Effects of climate change on a nocturnal gecko: an integrative analysis

Climate change poses a major threat to biodiversity, especially for ectotherms whose physiology is tightly coupled to environmental conditions. While most studies on climate change emphasize diurnal species, nocturnal reptiles remain largely overlooked. We evaluated the vulnerability of Gymnodactylus amarali —a nocturnal saxicolous lizard from the Brazilian Cerrado—to future climate change scenarios. We combined correlative, mechanistic, and hybrid species distribution models (SDMs) with biophysical simulations to assess the species’ vulnerability. We evaluated range dynamics by measuring rear (lost), stable (retained), and lead (gained) areas under different climate change scenarios. By the end of the century, G. amarali is projected to lose more than 96% of its climatically suitable range across all climate change scenarios. Biophysical models revealed additional constraints: the number of annual hours within the species’ thermal range of activity will decline by more than half relative to current levels, despite reduced time needed for surface heating. Cold stress is unlikely to limit the species, but acute heat exposure is predicted to increase under warming scenarios. Rock–air decoupling remained stable, indicating that rocky habitats will continue to buffer microclimates, although not enough to counterbalance the sharp loss in activity time and distributional range. Our integrated approach shows that G. amarali faces severe contraction risks in both geographic range and functional performance under future climates. Conservation efforts should prioritize protecting rocky habitats and Cerrado remnants that maintain thermal heterogeneity for G. amarali and other saxicolous species, as these features can buffer the impacts of climate change.

Global climate change is one of the main threats to biodiversity, affecting organisms in all biomes and ecosystems (Urban, 2015). Ectotherms are particularly vulnerable because their activity, performance, and survival depend on environmental conditions for thermoregulation (Sinervo et al., 2010; Clusella-Trullas et al., 2021). Climate change can reduce thermally suitable periods for foraging and reproduction, ultimately affecting population viability and contracting species’ ranges (Sinervo et al., 2010; Diele-Viegas and Rocha, 2018; Cox et al., 2022). Among ectotherms, reptiles—particularly lizards—have been used as model systems in ecological studies due to their relative abundance and ease of capture (Camargo et al., 2010). Tropical lizards are considered especially vulnerable, as their environments are already close to their upper thermal limits (Huey et al., 2012; Cox et al., 2022).

Predicting species distribution under varying scenarios of climate change is essential for developing effective conservation plans and avoiding further biodiversity loss. Species distribution models (SDMs) are a powerful tool to accomplish this goal: these models associate species occurrence with environmental conditions to predict the areas with conditions suitable for species persistence (Franklin, 2023; Soley‐Guardia et al., 2024). The most common approach is to use correlative models, which typically relate species’ occurrences to macroclimatic conditions, although they may also incorporate non-climatic predictors such as vegetation, habitat structure, or geology. These models are based on statistical associations and therefore lack an explicit mechanistic basis, often overlooking the physiological processes that constrain species distribution (Dormann et al., 2012; Tourinho and Vale, 2023). In addition, because they implicitly capture the combined effects of environmental conditions and biotic interactions (e.g., competition, predation, facilitation), correlative models are generally interpreted as an approximation of the realized niche. In contrast, mechanistic models explicitly incorporate functional traits and physiological limits to describe how organisms interact with their environment, thereby approximating the fundamental niche. For this reason, mechanistic approaches are expected to perform better when extrapolated to novel climatic conditions, such as those imposed by climatic change (Kearney et al., 2010; Sillero, 2011). For reptiles, particularly for lizards, ecophysiological variables related to locomotor performance, such as those derived from thermal performance curves, are informative proxies for an organism’s function, linking body temperature to energy acquisition, refuge use, and reproduction (Angilletta et al., 2002). Hybrid SDMs, which combine correlative and mechanistic predictors, associate cause and effect to more accurately predict the impact of climate change on organisms (Rodríguez et al., 2019). These models tend to perform better than purely correlative and mechanistic models, as they capture both macroenvironmental patterns and proximate processes (Caetano et al., 2020; Machado et al., 2023).

