Evaluating bat boxes: design and placement alter bioenergetic costs and overheating risk

We assess bioenergetic costs and overheating risk of five artificial roost designs. For endothermic bats, energy expenditure was lower for solar-exposed locations and our enhanced thermal mass box design. Shaded landscape placements reduced energy expenditure for heterothermic bats. Increasing ventilation, surface reflectance, thermal mass and decreasing solar exposure decreased overheating.

While provisioning energetically beneficial bat boxes could help imperilled species recover from diseases like white-nose syndrome (Webber and Willis, 2018;Wilcox and Willis, 2016), many bat boxes are inadequate at buffering extreme temperatures (e.g. Flaquer et al., 2014;Griffiths et al., 2017;Hoeh et al., 2018;Martin Bideguren et al., 2019;Rueegger, 2019). Additionally, bats avoid some designs and suboptimal landscape placements, which leads to limited conservation benefit Whitaker et al., 2006). Identifying design and placement combinations supporting beneficial microclimates and promoting occupancy could enhance the efficacy of bat boxes as conservation tools by reducing overheating events (Martin Bideguren et al., 2019), promoting energy savings (Wilcox and Willis, 2016), facilitating disease recovery (Fuller et al., 2020) and enhancing pup development (Zahn, 1999).
For bats, finding roosts that suit their thermoregulatory needs is vital to survival and pup development during pregnancy and lactation. Due to high surface area to volume ratios and low body mass, bats often thermoconform to roost temperature (Czenze et al., 2020;Licht and Leitner, 1967a). Warm roosting conditions enhance pup development because less energy is spent on maintaining endothermy (i.e. maintenance of stable core body temperature through metabolic heat production; Zahn, 1999;Lausen and Barclay, 2006). Furthermore, maintaining endothermy in cold roosts can be energetically expensive (Geiser and Brigham, 2000;Willis et al., 2005a). Consequently, many reproductively active individuals will use torpor (i.e. become heterothermic) when roosts are cool (Bergeson et al., 2021;Willis et al., 2006), even though this behaviour involves reproductive costs of delayed parturition (Racey and Swift, 1981), reduced milk production (Wilde et al., 1999) and slowed juvenile development (Hoying and Kunz, 1998).
Roost temperatures occasionally exceed bats' heat tolerance limits (Flaquer et al., 2014;Griffiths, 2021;Griffiths et al., 2017;Martin Bideguren et al., 2019). In the absence of temperature refugia within a roost, bats have limited capacity to combat lethally high temperatures but may respond by shifting location (Licht and Leitner, 1967b;Lourenço and Palmeirim, 2004), increasing evaporative cooling (Czenze et al., 2020) and vasodilation of blood vessels (Reeder and Cowles, 1951) or using facultative hyperthermia (Reher and Dausmann, 2021). Temperatures exceeding 40 • C may induce heat stress, which increases energy costs; prolonged exposure to high temperatures can cause mortality (Licht and Leitner, 1967a). If bats preferentially select roosts prone to temperature extremes and that offer no refugia, the roosts may function as ecological traps (Crawford and O'Keefe, 2021a).
The physiological state of roosting bats may also impact bat box microclimate. For example, colony size and metabolic state (e.g. active or torpid) can influence box temperature and humidity (Bartonička andŘehák, 2007;Pretzlaff et al., 2010). When in a torpid state, a bat's body temperature is often near ambient conditions (Willis et al., 2005a(Willis et al., , 2005b; thus, individual torpid bats likely exert minimal influence on roost temperature. However, a large group of torpid bats adds thermal mass and insulation and, thus, should stabilize roost temperatures (Kurta, 1985). Conversely, socially thermoregulating bats (i.e. a group of normothermic individuals) could substantially increase roost temperature (Pretzlaff et al., 2010), thereby yielding energy savings. For instance, the presence of bats in a roost could raise roost temperatures by 7 • C, allowing group-roosting individuals to conserve ∼ 53% of their daily energy budget (Willis and Brigham, 2007).
While bat boxes are deployed for bats worldwide, a lack of rigorous empirical research on microclimate and the effects of roosting bats has led to limited success and a poor understanding of bat boxes as conservation tools. Working at two study sites, we deployed 480 temperature dataloggers to map roost temperature at varying heights and aspects within 40 rocket-style bat boxes of 5 designs and assessed use by endangered Indiana bats (Myotis sodalis). For bats seeking to maintain endothermy, we expected solar-exposed locations, box designs promoting warmer temperatures, large bat group sizes and warm, calm weather would result in the lowest daily energy expenditure (DEE). For facultatively heterothermic bats, we expected shaded bat box placements, box designs promoting cooler temperatures, small group sizes and cold, windy weather would decrease DEE by reducing roost temperatures and, thus, facilitating deeper torpor. We hypothesized box designs with higher thermal mass, enhanced surface reflectance or enhanced ventilation would reduce overheating events compared to a reference box design. We further expected high solar exposure would increase overheating risk in all box designs.

