On the measurement of microclimate

Many organisms live in environments in which temperatures differ substantially from those measured by standard weather stations. The last decade has witnessed a paradigm shift in efforts to quantify these differences and to understand their ecological, functional and evolutionary implications. This renewed interest in microclimate ecology has been accompanied by the development of various compact temperature sensors and radiation shields. However, it is clear that there are many pitfalls when measuring temperature using these devices. Here we address the problem of measuring temperatures in these microenvironments accurately. We first discuss the theory of measuring surface, ground and air temperatures with reference to energy fluxes and how these are modified by material, reflective properties and size of the device. We highlight the particular difficulties associated with measuring air temperature. We then report on the results of a series of experiments in which air temperatures recorded by various commonly used microclimate temperature loggers are compared to those obtained using research‐grade instruments and synoptic weather stations. While accurate measurements of surface and ground temperatures and air temperatures at night and in shaded environments can be relatively easily obtained, we show substantial errors are to be expected when measuring air temperatures in environments exposed to sunlight. Most standard sensors yield large errors, which can reach 25°C due to radiative fluxes operating on the thermometer. This problem cannot be wholly overcome by shielding the thermometer from sunlight, as the shield itself will influence both the temperatures being measured and the accuracy of measurement. We demonstrate that reasonably accurate estimates of air temperature can be obtained with low‐cost and unshielded ultrafine‐wire thermocouples that possess low thermal emissivity and a highly reflective surface. As the processes that create microclimatic temperature variation are the same as those that cause errors, other logger types should be used with care, and generally avoided in environments exposed to sunlight and close to the ground where wind speeds are lower. We urge researchers interested in microclimates and their effects to pay greater heed to the physics of heat exchange when attempting to measure microclimate temperatures and to understand the trade‐offs that exist in doing so.


| INTRODUC TI ON
Temperature influences every aspect of the physical environment within which terrestrial, freshwater and marine organisms reside. It sets limits on the survival, reproduction and behaviour of organisms and governs the rates of biological processes within these limits (Clarke, 2017). The increasing availability of global gridded climate data-for example, ERA5 (Copernicus Climate Change Service, 2020), WorldClim (Fick & Hijmans, 2017), CHELSA (Karger et al., 2017) and Terraclimate (Abatzoglou et al., 2018)-interpolated from weather stations has greatly facilitated macroecological research on links between organisms and climate. However, many organisms live in environments with temperatures that differ substantially from those of weather stations (Suggitt et al., 2011), as close to the ground or is thus key to understanding how organisms interact with their environment, and is increasingly recognised as necessary for addressing applied challenges such as predicting the ecological consequences of climate change (Potter et al., 2013;Zellweger et al., 2020).
Growing recognition of the importance of this discrepancy has led to a paradigm shift towards microclimate ecology and biogeography (Lembrechts & Lenoir, 2020). Yet, many ecologists do not seem fully aware of the pitfalls associated with measuring microclimate.
Let us first consider the measurement of air temperature by a weather station. In 1954, the World Meteorological Organisation published the first edition of the 'Guide to Meteorological Instruments and Methods of Observation', which sets out standardised procedures for measuring air temperatures (WMO, 1954).
Since radiation from the sun, clouds, the ground and other surrounding objects passes through air without appreciably changing its temperature, but a thermometer exposed freely in the open can absorb considerable radiation, it is thus deemed necessary to protect the thermometer from radiation by a screen or shield. Without doing so, temperature differences between the air and a thermometer may reach 25°C (WMO, 1954). It is recommended that the size and construction of the screen is such that it allows ample space between the thermometer and the walls of the screen and that direct contact between the sensing elements and thermometer mounting is avoided to prevent conductive heat transfer. The screen itself is painted white or made of reflective material, and artificially ventilated and/or, more commonly, louvred to permit natural ventilation, thereby ensuring that convective heat exchange between the thermometer and the air inside the screen, and between the air inside the screen and that outside it, is maximised. It is also recommended that air temperature should be representative of the free air conditions surrounding the station over as large an area as possible. As such, temperatures are recorded at a height of between 1.2 and 2.0 m above-ground level in locations that are freely exposed to wind and unobstructed by nearby vertical objects in the landscape such as trees, buildings and surrounding terrain. In other words, microclimatic 'noise' is deliberately minimised. Yet, what is considered 'noise' by climatologists matters for biologists interested in biotic responses to climate.
Let us now consider air temperature close to the ground or vegetation. Just above the ground close to other opaque surfaces such as rocks, soil and leaves, conductive and convective heat transfer leads to significant fine-scale variation in air temperature, because reduced airflow maintains strong vertical and horizontal gradients in temperature (Geiger, 1927;Monin & Obukhov, 1954;Richardson, 1922). If the intention is to measure temperature in environments exposed to radiation, then the issue of radiation absorption arises: when exposed to solar radiation, the temperature of an unshielded thermometer will be influenced by these radiative fluxes.
The issue of radiation fluxes operating on temperature loggers has prompted many ecologists to deploy radiation shields (Table 1).
However, this can be problematic for two reasons. Firstly, whereas at the height of a standard weather station, airflow generally ensures that the temperatures underneath a shield are similar to those of its surroundings, this is not the case when microclimatic variation exists. Here, the temperature variation owes its existence to low wind speed (Geiger, 1927;Prandl, 1953) and a shield will alter the temperature through shading and reduced wind speed. Consequently, the temperature being measured ceases to be representative of that in the absence of a shield. Mechanical ventilation through the use of an aspirator is also not a solution. Artificially increasing the airflow alters the convective heat exchange processes that are ultimately responsible for microclimatic variation (Prandl, 1953), and thus alters the temperature of the air itself. Secondly, whereas a standard weather station is large enough to ensure that convective heat transfer between the shield and thermometer is negligible, the measurement of microclimate temperatures has often involved the We urge researchers interested in microclimates and their effects to pay greater heed to the physics of heat exchange when attempting to measure microclimate temperatures and to understand the trade-offs that exist in doing so.