Species depend on suitable macroenvironmental and microenvironmental conditions for survival (Kearney and Porter, 2009; Parlin et al., 2020). Biophysical models provide a process-based framework for simulating these conditions and organismal responses. By applying biophysical equations of energy and mass flow that depend on environmental conditions, these models capture the balances of heat, water, and energy, which are translated into performance metrics (Kearney and Porter, 2020; Briscoe et al., 2023). Biophysical models also allow the simulation of thermoregulatory behaviors such as basking and shuttling between sun and shade, and burrowing. By representing the costs and benefits of these behaviors, they can show if behavioral adaptations can mitigate the impacts of macroclimatic variation and minimize the harm from climate change—including shifts in temperature, precipitation regimes, and the frequency of extreme events (Parmesan and Yohe, 2003, IPCC, 2021)—by altering patterns of daily activity (Adolph, 1990; Smith and Ballinger, 2001; Briscoe et al., 2023).

Climate change studies often assume global warming is constant throughout the day or focus only on diurnal patterns, treating nocturnal warming as negligible (Rutschmann et al., 2024). This bias extends to the species level, as most studies focus on diurnal species, even though more than one-fourth of lizards have nocturnal habits – primarily members of Gekkota (Avery, 1982, Huey, 1982, Kearney and Predavec, 2000). The nocturnal, saxicolous gecko Gymnodactylus amarali Barbour, 1925, from the Brazilian Cerrado, provides an ideal model organism for evaluating how climate change may affect the thermal landscape available to tropical reptiles. This species is widely distributed, mainly in central and northern Cerrado (Vanzolini, 2005), a biome with high levels of endemism that is under severe anthropogenic threat and, for this reason, is considered a biodiversity hotspot (Myers et al., 2000). The Cerrado is primarily threatened by habitat conversion for agroecosystems and by climate change, including rising temperatures and intensified droughts (Colli et al., 2020; Rodrigues et al., 2020). Habitat conversion can also alter microclimatic conditions by reducing vegetation cover and modifying substrate properties, thereby affecting the availability and thermal buffering capacity of microhabitats such as rock outcrops (Scheffers et al., 2014; Sears et al., 2016).

Here, we combined ecophysiological measurements (preferred temperature, voluntary thermal limits, locomotor performance, and thermal tolerance), species distribution models, and biophysical models to evaluate the vulnerability of Gymnodactylus amarali to future climate change. We assessed models including environmental and ecophysiological variables by comparing their predictive performance. Our ecophysiological variables focused on thermal ecophysiology since temperature is a primary determinant of performance in ectotherms and is central to mechanistic modeling approaches. We acknowledge that hydric processes may also influence species responses, particularly in geckos, and that these effects were only indirectly captured here through precipitation-related variables. We projected the best model under present and future conditions and evaluated range dynamics, identifying rear (lost), stable (remain stable), and lead (gained) areas. Finally, we simulated microclimates and lizard activity in these areas to further illustrate the effects of future climate change on G. amarali, including thermal constraints experienced during both nocturnal activity and daytime inactivity, since daytime inactivity within refuges can be a critical component of the thermal environment in nocturnal geckos (Angilletta et al., 1999; Vasconcelos et al., 2012). From these simulations, we derived biophysical metrics to provide proximate evidence of climate change impacts on the species. We expected the hybrid SDM to outperform other approaches, reflecting the integration of macroenvironmental and ecophysiological predictors. We predicted significant reductions in suitable habitat and activity opportunities for G. amarali, consistent with previous studies on tropical lizards. Regarding biophysical variables, we predicted that, within suitable habitats, G. amarali would experience a reduction in hours of activity as the climate warms, with shorter periods within its preferred temperature range.

To predict the effects of climate change on the future distribution of Gymnodactylus amarali, we created three SDMs: (1) a model with only bioclimatic predictors, (2) a mechanistic model with only predictors of thermal ecophysiology, and (3) a hybrid model combining both types of predictors. We compared their performance, projected the best model to present and future environmental conditions, evaluated range dynamics, and ran biophysical simulations to assess the effects of climate change on microclimates and the thermal environment available to G. amarali.

Occurrence data

We assembled an expert-evaluated occurrence dataset from the literature and scientific collections, comprising 85 records of Gymnodactylus amarali. We used Kernel Density Estimation with the adehabitatHR package (Calenge and Fortmann-Roe, 2024) to represent the potential distribution (Franklin, 2023; Soley‐Guardia et al., 2024), applied a geographic filter (distance = 50 km) to reduce sampling bias (Boria et al., 2014), and retained 42 occurrence records in the final dataset (Figure 1).