Study sites
We  Divoll and O'Keefe, 2018), and was bordered by urban housing and warehouse districts. During this study, the Indiana site accumulated 640 mm of total rainfall and outside air temperature (T a ) ranged from 1.3 • C to 34.7 • C. Daily minimum and maximum temperatures averaged 13.5 ± 0.4 • C and 26.7 ± 0.5 • C, respectively. The Kentucky site, located in Scott County, was ∼ 1010 ha and ∼74% forested. The Kentucky site was characterized by mostly forested, rolling hills containing predominantly oak (Quercus spp.), hickory (Carya spp.) and eastern redcedar (Juniperus virginiana; Woods et al., 2002). During the study, the Kentucky site accumulated 804 mm of total rainfall and T a ranged from 2.5 • C to 36.4 • C. Daily minimum and maximum temperatures averaged 15.1 ± 0.4 • C and 27.7 ± 0.4 • C, respectively. Animal care and use protocols were approved by Eastern Kentucky University (no. 01-2019) and Indiana State University (no. 559972).

Rocket box deployment
We constructed eight replicates of five rocket box designs (i.e. 40 rocket boxes total; 20 per site). All design variants were modifications of the reference design described by Tillman et al. (2021). Our designs were reference (REF), vent removal (VR), chimney (CH), white tile roof (WTR) and external water jacket (EJW) (Fig. 1A). Design alterations were intended to promote microclimates different from the REF design while providing equal entrance area, roosting surface area and volume. Compared to REF, the VR design increases minimum temperatures by reducing ventilation, the CH design decreases maximum temperatures by venting rising hot air out a black chimney, the WTR design decreases maximum temperatures by reflecting solar radiation with a white ceramic tile roof and the EJW box buffers both high and low temperatures as a result of a sealed external chamber filled with three 750-ml water packets per box side, which increases thermal mass and, thus, time to heat and cool (Tillman et al., 2021). We constructed boxes from 1.91cm thick pine boards, offering a 1.91-cm chamber spacing, and coated exterior surfaces in two layers of medium brown paint. Seams were sealed with latex caulk to improve weather resistance.
At each site, we deployed four box clusters (hereafter, landscape placements; Fig. 1B and C), placing one of each of the five designs in each landscape placement to facilitate discovery and roost switching and provide bats a variety of microclimates in one locality (Lewis, 1995;Rueegger, 2016). Open solar treatment placements were located > 30 m away from tree lines and boxes received no shading. Forest placements were in a closed-canopy condition so boxes received little to no direct solar exposure. Easterly and westerly sun placements were ∼5 m from east-and west-facing tree lines such that boxes primarily received morning and evening solar exposure, respectively. All boxes were deployed by 1 April 2019, which is earlier than the mean arrival date for Indiana bats at the Indiana site (3 April, Pettit and O'Keefe, 2017). Landscape placements ran along a north-south axis (box vents facing north and south), with box designs randomly ordered and spaced 2 m apart. The top of each rocket box was ∼ 6 m above ground.