K E Y W O R D S
air temperature, climate change, ecology, ground surface temperature, microhabitat, microrefugia, soil temperature, thermocouple TA B L E 1 Examples of the variety of devices and shielding methods used to measure microclimate temperatures (*not specified). The full reference for in- Appendix S1): where R abs and R em are absorbed and emitted radiation (W/m 2 ), respectively, and k Δz is the conductivity (W m −1 °C −1 ) over distance (m).
Assuming our purpose is to measure temperature as closely as possible, an accurate device will thus have high thermal conductivity and minimise the effects of the absorbed and emitted radiation. Let us now consider each of these terms in detail.

| Thermal conductance
Heat transfer is usually measured in units of W/m 2 . When two objects are in direct contact, heat is transferred by conduction-a process in which thermal energy is transferred by the collisions of molecules to propagate energy from hot to cooler mediums-just like when walking bare foot on a hot sandy beach. This form of heat transfer is relevant to consider when determining, for example, the exchange of heat between a thermometer and a leaf or rock in direct physical contact with the thermometer or when considering how a thermometer might be influenced when physically in contact with a radiation shield. Here the heat transfer (W/m 2 ) is the product of the conductivity (k, in W m −1 °C −1 ) between the surface and heat-sensing element of a thermometer and the temperature gradient (°C/m). It is thus influenced by both the distance over which heat must travel and the thermal conductivity of the substance through which the heat travels. Copper, for example, has a higher conductivity than plastic. Conversely, it can be seen that the error in measurement (the difference between the temperature of the surface and that of the thermometer) is thus the heat transfer to the thermometer in form of radiation divided by the product of the conductivity and the distance through which heat must travel. Strictly speaking it is also necessary to consider the surface area in contact, as this scales the rate of heat transfer per unit area to the overall rate of heat transfer.
In practical terms, however, any gains from using a larger thermometer in terms of increased heat exchange between the thermometer and the surface are counteracted by the increases in radiative energy received. Irrespective of the surface area of the thermometer, since it is usually possible to maintain a very small distance between the heat-sensing element of the thermometer and the surface being measured, the overall conductivity per unit distance is very high and the errors caused by radiative fluxes are minimal.
When measuring the temperature of soil below the surface, no radiative heat is supplied to a thermometer and the errors in measurement are likely to be negligible. Here, the primary consideration is the any waterproof casing surrounding the thermometer, which may impede the conductance of heat and thus decrease the rate at which a thermometer's temperature attains equilibrium with that of the soil. Nevertheless, except near the soil surface, rates of change in temperature are relatively slow (Campbell, 1985). In consequence, even when housed in relatively solid casing made of a material with low conductivity, the temperature of a thermometer will generally attain equilibrium with that of the soil. However, weather proof casing surrounding a thermometer will affect its ability to accurately determine surface temperatures above-ground. Here, conductance between the surface and thermometer is imbedded, but the casing still receives radiative heat and transfers this heat to the thermometer itself.
For a thermometer suspended in a fluid such as air, however, the predominant heat transfer mechanism is by convection. This involves conduction between a substance and the fluid, simultaneously accompanied by transport of heat to or from the fluid.
Equation (1) can still be applied, but since the temperature gradient at the surface is maintained by the velocity of the fluid, conductivity must be appropriately defined. Here, the overall rate of heat transfer is defined by the Fick's Law and is the product of the volumetric specific heat of the fluid (J m −3 °C −1 ), its conductance (K expressed in m/s-see Appendix S1 for an explanation of the different units of measurement used) and the temperature difference between the fluid and the thermometer. Conversely, therefore, the error in measurement is thus the net radiative heat transfer to the thermometer divided by the product of the conductance and its volumetric specific heat. In contrast to the situation in which a thermometer is in direct surface contact with the substance, the conductance is not so high, and the radiative fluxes become important (Campbell & Norman, 2012). This is true in both air and water. Though in water, a significant portion of the radiation is attenuated, and the volumetric specific heat of the fluid is higher, overall conductive heat transfer is only c. 20% as efficient in water as in air owing to the much lower thermal diffusivity and kinematic viscosity of water (Appendix S2).