Environmental predictors

We downloaded the 19 bioclimatic variables from the WorldClim 2 database at 2.5 km (Fick and Hijmans, 2017) and cropped them to Brazil’s extent. We used Variance Inflation Factors (VIFs) analyses with the usdm R package (Naimi et al., 2014) to assess multicollinearity and retained variables with VIF < 4 and maximum pairwise correlation of 0.7: mean diurnal range (BIO 2), isothermality (BIO 3), mean temperature of wettest quarter (BIO 8), precipitation seasonality (BIO 15), precipitation of the warmest quarter (BIO 18), and precipitation of the coldest quarter (BIO 19), encompassing both temperature- and precipitation-related predictors.

Ecophysiological predictors

We collected Gymnodactylus amarali in the Bacaba Municipal Park (14° 42’ 23.87” S, 52° 21’ 9.59” W; Nova Xavantina, Mato Grosso, Brazil), located in the western portion of the Cerrado, near the ecotone with Amazonia (Abad and Marimon, 2008). We conducted sampling by active search in rupestrian cerrado with rocky outcrops during August-September 2015, December 2016, August-September 2017, September 2018, and April 2019. Searches began in the twilight and lasted approximately three hours, coinciding with the species’ activity period. We identified captured lizards and transported them to the laboratory.

We measured four sets of physiological variables of 43 individuals following Sinervo et al. (2010): preferred temperature, voluntary thermal limits, locomotor performance, and thermal tolerance. We took all measurements at night to match the species’ activity period. We obtained preferred temperatures (T pref) by placing individuals in a 20-50 °C gradient for one hour and recording body temperature every minute with thermocouples. Although longer sampling periods may capture broader temporal variation in thermal preferences, the one-hour protocol follows standard approaches in reptile thermal biology and reflects logistical constraints inherent to field-based studies. We acknowledge that repeated measurements at short intervals may introduce temporal autocorrelation; however, as these data were used to estimate summary statistics (e.g., percentiles) rather than temporal dynamics, this limitation is unlikely to affect our conclusions. We excluded the first 10 minutes as an acclimation period and defined voluntary thermal limits as the 5 th and 95 th percentiles of the temperatures measured along the gradient. By using percentiles, we avoided the potential effects of outliers and represented individuals’ movements to avoid excessive warming or cooling in their natural habitats. For each lizard, we measured locomotor performance at T pref and T pref ± 5 °C, with each individual tested twice per temperature. To reduce the influence of lack of motivation on performance estimates, the maximum speed recorded for each individual at each temperature was used for analyses, following common practice in performance studies (Losos et al., 2002). We recorded lizards at 420 fps as they ran on a 2 m track and extracted maximum speeds with Tracker 4.80 (https://opensourcephysics.github.io/tracker-website/). Individuals rested for at least 30 minutes between runs at different temperatures. We fitted thermal performance curves with generalized additive mixed models (GAMMs), including body temperature and body size (snout-vent length, SVL) as fixed effects and individual as a random effect. The model considering body size performed best and was retained (Supporting Information Table 1). We measured critical thermal limits by heating individuals under lamps for the critical thermal maximum (CT max) and cooling them with gel ice packs for the critical thermal minimum (CT min). The loss of righting response defined each threshold, and we measured body temperature with a quick-reading cloacal thermometer with a precision of 0.2 °C (T-6000, Miller & Weber, Inc.).

We collected microclimate data with HOBO dataloggers (HOBO Pro v2 Temperature/Relative Humidity, Onset Computer Corporation, MA, USA) that recorded air temperature and relative humidity every 10 min. We placed sensors near pitfall trap arrays in the Bacaba Municipal Park (50 cm above ground and sheltered from rain and direct solar radiation by a PVC cover). Although we sampled lizards on nearby rocky outcrops rather than directly at the traps, these sensors provided representative microclimatic conditions for the study area. From these records, we estimated hourly/daily/trap averages and monthly absolute values (minimum, maximum), averages, and standard deviations. We used these microclimate data to calculate the hours of activity rasters as a representation of the range of thermal conditions potentially experienced by the individuals.

