Ecologists often use indices or proxies to communicate complex ecological entities. Indices commonly known as Thermal Safety Margin, Habitat Thermal Quality and Hours of Restriction describe species' vulnerability to climate change by comparing organisms' thermal limits or preferences to available habitat temperatures. Ready access to temperature data, from global gridded datasets or limited in situ measurements, has made these indices popular for vulnerability assessments across taxonomic groups and regions. However, such coarse descriptions of thermal landscape mask the spatio-temporal heterogeneity that organisms experience, compromising the value of these indices. Full understanding of how scale affects index estimates is lacking, leaving ecologists and conservation managers with little guidance for applying or interpreting indices. Here, we show that incomplete temperature sampling, in space or time, provides erroneous assessments of vulnerability. Gradually sub-sampling a long-term, fine-scale dataset of operative environmental temperature altered the index estimates for a lizard. Uncertainty associated with the selection of data increased with coarser scales, often leading to contrasting interpretations about the species' vulnerability to climate change depending on the data subset used. Compressing the environmental temperature data into central or extreme tendencies, as traditionally done to compute these indices, further masked the thermal variation that animals exploit to buffer warming. We suggest the use of improved index formulations that better describe temperature availability at scales that are appropriate to the study organism.