The Conduction under forced convection is generally greater than under free convection, and also increases with the strength of the wind (Appendix S2). Thus, close to the ground, where wind flow tends to be much lower, the influence of absorbed and emitted radiation on the temperature of a thermometer will be greater as the thermometer is less able to exchange heat with the air. Conductance also decreases as the size of the thermometer increases (Appendix S2), as there is more potential for airflow along the object to develop into orderly laminar layers. Thus, size matters and only very small thermometers would be expected to provide accurate temperature measurements in areas with low wind speed ( Figure S1a).
In turbulent flow, rapidly fluctuating eddies (i.e. small whirlpools or vortices) transport heat, as occurs when the layered movement of fluid particles breaks down. This is relevant when considering heat transported through louvered radiation shields in open areas, for example, where the air is naturally turbulent. It thus dictates the extent to which the temperature of the air underneath the shield is similar to that away from the shield and, as with laminar flow, increases with wind speed. The equations that govern turbulent flow also determine the wind profile above-ground, which typically increases logarithmically with height. Since turbulent conductance also increases with wind speed, close to the ground a thermometer will be influenced more strongly by radiation emitted by the shield ( Figure S1b).

| Radiation
Radiation is generated by the thermal motion of particles in matter and no intervening medium is required for heat transfer. It is the underlying reason that one feels warmer in sunshine-here one's body is absorbing solar radiation. Any radiation received by an opaque object is then either absorbed or reflected, the latter depending on the wavelength-specific reflectance of the surface. Materials such as white plastic, polished steel or aluminium typically have a shortwave reflectivity of 75%-90%, whereas darker surfaces on average absorb more than 90% of shortwave radiation (Tarara, 2000), and hence reflect only 10%. All objects also emit radiation as a function of their absolute temperature to the power of 4. An absorption of radiation causes an object to heat up and thus emits more radiation.
The radiation received by a thermometer has three sources. The first is radiation from the sun, which can reach the surface of a thermometer either directly, or in the form of diffuse radiation, which is scattered by particles and clouds in the atmosphere. Direct radiation received by a thermometer depends on the angle of the surface relative to perpendicular. Thus, close to solar noon, the radiation absorbed by a horizontal thermometer will be greater. Diffuse radiation depends instead on the fraction of the hemisphere in view (Campbell & Norman, 2012). Thus, even on a cloudy day, the radiation absorbed by the thermometer will be greater in unshaded environments. The second source is solar radiation reflected from surrounding surfaces, which in turn depends on the reflectance or albedo of those surfaces (the reflectance of objects is wavelength specific, and albedo is the average reflectance of radiation in the shortwave spectrum). For example, ice and snow have a high albedo and reflect far more radiation than rock, bare soil and asphalt (Hay, 1993). The final source is longwave radiation emitted from surrounding surfaces such as vegetation, soil and the sky. This in turn depends on the temperatures of those surfaces, the proportion of each surface in view. A radiation shield will also emit longwave radiation, some of which is received by the thermometer even when sufficient distance is maintained so as to limit convective heat transfer.
Emitted radiation, in addition to temperature, depends on the emissivity of the object. Emissivity is one minus its reflectivity, so surfaces with low emissivity at a given wavelength have high reflectivity at that wavelength and vice versa, and since emitted radiation by passively heated objects is in the longwave spectrum, it is reflectivity and emissivity in the longwave spectrum that is relevant to consider. Whereas metals also have relatively low emissivity (and high reflectivity) of longwave radiation, the converse is true of plastics (Tarara, 2000). An ideal temperature sensor should therefore have a surface coating of polished metal.