Microclimate data collection
For all 40 boxes, we recorded internal air temperature (T roost ) with Thermochron iButtons (Model DS1921G, Maxim Integrated, 0.5 • C increments, ± 1.0 • C accuracy). We placed iButtons at 12 identical positions in each box: top, middle and bottom on all four box faces, and only in the more variable outer chamber (Brittingham and Williams, 2000), as we were fundamentally interested in temperature extremes.
The iButtons were housed in plastic bushings that prevented bats from touching the iButton surface but did not affect temperature recordings and iButtons made no discernable ultrasonic noise when tested (Crawford, 2020). Our iButtons recorded temperature bi-hourly, on even or odd hour intervals to conserve memory space. Even and odd hour iButtons were alternated at each level within the box to ensure that temperature data were collected every hour at every level (i.e. top, middle, and bottom). We removed iButtons at the end of the study following 3 consecutive site visits with no bat detections (in October for both sites).

Roost checks
To survey all 40 rocket boxes for presence/absence and abundance of Indiana bats, we performed spotlight checks and emergence counts 2-4 times per week per site, from April to October 2019 at both field sites. Spotlight checks, typically performed between 16:00-20:00 h (EDT), with sunset occurring around 21:00 h, involved shining a ∼1000 lumen spotlight into each box and visually determining presence/absence of bats (Hoeh et al., 2018;Whitaker et al., 2006). Further, bats were visually counted to estimate abundance and to aid in determining where to conduct emergence counts. We classified bats to genus visually via spotlight checks; the only other bat species detected was the larger big brown bat (Eptesicus fuscus) on < 5% of observations and the presence of Indiana bats was reaffirmed via molecular diagnosis from faeces (Walker et al., 2016) collected from guano traps beneath boxes. For emergence counts, observers arrived at boxes ∼ 30 min before sunset and stayed 10 min after the last bat emerged or 30 min after sunset if no bats emerged (Hoeh et al., 2018). Observers recorded times of first and last emergence and number of bats. Observers typically watched approximately three boxes within a box cluster. Observations of big brown bats were excluded from analyses.

Weather data collection
We collected hourly weather data at each box cluster via Ambient Weather   per site). Each weather station was mounted 3 m above ground and 2 m from the south side of each box cluster to prevent shading. Weather stations recorded temperature ( • C), solar radiation (W/m 2 ), rainfall (mm) and wind speed (m/s). Because recordings were hourly, we recognize that we may have missed some extreme weather observations. We discarded microclimate and bat count data for days when weather stations suffered power failures.

Bioenergetic models
We used a bioenergetic modelling approach (following Humphries et al., 2002Humphries et al., , 2005Wilcox and Willis, 2016), modified to assess the DEE of an Indiana bat that we assumed to use an optimal roosting position in each bat box over the study duration. By calculating DEE, we directly compare estimated energy expenditure of bats in each box design under different environmental conditions, while simultaneously accounting for the variety of temperatures available to exploit.
Because Indiana bats' thermoregulatory behaviour can range from near perfect endothermy to extreme heterothermy (Bergeson et al., 2021), we decided to model energy expenditures of two distinct thermoregulatory scenarioscontinuous endothermy and facultative heterothermy. As detailed metabolic data are unavailable for the Indiana bat, we used data for the well-studied, morphologically similar little brown bat (Myotis lucifugus). Using parameters listed in Table 1 and bioenergetic equations presented in Table 2, we calculated the DEE for a behaviourally thermoregulating bat occupying each bat box. For each scenario, we calculated mass-specific metabolic rate (mlO 2 g −1 hr −1 ) for a bat present in each bat box for a 24-h period. We then converted mass-specific metabolic rate to whole animal metabolic rate by multiplying by the mean body mass of reproductive female little brown bats (8.44 g; Kurta and Kunz, 1988). Then we converted hourly whole animal metabolic rates to energy expenditure in joules using a conversion factor of 1-ml O 2 = 20.083 J (Schmidt-Nielsen, 1997). We summed the energy expenditure for each bat box for each day and converted joules to kilojoules for modelling.

