selection imposed by these infrequent events can be eroded back to a phenotype that copes better with averages than extremes in subsequent generations (Bryant and Jones 1995). As such, microclimatic refugia may play a critical role in allowing species to cope with extreme conditions irrespective of adaptation, with long-term ecological and potentially evolutionary implications. There is a growing body of knowledge as to how microhabitat structure alters the temperatures that species experience at very local scales, particularly relating to ground-shading structures such as topography or vegetation cover. For example, taller vegetation resulted in lower soil temperatures (Song et al. 2013), and micro-topographically complexity altered microclimate compositions in alpine systems (Opedal et al. 2015). Similarly, there is evidence that microclimatic conditions vary not only over the horizontal plane, but also the vertical plane, with large differences in thermal conditions with only minor changes in vertical height or distance from the ground, thereby altering the conditions experienced by species occupying different horizontal and vertical parts of the landscape (Kearney et al. 2021; Klinges and Scheffers 2021; Zellweger et al. 2024). Much recent research has demonstrated the importance of considering these variations in local climate when predicting species’ responses to climate change (Scheffers et al. 2014; Baudier et al. 2015; Suggitt et al. 2018), particularly in relation to species distribution modelling (Lembrechts et al. 2019; Stark and Fridley 2022). However, although topography and vegetation structure are known to influence microclimate temperatures (Bennie et al. 2008), it is unclear whether microclimatic offsetting (the difference between microclimate temperature and ambient air temperature, following the definition of De Frenne et al. (2021)) Frontiers of Biogeography 18, 2025, e164843 Microclimates often amplify extreme temperatures 3 changes during extreme weather events. Furthermore, improving understanding of which environmental structures provide the most reliable cool refugia (microclimate temperatures below ambient, which allow organisms relief or escape from high temperatures), or result in heat traps (microclimate temperatures above ambient, which put individuals that live in these areas at greater risk from high temperatures) during extreme weather events, can inform management of landscapes to promote climate resilience. With a greater understanding of which environmental characteristics influence microclimate composition, land-managers could develop landscapes that maximise species persistence under a changing climate, and focus limited resources on the management of characteristics that have the strongest influence. This study takes place within a managed highly heterogenous calcareous grassland nature reserve in Bedfordshire, UK. As a calcareous grassland, this ecosystem contains high biodiversity, but is also innately exposed, with little shelter and dry soil (WallisDeVries et al. 2002). As a result, calcareous grasslands are at particular risk under future climate change, with the flora and fauna that occupy them particularly exposed to extreme weather events such as heatwaves. Using a network of 131 temperature data loggers, this study addresses the following questions: 1. Does the offsetting effect of microhabitats change with environmental structure: topographic aspect, slope steepness, amount of bare ground, degree of shelter, vegetation type, or vegetation height, and which structures are the most important? 2. Does the offsetting effect of microhabitats change with increasing distance from the ground? 3. Does the offsetting effect of microhabitats change with increasing air temperature, and is there a particular temperature at which this relationship changes? 4. Which combinations of environmental structures that land-managers could alter result in the strongest microhabitat offsetting effect with increasing air temperatures? We expect that microclimate temperatures will depend on microhabitat structure, but that particular characteristics will have stronger effects than others. We expect that topography (aspect, slope) and vegetation (type and height) will have strong effects on microclimate temperatures, as these are known to result in local temperature differences (Bennie et al. 2008; Song et al. 2013; Opedal et al. 2015). We expect that shelter may have a strong impact on the maintenance of cool refugia and heat traps as a result of reducing air mixing and trapping hot or cold air. We expect that percentage of bare ground will have an impact on heat traps due to the chalk in the soil reflecting solar radiation near the ground, resulting in amplified temperatures. However, the high albedo of the chalk should also reduce the amount of solar radiation absorbed, and so the impact of bare ground may be weaker than other environmental characteristics. Our