| Overview of experiments
Since the measurement of air temperature is most problematic, three sets of experiments were conducted to determine the accuracy of so doing. Our intention was to test a range of different types of temperature loggers and shields used commonly in ecological research (Table 1; Figures S2-S4). The experiments were designed to complement one another, each testing different facets of microclimate air temperature measurement. Experiment 1, conducted between 7 April and 10 July 2020 over several short intervals (Table S1)

| Temperature loggers and shields tested
In Experiment 1, measurements obtained using a research-grade ultrafine-wire thermocouple were compared with those obtained using consumer-grade ultrafine-wire thermocouples, standard

| Experimental set-up
In Experiment 1, air temperatures were measured 10 cm above a short grass lawn. Apart from the TMS4 loggers, each sensor was attached to a thin garden stake, and suspended c. 10 cm above the grass. This was achieved by counterweighting the stake on the surface of a concrete block located c. 10 cm away from the measurement area. TMS4 loggers were positioned in the ground c. 1 m away from other loggers, and inserted into the ground partially so that the above-ground sensor, used in this experiment, was also 10 cm above-ground. Research-grade equipment was programmed to obtain 20 temperature readings per second for 30 s at 10-min intervals. Consumer-grade thermocouples were programmed to record temperatures at 5-s intervals and the iButton thermochrons and TMS4 dataloggers to record temperatures at 1-min intervals. The number of devices of each type deployed on each occasion is shown in Table S1. To provide a proxy estimate of the effect size being measured, namely differences from macroclimate, we sourced 25km grid resolution hourly ambient air temperature data for the same location and time periods from ERA5 (Copernicus Climate Change Service, 2020) and compared these temperatures to those obtained using the research-grade ultrafine-wire thermocouple.
In Experiment 2, the set-up was duplicated in two vegetation

| Experiment 1
During the periods of bright sunshine, both research-and consumergrade ultrafine-wire thermocouples detected large fluctuations in temperature caused by eddy turbulence ( Figure S5). When averaged over hourly periods, however, only the consumer-grade ultrafinewire thermocouple gave estimates of hourly temperatures comparable to the research-grade thermocouple, with a root-mean-square (RMS) error of 0.93°C. All other devices, irrespective of them being shielded or not, resulted in measurements that in general differed from those obtained using the research-grade thermocouple by an amount that exceeded our proxy of the effect size being measured (Table 3; Figure S6). Both shielded and unshielded iButton thermochrons yielded substantial differences from the research-grade thermocouple, with the difference of unshielded iButtons on occasion exceeding 15°C. More accurate readings were obtained by shielding iButtons, but even when shielded, the RMS error was never lower than 3.16°C. Shielding had little effect on the accuracy of the TMS4 dataloggers, which in both cases gave measurements closer to those obtained by the research-grade thermocouple than iButton thermochrons. Nevertheless, the overestimation of temperatures of ~9°C was recorded by both shielded and unshielded loggers, though the RMS error both when shielded and unshielded was lower-2.7°C.
Variation in temperatures measured by each device over a typical 24-hr period (12 July 2020 GMT) is shown in Figure 1. Full results are shown in Table 3. Errors were generally larger during the day than at night, with RMS errors of the latter for all logger types, around or below 1.5°C.