Criteria Formula
When T roost ≥ T lc and T roost ≤ T uc BMR When T roost < T lc and T roost > T onset BMR + (T lc -T roost ) * C eu When T roost ≤ 25 • C and T roost > T tor-min TMR min * Q (Troost-T tor−min )/10 10 Equations are based on those presented in Humphries et al. (2002Humphries et al. ( , 2005. Parameters defined in text; values given in Table 1. Our first scenario modelled DEE for a continuously endothermic bat (DEE endothermic ), such as a reproductively active female attempting to maintain high body temperature to facilitate pup development (Lausen and Barclay, 2006;Zahn, 1999). We assumed hourly roost temperatures within the thermoneutral zone result in metabolic costs equal to the basal metabolic rate (BMR; see Tables 1 and 2). When T roost was less than the lower critical temperature (T lc ), we assumed the energy expenditure was BMR plus the product of the temperature difference (between the critical temperature and roost temperature) and euthermic thermal conductance (C eu ). Above the upper critical temperature (T uc ), we assumed energy expenditure was BMR plus the product of the temperature difference and C eu , which increases linearly until lethal temperatures are reached. We note that some tropical bats may use hyperthermic torpor during heat stress (Reher and Dausmann, 2021), but this behaviour is currently documented for only a single bat species. Because we did not document any mortality during our study period (2019), we assumed bats could move to avoid experiencing lethal temperatures (i.e. ≥ 45 • C).
Using the same microclimate data presented in scenario one, but applying a different set of equations, our second scenario modelled the DEE of a facultatively heterothermic bat selecting the temperature that minimizes energy expen- diture each hour of the day (DEE heterothermic ). This scenario assumes a reproductive or non-reproductive bat will attempt to maintain an endothermic body temperature until T roost drops to the torpor onset value (T onset ) when it will then enter torpor (see Tables 1 and 2). We chose T onset = 25 • C as several bat species readily enter torpor near this temperature (Geiser and Brigham, 2000;Cryan and Wolf, 2003;Willis et al., 2005a;Geiser et al., 2011). This scenario assumes bats will increase energy expenditure and attempt to maintain a warm body temperature when T roost is below T lc and above T onset but once energy costs become too high (i.e. T roost ≤ 25 • C) bats will conserve energy through torpor. We concede that this cut off may over or underestimate energy expenditure based on the individual bats' behaviour and physiological condition but is still a useful proxy for inference. While in torpor, metabolic rate is equal to the product of the minimum torpid metabolic rate (TMR min ) and the temperaturedependent change in metabolic rate (Q 10 ). If T roost reaches the minimum torpid set point (T tor-min ) of 2 • C (noting the lowest temperature recorded in a box in this study was −7.5 • C), the bat will increase energy expenditure to defend a minimum body temperature as a function of the temperature difference and torpid thermal conductance (C t ), plus TMR min . Following Willis and Brigham (2007), we did not calculate costs of entry into and arousal from torpor, as the cost, duration and frequency of these events can be highly variable. After calculating DEE values, we paired these data with available bat counts and weather data (maximum daily temperature [ • C] and windspeed [m/s]), yielding 2530 DEE estimates to analyse for each scenario.
Our DEE heterothermic scenario assumes a bat moves among positions in the box, always choosing the position that minimizes DEE, even while torpid; however, we acknowledge that this is unlikely to happen in reality. Conservatively, we chose to model DEE using only roost positions with known temperature data. Because iButtons were recording every 2 h, this necessitated using different roosting positions in our calculations for hourly DEE. An alternate approach could assume the bat chooses the best position within the roost for torpor at the start of each day (assuming temperatures ≤ 25 • C are available) and stays there until T roost > T onset , thus allowing movement. One problem with this approach is that we would have to estimate T roost for alternating hours when iButtons were not recording. Further, in our estimation, we could not account for the elevated temperatures at iButton positions within close proximity of groups or individual free-ranging bats that move non-randomly within roosts during the day. Our modelling approach for a facultative heterothermy scenario allows for a bat to move to the best roost location available for torpor and, thus, reduces this bias and models a clearer effect of box design itself on DEE.