aim is to identify key factors for land-managers and produce actionable advice to support nature within climate-resilient landscapes. Methods Study site Data collection took place in Totternhoe Quarry, Bedfordshire, UK (51°53'30.75"N, 0°34'09.37"W) (Fig. 1), a highly heterogeneous nature reserve managed by the Wildlife Trust for Bedfordshire, Cambridgeshire and Northamptonshire (110–130 m above sea level). Totternhoe Quarry is an 8.9-hectare Site of Special Scientific Interest (SSSI), composed of a mixture of bare chalk, mixed length grasses, and scrub, with highly variable topography as a result of medieval quarrying. There is no natural surface water on site. The large variation in topography and vegetation structure within a small area generates a variety of microclimates which can be studied simultaneously while controlling for variations in local weather conditions. The reserve was originally mapped in 2009 with various updates in 2016–2017, as described in Hayes et al. (2018), and further updated in 2018. A backpack GPS (Trimble GeoExplorer 2008 Series Geographica Positioning System Handheld Recorder, running Fastmap Mobile 7, and Geobeacon Receiver) was used to map out blocks of continuous habitat by walking habitat perimeters, which were assigned slope, aspect, and vegetation type characteristics. In 2020, environmental characteristics (topographic aspect and vegetation type) across the whole site were updated using images from drone surveys and ground-truthed with on-the-ground surveys. High resolution digital elevation models were used to calculate aspect. There have been no general changes to the distribution of vegetation types since 2020. Microclimate data Microclimate data were collected continuously from May 2018 to August 2019 (with a gap from April 1st to 25th where data loggers were inactive, as they had to be collected and reset), and again from May to August 2022. In 2018, sampling locations were selected using R (R Core Team, 2018), by identifying the three largest continuous areas of each combination of four vegetation types (exposed chalk, short grass, long grass, encroaching scrub – see below for definitions) and five aspects (north, east, south, west, flat) based on fine-scale habitat maps of the reserve from Hayes et al. (2018). We chose to stratify our sampling in this way to capture the range of microclimatic conditions represented across the reserve in relation to both aspect and vegetation type, factors which are known to influence microclimatic conditions. Not all combinations of vegetation type and aspect had three replicates large enough to appear on the maps (at least 5 m2), therefore a total of 51 locations were used. The geometric Frontiers of Biogeography 18, 2025, e164843 Esme Ashe-Jepson et al. 4 centroid of each polygon was calculated using R, and two data loggers (Thermochron iButton (DS1921G-F5), Maxim Integrated Products Inc) (accuracy ± 1 °C) were placed there, at 5 cm and 50 cm above the soil. These heights were chosen to represent near-ground conditions that organisms on low vegetation might experience (such as caterpillars), and conditions flying insects (such as adult butterflies) or insects climbing higher in the vegetation might experience. Data loggers recorded temperature hourly (on the hour) from April to October, and every two hours (to extend the time they could be left) from October to April. Data loggers were always orientated to face north to minimise direct sunlight on the temperature-sensing face and were wrapped in shiny metal foil to reflect rather than absorb solar radiation, reducing rapid temperature shifts due to short-term changes in solar radiation, and to protect them from water damage. They were then attached to wooden stakes (approximately 1 cm diameter) by zip ties (around the non-sensing outer edge) with a piece of shielded garden wire (in contact only with the zip tie), and they were suspended in the air (Suppl. material 1: fig. S1). We acknowledge that the exposed positions of the data loggers means that direct solar radiation hit the reflective surface, and may influence the temperatures recorded (Maclean et al. 2021). We view the large number of data loggers deployed as a compromise to more expensive radiation shields, and reduces the risk of artificial temperature amplification caused by shields heating in direct sunlight. We also note that animals, such as insects, will also experience direct solar radiation in this habitat, which will make our results perhaps more realistic than loggers housed in shields. In 2022, a different 45 sampling locations were randomly generated in QGIS version 3.36.3 (QGIS.org 2024) across three replicates of each combination of five aspects (north, east, south, west, flat) and three vegetation types (short grass, long grass, encroaching scrub), with a minimum plot size (consistent aspect and vegetation type) of 5 m2, and a minimum distance of 10 m between each point. All