Ad hoc instrumentation methods in ecological studies produce highly biased temperature measurements

Abstract

In light of global climate change, ecological studies increasingly address effects of temperature on organisms and ecosystems. To measure air temperature at biologically relevant scales in the field, ecologists often use small, portable temperature sensors. Sensors must be shielded from solar radiation to provide accurate temperature measurements, but our review of 18 years of ecological literature indicates that shielding practices vary across studies (when reported at all), and that ecologists often invent and construct ad hoc radiation shields without testing their efficacy. We performed two field experiments to examine the accuracy of temperature observations from three commonly used portable data loggers (HOBO Pro, HOBO Pendant, and iButton hygrochron) housed in manufactured Gill shields or ad hoc, custom‐fabricated shields constructed from everyday materials such as plastic cups. We installed this sensor array (five replicates of 11 sensor‐shield combinations) at weather stations located in open and forested sites. HOBO Pro sensors with Gill shields were the most accurate devices, with a mean absolute error of 0.2°C relative to weather stations at each site. Error in ad hoc shield treatments ranged from 0.8 to 3.0°C, with the largest errors at the open site. We then deployed one replicate of each sensor‐shield combination at five sites that varied in the amount of urban impervious surface cover, which presents a further shielding challenge. Bias in sensors paired with ad hoc shields increased by up to 0.7°C for every 10% increase in impervious surface. Our results indicate that, due to variable shielding practices, the ecological literature likely includes highly biased temperature data that cannot be compared directly across studies. If left unaddressed, these errors will hinder efforts to predict biological responses to climate change. We call for greater standardization in how temperature data are recorded in the field, handled in analyses, and reported in publications.

Modelling high‐latitude summer temperature patterns using physiographic variables

“…In contrast to the earlier findings of Lundquist and Huggett (), non‐standard shielding has since been associated with temperature bias and enhanced diurnal variability (Terando et al, ). While it was not possible to employ Gill shields at all sites, we applied a statistical correction to tree‐mounted sensor measurements using models developed at locations having paired sensors in Gill shields and mounted on trees (including data from three LTEM sites for a total five sites).…”

Which is more biased: Standardized weather stations or microclimatic sensors?

“…A recent article in Ecology and Evolution by Terando, Youngsteadt, Meineke, and Prado (2017) compared observations from standardized weather stations (shielded at a height of ~2 m) with those from cheaper microclimatic sensors with custom‐built radiation shields. They found there could be biases of up to 3–5°C under full sun conditions on hot days.…”

“…Terando et al. (2017) start with the assumption that the truth is air temperatures in radiation shields at a height of ~2 m. However, using the terminology of Austin (2002), air temperatures are only an indirect predictor of temperatures that are likely to be ecologically meaningful, like body, soil, or leaf temperatures. Indeed, Terando et al.…”

“…Indeed, Terando et al. (2017) note that only a third of studies with microclimatic sensors were used to record air temperatures, so the majority were observing soil, water, or other temperatures that were presumably targeted to be more biologically relevant to the species or communities of interest. We should not be discouraging researchers from observing conditions that more closely reflect those of our target species, so greater standardization of observations may appear beneficial, but actually lead to observations that are less ecologically relevant.…”

“…Similarly, Terando et al. (2017) conclude that observations of minimum temperatures are relatively unbiased, but this also overlooks the importance of sensor height. It has long been known that there can be frosts at surface level even when air temperatures at 2 m are positive (Geiger, 1950), and given that frosts can be biologically destructive, this is a crucial bias in minimum air temperatures.…”