Overheating risk
In addition to measuring the metabolic costs associated with each box design, we investigated overheating risk. Here, we deemed temperatures > 40 • C to be unsuitably hot, as bats avoid these high temperatures (Licht and Leitner, 1967b;Lourenço and Palmeirim, 2004), and prolonged exposure can result in heat stress and mortality (Alcalde et al., 2017;Flaquer et al., 2014;Griffiths, 2021). For each bat box for each day, we summed the total number of hourly overheating events (any hourly recording > 40 • C) from the 144 recordings per box on each 24-h day. Because forest placements logged few overheating events (n = 25), models incorporating this landscape placement failed to converge; therefore, we removed forest placements from this analysis. We paired daily counts of overheating events with observations of bat abundance and weather, resulting in 1932 daily observations.

Analysis
To model DEE under two scenarios and the daily count of overheating events, we followed an information theoretic approach (Burnham and Anderson, 2002), constructing an a priori candidate set of 14 models (Table 3). Models comprised combinations of box design, landscape placement, total number of bats, maximum daily T a , and maximum daily wind speed. We used linear models (LMs), using qq-plots and histograms of residuals to assess normality of DEE. We log transformed DEE heterothermic to achieve normality. We used generalized LMs fit to a Poisson distribution to assess the daily count of overheating events. We checked models for overdispersion and goodness of fit via overdispersion parameters and R 2 values. For all analyses, we checked for multicollinearity among predictors via variance inflation factor (VIF) tests; all predictors had VIFs < 2. The REF design and open landscape placements were used as reference levels during modelling, as all box designs are variants of the REF and high solar exposure is important for maternity colony formation. Although our dataset contained repeated measures within each of the 40 boxes, we did not include boxID as a random effect in models. We reason that under identical conditions each box design should function the same and, thus, yield the same results. As such, the differences between identical boxes in different landscape positions are due to environmental effects, not the boxes themselves.
All modelling was conducted in R version 3.6.2 (R Core Team, 2019). We ranked models via AIC C (Akaike's Information Criterion corrected for small sample sizes) using the package 'bbmle' (Bolker, 2020) to identify the top overall model(s). We considered models to be competing if AIC C was ≤ 2 from the top model. If competing models were present, we constructed a 90% confidence set for model averaging (Burnham and Anderson, 2002) using the package 'MuMIn' (Barton, 2020). To reduce the possibility of excluding biologically important parameters that were in top model(s), we identified informative parameters as those with 85% confidence intervals not overlapping 0 (Arnold, 2010). Means are presented asx ± SE unless otherwise stated. Predicted means were obtained through the package 'emmeans' (Lenth, 2020).