data loggers were placed at 1 cm above the soil facing north, again shielded with shiny metal foil, and recorded temperature hourly. This height was chosen to represent near-ground conditions and was also partially chosen to reflect other data collected as part of a wider study. These temperatures should reflect the conditions that ground-dwelling organisms experience. Figure 1. Totternhoe Quarry nature reserve, with vegetation type (mapped from drone images in 2018 and updated in 2020, shown in colour) and topographic aspect (calculated from the OS Terrain 5 map at 5-m resolution, shown with hatching lines) illustrated. Points show locations with different combinations of topographic aspect and vegetation type where temperature data loggers were placed in 2018-2019 and 2022, and environmental structure was recorded. Base layer ESRI Satellite imagery, accessed via QuickMapServices plugin in QGIS 3.42.2. This includes imagery from Esri, DigitalGlobe, GeoEye, i-cubed, USDA FSA, USGS, AEX, Getmapping, Aerogrid, IGN, IGP, swisstopo, and the GIS User Community. Frontiers of Biogeography 18, 2025, e164843 Microclimates often amplify extreme temperatures 5 After placing the data loggers, environmental variables were recorded from the 5 m2 surrounding each point. Topographic aspect was recorded as the predominant slope direction (assuming there was a slope, if not the point was recorded as Flat), recorded on a compass by the observer turning to face directly downhill. Slope was estimated in degrees in categories (0°, 1–10°, 11–20°, 21–30°, 31–40°, or > 40°). Shelter was recorded by scoring one point for each cardinal direction in which there was an obstruction (vegetation or topography) within 5 m of the point at chest height (from 1 (no obstructions) to 5 (obstructions in all directions), common obstructions included steep slopes and scrub). Vegetation type was recorded in three categories: encroaching scrub (25–75% of the surrounding 5 m2 was scrub), short grass (<25% scrub, average grass height < 10 cm), long grass (<25% scrub, average grass height > 10 cm). In 2022, two additional variables were collected due to the likelihood of the ground-level data loggers being particularly affected by fine-scale local structure: the percentage of bare ground within 5 m2 of the data logger was estimated by eye, and the average vegetation height at time of data logger deployment was recorded by gently lowering a clipboard onto the vegetation until it was supported, and the height of the clipboard above the ground was measured. After three months (or six months in winter), the data loggers were retrieved (and replaced in 2018–2019 to allow for continuous monitoring). Missing, moved, or tampered data loggers were noted and excluded from analysis. The temperature recordings were checked for outliers by visual inspection of the data, removing incomplete datasets, or complete datasets for data loggers where temperature recordings appeared nonsensical for the location and time of year (e.g. temperatures <-20 °C, n = 20). Macroclimate data Macroclimate data were acquired through the Met Office MIDAS Open database, where hourly air temperature data were acquired from a nearby weather station (Woburn, 52.01400, -0.59457) (approximately 16 kilometers from the study site) (Met Office, 2019) (89 metres above sea level). The weather station is positioned in a similar habitat composition to the landscape of interest (in that it is placed within an open grassy area comparable to a grassland). The weather station thermometer recorded air temperature at a height of 1.25 m (Met Office, personal communication, February 5th 2025), and was shielded with a Stevenson screen. A weather station was chosen for comparison rather than microclimate models due to weather station data still being commonly used in ecological studies, providing real-world temperature measurements for comparison, and its temperature recordings coming from a similar landscape. To date, microclimate modelling methods are still at relatively coarse resolution (e.g. ERA5 at 31 km, ERA5-Land at 9 km), which would incorporate a significant proportion of the surrounding landscape which is not similar in composition to the study site (being largely agricultural land and urbanised areas). Data analysis Data analysis took place in R version 4.3.3 (R Team, 2024). The data were initially prepared using the ‘dplyr’ (Wickham and Romain, Francois 2023) and ‘lubridate’ (Grolemund and Wickham 2011) packages. Plots were produced in ‘ggplot2’ (Wickham 2016) and ‘sjPlot’ (Lüdecke 2021). All temperature recordings (microclimate and macroclimate) were cut to the same date-times, starting from midnight after the last data logger was placed out, and running to midnight before the first data logger was collected. The difference in temperature between each microclimate recording and the macroclimate was calculated for each date-time (hereafter: microclimate offset). A