We created hours of activity and performance rasters with the Mapinguari R package (Caetano et al., 2023). We derived performance from a generalized additive mixed model (GAMM) with size as a covariate, applied to temperature-related bioclimatic variables. We calculated hours of activity from nocturnal microclimatic data recorded by dataloggers at collection sites, then fitted a logistic model relating these hours to macroclimatic temperatures. We applied this model to rasters of temperature-related bioclimatic variables, generating candidate extrapolations of hours of activity across the species range (Caetano et al., 2023; Machado et al., 2023). We defined activity hours as the proportion of time when environmental temperatures fell within the voluntary thermal limits. After screening for multicollinearity we retained variables with VIF < 3 and maximum pairwise correlation of 0.7: hours of activity for the maximum temperature of the warmest month (BIO 5), and performance rasters for maximum temperature of the warmest month (BIO 5), minimum temperature of the warmest month (BIO 6), mean temperature of the driest quarter (BIO 9), mean temperature of the warmest quarter (BIO 10).

Hybrid SDM

We created a hybrid SDM that included both bioclimatic and ecophysiological predictors. We ran a multicollinearity check with all the macroclimatic and ecophysiological variables (related to both performance and hours of activity), and retained variables with VIF < 4 and maximum pairwise correlation of 0.7: mean diurnal range (BIO 2), isothermality (BIO 3), precipitation seasonality (BIO 15), precipitation of the warmest quarter (BIO 18), precipitation of the coldest quarter (BIO 19), hours of activity for the mean temperature of the wettest quarter (BIO 8), and performance rasters for the maximum temperature of the warmest month (BIO 5), mean temperature of the wettest quarter (BIO 8), and mean temperature of the driest quarter (BIO 9).

Model fitting and evaluation

We fitted models using the random forest (RF) algorithm, which produces highly accurate models with short computation time (Valavi et al., 2022). We used the down-sampled approach, as RF performs better with the same number of pseudo-absences as presences (Barbet‐Massin et al., 2012). We created 20 pseudo-absence datasets with the same number of occurrences, representing the number at which TSS values approached an asymptote for RF models run with 100 occurrences. We sampled pseudo-absences with a surface range envelope (SRE), selecting points outside a defined proportion from presence conditions, an approach suitable when the species distribution is well sampled (Thuiller et al., 2024). We split the dataset into calibration and validation using k -fold cross-validation (k = 4). We ran models with 10 permutations and used 10 permutations to evaluate variable importance. We checked the area under the ROC curve (AUC) and TSS as these are the most accurate metrics (Allouche et al., 2006). Our final model was an ensemble with models with TSS above 0.7. We used the weighted mean (i.e., probabilities from selected models are weighted by their evaluation scores) because this ensemble approach performs best compared to other techniques (Marmion et al., 2009; Hao et al., 2019). The performance metrics for the single models and the final ensemble model are in the Supporting Information. We used the biomod2 package (Thuiller et al., 2024) for the SDM workflow.

Comparing SDM performance

We assessed model performance with linear mixed-effects models (LMMs). We used calibration scores as the response, the model type (correlative, mechanistic, hybrid) as the fixed effect, and the pseudo-absence set and the algorithm run as random effects. We compared models with and without “model type” to test its effects on model fit. We extracted the conditional modes (Best Linear Unbiased Predictors, BLUPs) and their associated posterior variances to partition variance. We conducted all analyses in R with lme4 (Bates et al., 2015) and nlme (Pinheiro and Bates, 2000; Pinheiro et al., 2025).

Projection and range dynamics

To estimate the effects of future climate change on the species range, we projected the final ensemble model to present and future conditions. We used future conditions until the end of the century (2041-2060, 2061-2080, and 2081-2100) under optimistic (SSP245), intermediate (SSP370), and pessimistic (SSP585) scenarios of climate change. We used the EC-Earth3-Veg Global Circulation Model (GCM), which performs well for South America (Almazroui et al., 2021; Bazzanela et al., 2023). We obtained the future conditions from the geodata database, directly downloaded with the cmip6_world function from the geodata package (Hijmans et al., 2024). We created binary maps using a threshold that balances sensitivity and specificity (true positive and negative rates, respectively), aiming to balance both commission and omission errors (falsely predicting presence or absence, respectively) (Guisan and Thuiller, 2005; Liu et al., 2005). We used the biomod2 function BIOMOD_RangeSize to quantify range dynamics between present and future conditions by the end of the century. We classified areas as rear (lost, no longer suitable in the future), stable (retained, suitable today and likely to remain so), or lead (gained, not suitable today but likely to become so). We reported these values as pixel counts and as percentages.