Continuous endothermy scenario
For the continuous endothermy scenario, Model 10 was the only competitive model (w i = 0.99, R 2 = 0.86; Supplementary Material, Table S1). We identified 10 informative parameters related to box design, landscape placement, bat abundance, and weather (Supplementary Material, Table S2). DEE endothermic ranged from 10.8 to 37.5 kJ (mean = 19.1 ± 0.09 kJ) and showed substantial variability over time (Fig. 2). DEE endothermic was most variable during pregnancy, with no patterns in differences between landscape placements. In late May, when temperatures warmed and leaf-out occurred in the forests, it was easier to see differences in DEE endothermic by box design ( Fig. 2A) and landscape placement (Fig. 2B) Predictably, increasing numbers of bats decreased DEE endothermic for all box designs (Fig. 3). There was a strong decreasing trend in DEE endothermic with increasing bats for the EJW design, but the low maximum emergence count of 59 bats limited our ability to assess the effect of bats for this design. For a group of 50 bats, DEE endothermic was 7.9% lower in the EJW design and 3.9% lower in the WTR than the REF design (Supplementary Material, Table S5). Comparing a roost occupied by 150 bats to an unoccupied roost, predicted DEE endothermic was 29.4% (5 kJ) lower in the WTR design, whereas adding 150 bats to the REF design decreased DEE endothermic by only 12.7% (2.3 kJ). This is due to the WTR design reducing instances of T roost > T uc , thus lowering cooling costs at large group sizes.
Increasing maximum daily T a substantially decreased DEE endothermic regardless of placement, and effect sizes were similar for all placements. For instance, for a change in maximum daily T a from 10 • C to 30 • C, DEE endothermic decreased by 64.4% (17.1 kJ) for forest placements and decreased by 69.5% (16.6 kJ) for open placements. In contrast, DEE endothermic increased substantially as maximum daily wind speed increased, most noticeably in forest placements where T a and solar radiation were lowest. For an increase in maximum daily wind speed from 0 to 7 m/s, DEE endothermic increased by 21.9% (4.7 kJ) at forest placements and by 16.6% (3.1 kJ) at open placements.

Facultative heterothermy scenario
Model 14 was the top overall model for the facultative heterothermy scenario (w i = 0.68, R 2 = 0.91; Supplementary Material, Table S6). Additionally, model 10 was competitive, with AIC C = 1.5, w i = 0.32, and R 2 = 0.91. From the 90% confidence set containing models 14 and 10, we identified 10 informative parameters related to box design, landscape  Table S7). DEE heterothermic ranged from 0.2 to 10.4 kJ (mean = 4.1 ± 0.05 kJ), and variability in DEE heterothermic was highest during lactation and post-lactation (Fig. 4). Box design alone did not have a substantial impact on DEE heterothermic , which was ∼ 4 kJ for all box designs (Supplementary Material, Table S8, Fig 4A), but landscape placement was an important predictor. Compared to the open placements, forest placements decreased DEE heterothermic by 6.9% (Supplementary Material, Table S9, Fig. 4B).
Regardless of box design, increasing numbers of bats increased DEE heterothermic , though effects were weak (Supplementary Material, Table S7). Increasing daily maximum T a had a stronger effect by increasing DEE heterothermic in forest placements vs. open placements (Supplementary Material, Table S7). Further, increasing maximum daily wind speed resulted in substantial decreases in DEE heterothermic for all placements. This effect was most prominent for forest placements which had low solar radiation. For instance, for a change in maximum daily wind speed from 0 to 7 m/s, DEE heterothermic decreased in forest placements by 19.7% (0.5 kJ), whereas for open placements DEE heterothermic decreased by 15.9% (0.5 kJ).

Overheating risk
In total, boxes recorded 9171 overheating events (i.e. temperatures > 40 • C) out of 945 060 total temperature recordings. Most overheating events occurred in the VR and REF designs in solar-exposed locations (66%; Fig. 5A). Further, roosting  From this model, we identified 18 informative parameters related to box design, landscape position, weather and bats (Supplementary Material, Table S11). Design was important; compared to the REF design, the VR design was similar, while the EJW, CH and WTR designs logged substantially fewer daily overheating events ( Fig. 6A and Supplementary Material, Table S12). Boxes in easterly and westerly sun placements logged considerably more daily overheating events than boxes in open placements (east: 3.1 ± 0.2, range = 0-27; west: 2.4 ± 0.2, range = 0-22; open: 1.6 ± 0.1, range = 0-22; forest: 0.01 ± 0.01, range = 0-2). Boxes in the open cluster were unused in 2019; overheating should be less likely in the absence of many warm bat bodies. Overheating counts for a given box design varied across placements. Regardless of box design, higher numbers of bats increased overheating risk. This effect was strongest in the VR design (Fig. 6B). Increasing maximum daily temperature increased overheating risk for all three landscape placements tested (Supplementary Material, Table S11); however, easterly and westerly sun placements did not respond as strongly as open placements. On days when T a was ≥30 • C, bat boxes logged an average of 5.5 ± 0.  Table S11).