positive value indicated that the microclimate was warmer than the macroclimate (a ‘heat trap’), and so the weather station underestimated the temperatures species would experience locally. A negative value indicated the microclimate was cooler than the macroclimate (a ‘cool refugia’), and so the weather station overestimated the temperatures species would experience. A value of zero indicated that the microclimate and macroclimate temperatures were the same, and so the weather station accurately reflected the local conditions species would experience. Does the offsetting effect of microhabitats change with environmental structure: topographic aspect, slope steepness, amount of bare ground, degree of shelter, vegetation type, or vegetation height, and which structures are most important? Due to the large number of variables and data points, and the possibility of complex interactions between variables, a random forest classification model was implemented to predict which environmental structures had the highest predictive power when it came to microclimate offsetting, and the highest probability of maintaining cool refugia, using the ‘randomForest’ package (Liaw and Wiener 2002). Note that only temperature records from the 1 and 5 cm height data loggers was used, with the 50 cm loggers excluded from the random forest model. To understand which environmental variables were important in altering the offsetting capacity of the microhabitats, the first random forest classification model was trained on the environmental variables; aspect, slope steepness, shelter, vegetation type, and macroclimate temperature, with the response variable as microclimate offsetting above or below zero (a binary categorical response). Above zero indicated a heat trap, whereby microclimate temperatures were hotter than ambient (amplified), and below zero indicated cool refugia, whereby microclimate temperatures were cooler than ambient (buffered). We chose to use a binary approach to specifically focus on what land-managers invest time and resources in to benefit nature (cool refugia), or should be avoided (heat traps). A second model was also Frontiers of Biogeography 18, 2025, e164843 Esme Ashe-Jepson et al. 6 trained on only the 2022 data to include all previously mentioned environmental variables and also percentage bare ground and vegetation height, which were only recorded in that year. The random forest models were built using 500 decision trees. These models then ranked each variable in importance in improving the predictions by calculating the mean decrease in accuracy. This metric quantifies how much the model accuracy drops when each variable is permutated: a higher mean decrease in accuracy corresponds to a variable that is more important for accurately predicting offsetting capacity. Model performance was checked using out-of-bag (OOB) error estimations. The model was trained on a bootstrap sample (67%) with the remaining (33%) used to test tree predictions. Each tree in the model was trained on equal-sized class samples, which slightly elevated the OOB (from 33.5% to 34.7%), but resulted in more balanced prediction accuracy across classes (55.2% and 17.8% to 36.0% and 33.8% respectively). To account for minor fluctuations in variable importance due to the random sampling inherent in random forests and bootstrapping, we re-ran the model 10 times and averaged the mean decrease in accuracy for each variable to produce more stable importance rankings. Similar random forest models were fitted with the same structure, except that all categorical variables were transformed to dummy variables. This was done to more specifically test how each different level of each environmental variable influenced the likelihood of a microhabitat maintaining cool refugia specifically. Again, a separate model was run for 2022 to include vegetation height and percentage bare ground (as continuous variables these were not dummied). The difference in predictive power within each variable was then visualised using partial dependence plots, which show differences within variables while holding all other variables constant (for example to identify differences in the dependent variable (positive or negative microclimate offsetting) between north and south facing slopes with all other variables held constant). This was done by calculating log-odds of microclimate offsetting by averaging the models’ predicted probabilities of a microclimate being cooler than ambient (cool refugia) while changing a single variable and holding all other variables constant. A higher log-odds value indicates a greater likelihood that the model predicts cool refugia in that microhabitat. Does the offsetting effect of microhabitats change with increasing height above the ground? To test whether microclimate offsetting varied with height above the ground, a linear mixed effects model was fitted with microclimate offsetting as the response variable, and