Microclimate and lizard simulation with biophysical models

We simulated microclimates and lizard activity under present conditions and scenarios of climate change with the NicheMapR package (Kearney and Porter, 2017, 2020). We aimed to compare the microclimate and lizard activity among the rear, stable, and lead areas created under these climate change scenarios. We randomly sampled 100 coordinates from each of the three range-dynamic areas (rear, stable, lead) for each of the three climate change scenarios, totaling 900 coordinates (3 areas x 3 SSPs). For each of these coordinates, we simulated the microclimate and lizard activity under present conditions and under the corresponding climate change scenario. We used the warm argument from the ectotherm function, with these offset warming temperatures: SSP245 = +2 °C; SSP370 = +3 °C; SSP585 = +4 °C. This approach allowed evaluating the effects of gradual warming scenarios on the microclimates available for Gymnodactylus amarali and the resulting lizard activity under broad-scale climatic conditions, without explicitly accounting for local land-use or hydrological variation.

The NicheMapR workflow has two primary steps. First, we used the micro_global function to compute microclimates for a whole year for a nocturnal, crevice-dwelling lizard, with conditions matching the species’ biology (Kearney and Porter, 2017). micro_global uses habitat properties from the WorldClim dataset, thereby ensuring consistency between macroclimatic and microclimatic inputs. To represent the most thermally extreme environment available for Gymnodactylus amarali, we parameterized the model with minimum shade (minshade = 0) and extracted only the open-surface output. Although the species is primarily nocturnal and uses rock crevices, the minimum shade scenario captures the upper limit of environmental heating and the slow cooling of exposed rock surfaces at night. This approach provides an estimate of potential thermal constraints under present and future climates, enabling evaluation of the species’ vulnerability to extreme heat. However, it may underestimate thermal stress if diurnal temperatures in superficial refuges approach or exceed critical limits, particularly in open and seasonally dry environments such as the Cerrado, where microhabitat buffering may be limited. The second step used the ectotherm function, which takes the output from the microclimate model and calculates heat and water exchange with the environment, considering morphological, physiological, and behavioral traits, and computes body temperature, activity states, and other aspects of the mass budget (Kearney and Porter, 2020).

To capture a multi-dimensional view of the thermal niche for Gymnodactylus amarali and how it may be altered under climate change, we quantified a suite of variables. We predicted hourly body temperatures and activity states such as time spent surface heating, foraging, and active (surface heating + foraging), which served as proxies for time available for critical functions. We also measured annual exposure to critical thermal limits to represent the frequency of potentially lethal or damaging exposures across different body compartments and environments. We measured nighttime temperatures to better reflect the thermal environment used by the species. We also measured degree-hours related to critical thermal limits, as these integrate both duration and magnitude of stress. Finally, we measured rock–air decoupling indices to evaluate how much rock microclimates buffer or amplify air temperature. Our final dataset for biophysical variables included total annual hours spent foraging, surface heating, and active; annual and nocturnal hours above critical thermal maximum (CT max) and below critical thermal minimum (CT min) for body (T c), air (T a), air at lizard’s height (T aloc) and substrate (T sub); and maximum and minimum nocturnal temperatures for the same variables. We plotted the activity states for the whole year and selected months to represent different environmental conditions.

Ecophysiological predictors

Preferred body temperatures of Gymnodactylus amarali ranged from 16.3 to 39.7 °C, with the 5 th and 95 th percentiles as 17.8 and 34.5 °C. Minimum and maximum critical limits were respectively 7.9 and 40.5 °C. We present the variation in critical thermal limits in Supplementary Figure 1. Locomotor performance increased with body temperature, peaking at 31.6 °C. The generalized additive mixed model (GAMM) including body size as a covariate performed significantly better than the size-free model (Supplementary Table 1), indicating that size influences locomotor ability. We retained this size-inclusive model to generate the species’ thermal performance curve (Supplementary Table 2; Supplementary Figure 2).