Discussion Overview
Our study jointly quantifies the energy benefits and overheating risk of different bat box designs and placements, as well as providing a robust method for assessing the conservation value of bat boxes from two physiological perspectives. We found landscape placement was the most influential, controllable factor altering energy expenditure and overheating risk, but box design, bat group size, and weather were also important factors. For reproductively active bats maintaining endothermy, solar-exposed landscape placements resulted in the lowest DEE endothermic, and the EJW box (with greater thermal mass) promoted further reductions in DEE endothermic . For facultatively heterothermic bats, forest placements resulted in the lowest DEE heterothermic and virtually eliminated overheating events. At the same time, our study affirms overheating risk is higher in solar-exposed locations and risk is further increased by box designs with low surface reflectance, low thermal mass or poor ventilation (Griffiths et al., 2017;Martin Bideguren et al., 2019;Rowland et al., 2017;Tillman et al., 2021). Three box designs (CH, EJW and WTR) reduced overheating risk compared to the REF design, but only one (EJW) simultaneously decreased endothermic energy expenditure. These observations suggest there need not be a tradeoff in reducing box overheating risk and providing a microclimate suitable for reproductive bats.

Landscape
Landscape placement is a key consideration for practitioners when choosing to deploy bat boxes. For instance, both high solar exposure and aspects facing the sun are qualities of bat boxes facilitating maternity colony formation (Brittingham and Williams, 2000;Kerth et al., 2001) and can increase bat box uptake (Whitaker et al., 2006). For a continuously endothermic individual, easterly sun, westerly sun and open sun placements yielded the lowest DEE. This validates the assertion that female bats choose solar-exposed roosts to reduce the energy costs of normothermia (Callahan et al., 1997;Vonhof and Barclay, 1996), thus facilitating the avoidance of torpor use and the negative effects of low body temperature on reproduction (e.g. Hoying and Kunz, 1998;Wilde et al., 1999; but see Bergeson et al., 2021). In contrast, forest box placements minimized DEE for heterothermic bats. Less solar exposure of this landscape placement likely tempered warm roosting positions, thus facilitating deeper torpor bouts and, hence, greater energy savings. While reproductive bats gain both energetic and reproductive benefits from solar-exposed roosts, our study and others indicate high solar exposure and orientations facing the sun increases overheating risk (Griffiths et al., 2017;Martin Bideguren et al., 2019). The effects of landscape placement can be mitigated (to some extent) with careful box design selection; however, overheating risk must be minimized while still providing warm microclimates to facilitate pup growth (Crawford and O'Keefe, 2021a).

Box design
While past bat box bioenergetic research has focused on the impacts of artificial roost colour and heating on energy expenditure (e.g. Doty et al., 2016;Wilcox and Willis, 2016), few studies have specifically assessed overheating risk posed by varying box designs (but see, Flaquer et al., 2014;Griffiths et al., 2017;Martin Bideguren et al., 2019). We expand on prior work by interactively investigating the effects of a box's thermal mass, ventilation, and enhanced surface reflectance on DEE and overheating risk.  (Larson et al., 2018). Increasing bat box thermal mass should decrease metabolic costs for normothermic endotherms. In our study, boxes that decreased overheating events through higher surface reflectance (WTR) or enhanced ventilation (CH) also increased DEE endothermic . Thus, there is a tradeoff between heat stress risk and energetic benefits in some instances.
The number of bats occupying a bat box can alter temperature and energetic savings (Pretzlaff et al., 2010;Willis and Brigham, 2007), but this effect varies with bat box design. For example, with increasing bat group size, the WTR and EJW designs had stronger decreasing trends in DEE endothermic compared to REF. Similarly, artificial boxes insulated with polystyrene and occupied by individual great tits (Parus major) were significantly warmer than T a , whereas occupation by a single bird did not significantly elevate box temperature in uninsulated nest boxes (Veľký et al., 2010). Our work also highlights the value of altering thermal mass and thermal conductance in artificial roosts to reduce overheating risk and potentially enhance social thermoregulation at large bat group sizes.
Contrary to our prediction, box design was not an important factor for DEE heterothermic . Rocket boxes support vertical temperature gradients up to 10 • C (Hoeh et al., 2018), offer multiple aspect options (i.e. north, south, east and west faces), and ∼1 m of vertical space; these bat box traits should allow for bats to find temperatures suitable for torpor use in all our roost designs, assuming colony size does not hinder movement. We found increasing bat group size increased overall energy expenditure because of the concomitant increase in T roost and because metabolic rate increases with temperature for torpid bats (Geiser and Brigham, 2000;Willis et al., 2005a). Only in the CH design did we see that increasing group size did not impact energy expenditure, likely due to excess body heat generated by bats being vented out the chimney and, thus, keeping temperatures cooler for torpid bats.