macroclimate temperature, data logger height and their two-way interaction as explanatory variables, and data logger identity as a random effect. A significant interaction would imply that microclimate offsetting differs across air temperatures at different heights above ground. Model assumptions and performance of the random effect were checked before fitting. Does the offsetting effect of microhabitats change with increasing air temperature, and is there a particular temperature at which this relationship changes? To test whether microclimate offsetting changed with increasing air temperatures, and whether there was a particular macroclimate temperature which altered this relationship, a segmented linear mixed effects model was fitted using the ‘segmented’ package (Fasola and Muggeo 2018). First, a linear mixed effects model was fitted using the ‘nlme’ package (Pinheiro et al. 2023) with the microclimate offset as the response variable, and macroclimate temperature as the explanatory variable. As it is likely that environmental structure will influence this relationship, environmental variables were also included in the model: aspect, slope, shelter, vegetation type, vegetation height, and percentage bare ground. Interaction terms were also included between all environmental variables and macroclimate temperature. By including these environmental variables, the effect of macroclimate temperature on microclimate offsetting could be estimated independently of these variables. Data logger identity was included as a random effect. Performance of the random effect was checked using the ‘lmerTest’ package (Kuznetsova et al. 2017). The model was then tested for evidence of a change in slope using a Davies test, and the optimal breakpoint was extracted using the ‘segmented’ package (Fasola and Muggeo 2018). The segmented and non-segmented models were compared with AIC. To account for low heteroscedasticity, robust coefficients were calculated with the ‘broom.mixed’ package (Bolker and Robinson 2024). Which combinations of environmental structures result in the strongest microhabitat offsetting effect with increasing air temperatures? To determine which combinations of environmental variables were most effective in maintaining cool refugia during heatwaves, a linear mixed effects model was fit to the data from all years (therefore excluding the variables percentage bare ground and vegetation height, to reduce the number of models run and missing values impacting the results), with microhabitat offsetting as the response and all possible combinations of all environmental variables, macroclimate temperature, and an interaction term between the macroclimate temperature and environmental variables. Data logger identity was included as a random effect. Note that temperature recordings at 50 cm height were excluded from this analysis. We extracted the slope of the relationship between macroclimate air temperature and the microhabitat offset for all models. These slopes were then ranked according to the lowest slope value, as this would indicate combinations that could Frontiers of Biogeography 18, 2025, e164843 Microclimates often amplify extreme temperatures 7 maintain either cool refugia (negative slopes), or microclimate temperatures close to ambient temperatures without heat amplification (slopes close to zero). Results In total, 629,503 microclimate temperature recordings were recovered from 131 data loggers (309,072 in 2018 (172,436 at 5 cm, 136,636 at 50 cm), 270,600 in 2019 (154,161 at 5 cm, 116,439 at 50 cm) (Suppl. material 1: fig. S2A), and 49,831 in 2022 (all at 1 cm)) (Suppl. material 1: fig. S2B). Microclimate temperatures ranged from -16.0 to 63.0 °C, with a mean ± standard deviation of 14.8 ± 7.6 °C. Across 10,155 records, the macroclimate temperatures ranged from -7.1 to 39.3 °C, with a mean ± standard deviation of 14.5 ± 6.0 °C. Six heatwave events (as defined by the Met Office as extended periods of hot weather relative to the expected conditions of the area for that time of year) were captured during recording (June 22nd to August 8th in 2018, February 21st to 28th, July 21st to 29th, August 23rd to 30th in 2019, and June 15th to 17th, July 17th to 19th, and August 9th to 10th in 2022). During these heatwaves, microclimate temperatures ranged from 17.2 °C below ambient temperature to 36.0 °C above ambient. However, high temperatures were not only recorded during heatwave events, with 3,078 microclimate temperature recordings over 40 °C recorded during the surveyed period, 1,585 of which occurred outside of the defined heatwave events (51.3%). Does the offsetting effect of microhabitats change with environmental structure: topographic aspect, slope steepness, degree of shelter, vegetation type, amount of bare ground, or vegetation height, and which structures are most important? Overall, macroclimate temperature was the most influential variable in determining whether