| Experiment 2
At the grassland site, the accuracy of hourly temperature measurements obtained using the TMS dataloggers (relative to measurements obtained using consumer-grade ultrafine-wire thermocouples) was considerably greater than that of iButton thermochrons at all heights, with the greatest accuracy achieved by the shielded TMS4 data logger (Figure 2; Table 4). The iButton thermochrons consistently overestimated temperatures during the day, particularly when unshielded and coated, with errors reaching 25.96 and 18.50°C respectively. The iButtons housed in PVC tubes generally performed better than unshielded iButtons, though temperatures were consistently overestimated both during the day and at night. Both shielded TA B L E 3 Experiment 1. Root-mean-square (RMS) and maximum error of aggregated hourly temperature measurements. Error is defined as the difference between temperatures measured by a Campbell Scientific research-grade ultrafine-wire thermocouple and a variety of loggers (see text). As an indication of the effect size being measured, the RMS difference between the ultrafine-wire thermocouple measurement and estimates from coarse-gridded ERA5 data are also shown (right-hand column) TMS4 dataloggers systematically overestimated temperatures in sunny conditions, though errors were larger for unshielded dataloggers (Figure 2).
At the forest site, overall accuracy was higher, but the accuracy of hourly temperature measurements obtained using the TMS4 dataloggers was again consistently greater than that of iButton thermochrons at all heights, with greatest accuracy achieved by the shielded TMS4 dataloggers, which gave reasonably accurate estimates. The iButton thermochrons again overestimated temperatures during the daytime. Though temperatures were more F I G U R E 1 Experiment 1. Comparisons between temperature readings obtained using different types of loggers. In (a) and (b), daytime temperatures (04:20:00-20:30:00 on 12 July 2020 GMT) obtained using a research-grade 0.0127 mm chromel-constantan wire thermocouple (Campbell Thermocouple; black) are compared to those obtained with a selection of consumer-grade temperature measurement devices. In (c) and (d), the same comparisons are made using nighttime temperatures (20:30:00-04:20:00 on 12 and 13 July 2020 GMT). Solid lines indicate the mean and the shaded area indicate the standard deviation of multiple readings obtained at 10-min intervals. For further details of sensors and their recording frequency, see text F I G U R E 2 Experiment 2. Comparisons of hourly temperatures obtained using consumer-grade 0.08-mm K-type thermocouples, TMS4 dataloggers (shielded and unshielded) and iButton thermochrons (no housing, shielded by a PVC tube, water-proofed in clear plastic dip) during selected cloudy and sunny periods (sunny, cloud cover <25%: also lower in forest environments. Only the shielded TMS4 datalogger yielded errors that were consistently smaller than this difference (Figure 3; Figure S6). Full results for both habitat types, TA B L E 4 Experiment 2. Root-mean-square (RMS) and maximum (in brackets) error of hourly temperature measurements at 0, 15 and 150 cm above-ground. Error is defined as the difference between temperatures measured using the 0.08-mm Type K thermocouples and those measured using TMS4 datalogger and iButton thermochrons at the same height. As an indication of the effect size being measured, the RMS (and maximum) difference from thermocouple temperature measurements at 150 cm in the open grassland area, best representing the reference air temperature that would be measured by a weather station, is also shown (right-hand column)  Table 4. Temperature comparisons during representative sunny and cloudy periods are shown in Figure 2 and over the duration of the study in Figure S5. Errors computed for daily maxima and minima and during selected cloudy and sunny periods are shown in Tables S2-S5.

| Experiment 3
Maximum daily temperatures recorded in the open area using the Lascar loggers with internal thermometers shielded by home-made shields were frequently overestimated by several degrees in comparison to temperatures obtained by the synoptic weather station, particularly when the funnel shield was used ( Figure 4; Table 5). Mean daily temperatures were also overestimated, though by approximately half the amount, and again temperatures were overestimated more when the funnel shield was used. Minimum temperatures were relatively accurately estimated irrespective of which shield was used. Daily and monthly RMS and maximum errors for both shield types are shown in Table 5. In general, the temperatures measured using consumer-grade devices in the forest environment were much closer to those measured using the synoptic weather station in the open environment, despite the expectation that significant habitat effects would be evident (Figure 4).