Conservation implications
While our data are specific to bats, the insights derived from our analysis are broadly applicable to a variety of nest-box using mammals. Our work highlights several key considerations that may improve the success of bat boxes when deployed for the conservation of imperilled bat species. Landscape placement in conjunction with bat box design and colour are key aspects of mitigating overheating risk to bats and creating conditions conducive to endothermy. Deploying a design with inadequate features (e.g. poor ventilation, low surface reflectance, small size) in a solar-exposed location is likely to increase the risk of heat stress to roosting bats. When deploying bat boxes for maternal bat populations, practitioners should consider the use of box designs that increase thermal mass (EJW in this study) to buffer against overheating events while simultaneously decreasing DEE endothermic . We recommend further experiments altering thermal mass, thermal conductance and surface reflectance to improve upon current bat box designs, as these modifications will be critical to buffering the immediate effects of overheating and longterm effects of a warming climate (Larson et al., 2018).
Promoting warmer microclimates in roosts during the spring and summer could be critical to improve survival rates of bats impacted by diseases, like white-nose syndrome (Wilcox and Willis, 2016). For example, little brown bats maintain higher body temperatures when rapidly healing from white-nose syndrome during the spring (Fuller et al., 2020). Supplying afflicted bats with roosts that reduce the overall costs of maintaining higher body temperatures could promote recovery. Further, increasing the energy savings of bats during the summer by catering to both heterothermic and endothermic bats could enhance overwinter survival. For instance, little brown bats going into hibernation with larger fat stores have a higher probability of surviving hibernation (Cheng et al., 2018). Clearly, additional work is needed to improve upon current bat box designs to increase energy savings while simultaneously reducing overheating events (Crawford and O'Keefe, 2021a). While no mortality was documented during the study period (2019), we observed mortality (presumably from overheating) at boxes in both study sites during the summer of 2020 (11 bats in Indiana and 4 bats in Kentucky) and in Kentucky in 2021 (2 bats). We urge practitioners to recognize the importance of bat box design, material, colour and placement, and to carefully study microclimates of novel artificial roosts before provisioning artificial roosts to bats (Crawford and O'Keefe, 2021b).
While artificial roost microclimates are relatively easy to study and manipulate, we lack data on how artificial roost microclimates compare to natural tree roost microclimates. Many tree roosts (e.g. hollows, cavities and bark) buffer against temperature extremes, such that they are cooler than outside air temperatures during the day and warmer at night (Lacki et al., 2013;O'Connell and Keppel, 2016;Rowland et al., 2017;Sedgeley, 2001). Most solar-exposed artificial roosts exceed ambient temperature during the day (Griffiths et al., 2017;Martin Bideguren et al., 2019;Tillman et al., 2021, this study), sometimes by > 25 • C (Hoeh et al., 2018), and fall to ambient temperature within a few hours of sunset (Kerth et al., 2001;Lourenço and Palmeirim, 2004;Tillmanet al., 2021). We recommend future work to compare the microclimates of both natural and artificial roosts from both physiological and ecological perspectives to improve our understanding of artificial roost use.