microhabitats became heat traps or maintained cool refugia, followed by vegetation type, aspect, and shelter score, which all performed similarly well, followed by slope steepness (Fig. 2A). In 2022, which included the variables vegetation height and percentage bare ground, vegetation height ranked the highest of all the environmental variables, and percentage bare ground the lowest. The inclusion of these extra variables also changed the ranking of vegetation type and aspect, though they performed relatively similarly in their impact on model accuracy. In all cases, all environmental variables tested improved the model’s ability to predict whether microhabitats buffered or amplified temperatures. All environmental variables tested influenced the ability of microhabitats to maintain cool refugia specifically (Fig. 3). In particular, aspect (Fig. 3A), shelter (Fig. 3C), and vegetation type (Fig. 3D) performed well, with north-facing slopes, medium shelter scores, and encroaching scrub having the highest probabilities of maintaining cool refugia. To a lesser extent, intermediate slopes (Fig. 3B), low percentages of bare ground (Fig. 3E), and high vegetation height (Fig. 3F) also contributed to increased probabilities of maintaining cool refugia. Does the offsetting effect of microhabitats change with increasing distance from the ground? There was a significant change in microclimate offsetting with distance from the ground (χ2 = 11.082, d.f. = 1, p < 0.001) (Fig. 4, Suppl. material 1: fig. S3), with temperatures at 50 cm above the ground being similar to macroclimate temperatures and remaining relatively stable with increasing air temperatures, whereas temperatures near the ground (1 and 5 cm) were increasingly warm as air temperatures increased. Figure 2. Environmental variables ranked in importance according to how much they contributed to the prediction accuracy of the random forest models in all years (A) and only 2022 (B), which was tested separately due to including two new variables (vegetation height and percentage bare ground). All decrease in accuracy values are averaged from 10 models. Error bars show the standard deviation in the decrease in accuracy values across the 10 models. Frontiers of Biogeography 18, 2025, e164843 Esme Ashe-Jepson et al. 8 Does the offsetting effect of microhabitats change with increasing air temperature, and is there a particular temperature at which this relationship changes? There was a significant change in the relationship between macroclimate temperature and microclimate offsetting at a macroclimate temperature of 7.24 °C ± 0.07 (d.f. = 323217, p < 0.001), which results in a significant change in slope at this temperature (Δβ = 0.357, SE = 0.006, t = 62.00), this was confirmed by the Davies test (p < 0.001) (Fig. 5). Below this temperature, microclimate offsetting decreased (i.e. microclimate temperatures got marginally cooler relative to macroclimate temperatures) as macroclimate temperature increased (β = -0.159, SE = 0.006, t-value = -25.079, d.f. = 376329, p < 0.001). Above the breakpoint, microclimate offsetting increased (i.e. microclimate temperatures got warmer relative to macroclimate temperatures) as macroclimate temperature increased (β = 0.198). Figure 3. The marginal effects on the log-odds of a microhabitat having temperatures below ambient conditions (cool refugia) from the random forest model. Each environmental variable’s effect is shown separately with all other variables held constant, with scale bars standardised to ease comparisons. Higher marginal effects indicate a higher probability of a microhabitat being below ambient temperature (cool refugia). Frontiers of Biogeography 18, 2025, e164843 Microclimates often amplify extreme temperatures 9 Figure 4. The relationship between air temperature (macroclimate temperature, from nearby weather station) and microclimate offsetting (the difference between the microclimate temperature and macroclimate temperature) at different heights from the ground (1 cm, 5 cm, and 50 cm). Points represent individual data logger hourly temperature recordings, and are plotted semi-transparent to illustrate where data overlap. Figure 5. The relationship between air temperature (macroclimate temperature, from nearby weather station) and microclimate offsetting (the difference between the microclimate temperature and macroclimate temperature). Points represent individual data logger hourly temperature recordings and are coloured by the microclimate temperature. The black line illustrates the fitted segmented linear relationship, with the breakpoint highlighted with a vertical red dashed line. The dashed horizontal line highlights the microclimate offset values of zero (where the macroclimate and microclimate temperatures are the same). Above this line indicates the microclimate was hotter than the macroclimate, and below the line indicates where the microclimate was colder than the macroclimate.