| CON CLUS IONS
The physics of thermometer heat exchange demonstrates that it is the measurement of microclimate air temperatures that is most problematic. Below the surface of the soil, radiative fluxes do not affect the temperature of a thermometer and when a thermometer can be placed in direct physical contact with a surface, conductance is high, and the radiative fluxes become less important. But how should one measure microclimate air temperatures, and how much error can one expect in doing so? From a theoretical perspective, three properties of a thermometer influence its accuracy. Firstly, as conductance is inversely related to the size of the device, a very small thermometer will obtain more accurate readings. Since wind speeds close to the ground are generally low (Campbell & Norman, 2012;Geiger, 1927;Monin & Obukhov, 1954), only the very smallest of devices, for example thermocouples of <0.  (Tarara, 2000) and thus weather proofing a thermometer using plastic casing is potentially problematic. In terms of reflective properties and size, both the device measuring temperature and the logger itself are important, with the relative importance of each depending on the extent to which they are thermally isolated from one another. An ideal thermometer should thus be as small as possible, have a surface coating of polished metal and should be thermally isolated from the data storage unit and housing. Empirically, however, we show that iButtons are likely to yield measurements that differ substantially from those obtained using research-grade equipment. This is likely due to a high proportion of the thermal heat emitted by the temperature sensor element of the iButton being absorbed and remitted by the black plastic casing on the interior surface of the logger.
If estimates of air temperature in sunny and low-wind environments are required, our results demonstrate that standard non-finewire devices often used in ecological research are not well-suited to this purpose. Notwithstanding that our experiments were conducted in environments where radiative fluxes are not at their most extreme, for the most part, errors are so large that they exceed the differences between ambient air temperature and microclimate temperature. Even in partially sunny conditions, errors of several degrees can be expected. Shielding the device from radiation offers only a partial and unsatisfactory solution. The radiation shield itself absorbs radiation and is rarely sufficiently thermally isolated from the thermometer to prevent interference. The shield will also influence the very microclimatic conditions being measured. Shields are thus most appropriate to use where localised temperature differences from the surrounding air are of less concern, and where wind speeds are sufficiently high to ensure thermal mixing. To limit heat exchange between the sensor and the shield, a sufficient distance between the shield and the sensor must be maintained, particularly in low-wind environments.
Overall, we recommend that in sunny environments an ultrafinewire thermocouple is used. The consumer-grade ultrafine-wire thermocouple tested in this study provides estimates of temperature with adequate accuracy for most purposes, and substantially greater accuracy than the majority of devices used more commonly.
We show, however, that miniaturised thermocouples will be prone to measuring rapid, random fluctuations in air temperature, which can be significant above heated ground owing to the turbulent nature of heat transfer (Campbell, 1969). Since ultrafine-wire thermocouples are likely to be responsive to these temperature fluctuations, it is necessary to set a frequent recording interval, such that 30 measurements or more are obtained for each period for which average temperature is required. At night or in shaded environments, the problem of radiation absorption is less severe, and the TMS4 dataloggers provided reasonably accurate estimates of temperature.
There is little to differentiate between whether the devices should be deployed with shields or not. In the first experiment, greater accuracy was achieved when the TMS4 loggers were unshielded, though in the second experiment greater accuracy was achieved when shielded.
Nevertheless, in many circumstances, the purpose of collecting microclimate air temperatures is to quantify the difference from those that would be recorded by a standard weather station, for example by endeavouring to estimates near the ground surface. It is generally the case that the factors contributing to this difference are also those that result in errors of temperature measurement.
Consequently, the effect size being measured and the degree of error are often correlated. Thus, with the exception of the ulrafinewire thermocouples, errors in measurements obtained by the loggers tested in this study approach, or even exceed, the effect size being measured. Likewise, if comparisons between habitats with different degrees of shading are being made, the measured difference is likely to comprise both real differences and apparent differences caused by differential sensor errors. If high accuracy is of most concern, again an ultrafine-wire thermocouple should be used, though TA B L E 5 Experiment 3. Root-mean-square (RMS) and maximum error of daily and monthly minimum, mean and maximum temperatures obtained in an open field at 2 m above-ground. Error is defined as the difference between temperatures obtained at 150 cm above-ground using a Lascar ELUSB-1 logger and two different shield types with those obtained by an adjacent official synoptic weather station. in circumstances where this is unpractical, the TMS4 loggers are the most accurate alternative.
In summary, there is no perfect way to measure air temperatures in environments where thermometers are subject to radiative fluxes and wind speeds are low enough to limit conductance. In most ecological settings where spatial replication is needed, endeavours to measure temperature will inevitably have to make a trade-off between cost, ease of deployment and data retrieval and the desired accuracy of measurements.
Consumer-grade ultrafine-wire thermocouples will offer an affordable solution for most purposes. Nevertheless, in closedcanopy environments the options available are wider, and in some circumstances the use of other logger types, particularly TMS4 dataloggers, is appropriate. Such circumstances are likely to arise when the measured effect sizes are larger compared to the expected errors, such as may occur when regional or altitudinal variation in temperature is of primary concern. Overall, we urge ecologists to pay greater heed to the physics of heat transfer when attempting to measure air temperatures and to understand the trade-offs that exist in doing so. An improved understanding of these principles will reduce the risk that highly inaccurate measurements are taken.

PE E R R E V I E W
The peer review history for this article is available at https://publo ns.

DATA AVA I L A B I L I T Y S TAT E M E N T
Microclimate sensor data used in three experiments. Dryad Digital