Species distribution models

We compared model performance with random effect models. Performances were broadly similar, but the hybrid model performed slightly better, and we used it as the final model (Figure 2; Supplementary Figure 3). We provide the mixed-effects model results in Supplementary Table 3. The most suitable areas for Gymnodactylus amarali are concentrated in the central Cerrado and in a portion of the southern Amazonia-Cerrado transition (Figure 3). Mean diurnal range (BIO 2), isothermality (BIO 3), and precipitation seasonality (BIO 15) were the most important predictors. Performance on the maximum temperature of the warmest month (BIO 5) was the least important predictor. The remaining predictors (one precipitation-related and four ecophysiological variables) had similar importance values (Supplementary Figure 3d), suggesting that ecophysiological variables contributed additional, but not dominant, explanatory power relative to macroclimatic predictors. Gymnodactylus amarali is projected to lose a substantial portion of its suitable range under all scenarios of future climate change. We summarize range dynamics in Table 1 and Figure 4, and show successive shifts in Supplementary Figure 4.

Biophysical models

Biophysical simulations indicate that Gymnodactylus amarali will spend less time at the surface for heating in the future, and that its total annual active hours will decline sharply across all scenarios. Under present conditions, the species has approximately 12,500 active hours per year, but this is predicted to decrease to fewer than 5,000 under all scenarios (Figure 5A). Rock–air decoupling was never negative, indicating that rocks consistently buffered upward relative to air by storing heat during the day and releasing it at night. Decoupling metrics were stable across scenarios: both the mean and 95 th percentile intensities of decoupling (Figure 5B) and the cumulative positive degree-hours of divergence (Figure 5C) changed little with warming. Daily and nightly maximum and minimum temperatures increased consistently across all scenarios (Figures 6A, B). We observed no degree-hours below CT min or above CT max during either day or night. Air, substrate, and lizard body temperatures never fell below CT min at night or day. At night, body, air, and lizard-height temperatures never exceeded CT max. Under present conditions, no temperatures surpassed CT max; however, future warming scenarios showed exceedances (Figure 6C). Only lizard-height temperatures occasionally dropped below CT min, and this occurred only under present-day conditions (Figure 6D).

Our results indicate that Gymnodactylus amarali is highly vulnerable to climate change, driven by both distributional and functional constraints. The hybrid species distribution model (SDM), which integrated macroclimatic and ecophysiological variables, outperformed models based solely on environmental and ecophysiological predictors and projected substantial range contractions across all climate change scenarios. This improved performance reflects the incorporation of multiple climatic dimensions, including both thermal and hydric factors. Biophysical simulations further indicated a sharp reduction in annual activity hours, despite reduced reliance on surface heating. Although cold stress is unlikely to be a limiting factor, extreme heat events are expected to become more frequent. Rock–air thermal decoupling remained stable, suggesting that rocky habitats will continue to buffer microclimates, albeit not sufficiently to counterbalance the loss of active hours.

Integrating ecophysiological variables, which provide important biological information about proximal species’ traits, improved our SDM by capturing biologically meaningful constraints that models based solely on environmental predictors overlook. In addition, the improved performance of the hybrid model likely reflects the incorporation of multiple climatic dimensions, as macroclimatic predictors included both temperature and precipitation-related variables, whereas the mechanistic component was based exclusively on thermal ecophysiology. This mismatch indicates that hydric processes—captured indirectly through precipitation variables—represent an additional axis of the species’ niche not accounted for in the mechanistic framework. This integration revealed narrow thermal margins, consistent with expectations for tropical ectotherms, which are predicted to be highly sensitive to warming as they already live in climatic conditions close to their thermal limits (Huey et al., 2012; Clusella-Trullas et al., 2021). Furthermore, the sharp decline in locomotor performance above optimal temperatures mirrors patterns reported in other lizard species, reaffirming the limited capacity of species to cope with climate warming. Although reptiles are generally less constrained by water balance than amphibians due to their keratinized skin, water loss can still represent a relevant physiological challenge, particularly in geckos, which may exhibit comparatively higher permeability (Rato and Carretero, 2015). In this context, hydric processes may represent an additional axis of constraint not explicitly captured by thermal performance curves, potentially contributing to differences in model performance.

In this context, and consistent with broader patterns in tropical ectotherms, tropical reptiles usually have limited physiological ability for thermoregulation, depending on behavioral mechanisms, such as varying patterns of activity, adjusting posture, and shuttling between sun and shade (Cowles and Boger, 1944, Avery, 1982, Huey, 1982, Kearney and Predavec, 2000). Future climate change is predicted to affect lizards primarily by reducing the time available within suitable thermal limits, a key constraint linking environmental conditions to species distribution (Sinervo et al., 2010; Caetano et al., 2020). Our findings for Gymnodactylus amarali contrast with several other studies on lizard vulnerability. For instance, populations of the lizard genus Stenocercus will face a thermal deficit but will have thermal refugia that prevent reduced performance or thermal stress (Guerra-Correa et al., 2020). Other studies show more moderate effects, with the Amazonian lizard Kentropyx calcarata showing extinction risk of 26.1%-70.8% by 2050 (Pontes-da-Silva et al., 2018), and South American Teius showing only a minor decrease (Minoli et al., 2019). Conversely, the desert lizard Liolaemus ditadai will increase suitable habitats by 12–27% (Lara-Reséndiz et al., 2024). These patterns contrast with our findings for G. amarali and reinforce that the effects of climate change are species- and context-specific.

For ectotherms, integrating behavioral and other proximal variables into future climate change models is crucial, as appropriate environmental temperatures for thermoregulation are a vital resource for survival (Kearney and Porter, 2004; Pearson et al., 2006; Kearney et al., 2009). A more realistic evaluation of the impact depends on the capacity to buffer air temperatures through behavioral thermoregulation. Biophysical models allow integrating these traits and, consequently, obtaining more realistic predictions and a mechanistic understanding. Studies suggest that future macroenvironmental conditions may become unsuitable for the species, but microenvironmental conditions could compensate (Sears et al., 2016). G. amarali is predicted to have fewer total annual active hours under all scenarios of climate change when compared to the present. G. amarali will need to spend less time on surface heating, which, by itself, could be positive. Individuals are usually more exposed to predators when surface heating, and this should present a trade-off between thermoconforming and thermoregulating (Huey and Slatkin, 1976; Webb and Whiting, 2005). However, G. amarali will have fewer active hours in the future, such that the benefit of less time spent on surface heating will likely be outweighed by this overall decrease in active hours. Reduced active time in turn reduces opportunities to forage or reproduce, ultimately resulting in population-level effects, underscoring the complexity of behavioral responses (Kearney et al., 2009; Sinervo et al., 2010). Even in heterogeneous landscapes, active hours can decrease if preferred microclimates become too rare to locate without overheating (Sears and Angilletta, 2015). Thus, our results show the species’ potential to buffer adverse macroenvironmental conditions by adjusting their microhabitat. Such integration of macro and microecological perspectives will be increasingly needed in the context of climate change (Kearney et al., 2009). Biophysical evaluations provide a more accurate representation of how animals buffer air temperatures and are increasingly recognized as vital for evaluating the impacts of climate change at local scales.

The tropics tend to exhibit greater environmental heterogeneity, offering multiple opportunities for thermoregulation. The availability of microclimates within suitable temperatures is crucial for nocturnal lizards, since direct solar radiation is not available for thermoregulation. Nocturnal species are usually active at lower environmental temperatures than diurnal species, since nighttime temperatures are often lower than their preferred temperatures (Kearney and Predavec, 2000; Nordberg and Schwarzkopf, 2019). Our biophysical simulations indicate that G. amarali currently experiences a thermal environment that rarely imposes exposure to its critical thermal limits. Neither air, substrate, nor body temperatures were projected to fall below CT min, but temperatures exceeded CT max only under future-climate scenarios. These results suggest that lethal cold stress is unlikely to be a limiting factor for this species, while heat stress may become increasingly relevant with warming. The finding that temperatures at lizard height can occasionally drop below CT min only under present-day conditions highlights the potential for microhabitat use (e.g., crevices, ground-level shelter) to buffer individuals from rare cold exposures. Importantly, we did not detect degree-hours below CT min, and nighttime degree-hours above CT max were absent across all scenarios. These results indicate that sublethal stress accumulation from prolonged exposure to extremes is unlikely under current conditions. Thus, thermal constraints appear limited during nocturnal activity, but may arise during daytime inactivity in refuges, as diurnal temperatures in superficial refuges may approach or exceed critical limits, particularly in open environments (Vasconcelos et al., 2012).

G. amarali is a saxicolous species that primarily uses rock crevices and spaces beneath rocks as refuges, microhabitats that can provide thermal buffering and retain moisture but may still experience substantial heating depending on exposure and depth (Colli et al., 2003). Thus, individuals occupying shallow or exposed refuges may be vulnerable to extreme temperatures during the day. The occurrence of microclimates exceeding CT max under warming scenarios suggests a shift in risk toward acute heat events, which could reduce active hours or necessitate greater reliance on behavioral thermoregulation. Importantly, because our simulations represent open-surface conditions and do not explicitly model the thermal dynamics of specific refuge types, thermal stress during inactivity may be underestimated, particularly in shallow or exposed refuges. This threat is significant because tropical species are often close to their thermal safety margins and are highly vulnerable even to small temperature increases (Huey et al., 2012). By incorporating degree-hours, we account for both the intensity and duration of possible thermal stress and account for exposure time, a relevant measure regarding thermal effects (Rezende et al., 2014).

Rock–air decoupling dynamics provide further insight into the buffering role of microhabitats. The consistently positive decoupling values demonstrate that rocks act as thermal capacitors, absorbing heat during the day and releasing it at night. This buffering may reduce the risk of sudden cooling and provide extended opportunities for nocturnal or crepuscular activity. The stability of both mean and extreme (95 th percentile) decoupling intensities across climate scenarios suggests that the functional role of rocks as thermal refuges will persist under future warming, even as maximum and minimum environmental temperatures increase. The positive cumulative degree-hours of divergence reinforce this interpretation, highlighting the continued importance of rocky microhabitats in mediating thermal stress. Thus, our findings emphasize the role of rocks as thermal refugia that extend activity time (Scheffers et al., 2014; Sears and Angilletta, 2015). Taken together, our findings suggest that G. amarali will likely remain buffered from cold stress across its range, while warming climates may increasingly constrain activity through acute heat exposure. Nevertheless, the persistence of rock–air decoupling indicates that behavioral access to rock microhabitats will remain a key strategy for mitigating thermal stress, underscoring the importance of habitat structure in shaping this species’ resilience to climate change. Additionally, even slight increases in minimum night temperature could reduce the thermal gradient that rocks provide for nocturnal ectotherms.

Even though agriculture is currently the main threat to lizards, climate change is a looming threat, particularly when combined with other types of pressures (Cox et al., 2022). Importantly, the microclimatic patterns described above are likely to be strongly influenced by land-use and hydrological changes, which can alter habitat structure and thermal buffering capacity. Our simulations focused on broad-scale warming scenarios did not explicitly incorporate changes in precipitation regimes, drought intensity, or land-use dynamics. However, these factors may substantially alter microclimatic conditions at local scales. In the Cerrado, vegetation removal associated with habitat conversion can increase thermal exposure and reduce microhabitat buffering, whereas agricultural practices such as irrigation may locally mitigate thermal stress by increasing moisture availability. These contrasting effects highlight that microclimatic conditions experienced by individuals may deviate from macroclimatic predictions, particularly in human-modified landscapes, and should be considered when interpreting species vulnerability to climate change. Overall, vulnerability to climate change is not only species-specific but also strongly mediated by habitat structure and land-use context, particularly in heterogeneous systems such as the Cerrado.

Our results highlight the dual vulnerability of Gymnodactylus amarali to climate change combining large-scale range contractions with fine-scale reductions in active hours. Although rocky microhabitats will continue to buffer temperature variation through rock–air decoupling, this buffering is insufficient to offset the projected decline in functional activity. These findings underscore the importance of integrating ecophysiological and biophysical perspectives into distribution modeling, as models based solely on environmental predictors risk underestimating species’ vulnerability. Maintaining habitat heterogeneity in the Cerrado, particularly rocky outcrops that provide thermal refugia, will be critical for the persistence of G. amarali and other saxicolous species, as these structures can buffer thermal constraints while sustaining foraging opportunities and resource availability. More broadly, our results suggest that nocturnal tropical lizards may be particularly vulnerable to climate change, a group that remains underrepresented in the literature compared to diurnal species. Conserving structural microhabitats can enhance the resilience of nocturnal lizards to climate change and should be prioritized alongside broader strategies to address habitat loss and climate change.