A new study has found that inaccuracies in the temperature data used in climate impact research could lead scientists to underestimate the effects of extreme heat by 15% or more.
Published in the Journal of the Association of Environmental and Resource Economists, the study by researchers from Wake Forest University and the University of California, San Diego examined how temperature estimates are constructed when direct weather station measurements are unavailable for specific cities, neighbourhoods or agricultural areas.
The researchers found that these temperature “proxies” can contain substantial errors, particularly in areas with complex geography. In mountainous Boulder, Colorado, for example, estimated temperatures differed from actual temperatures by nearly 15°F on average, while errors in more densely monitored and relatively flat Chicago were as low as 1°F.
The study also found that temperature measurement errors can vary systematically by location and persist over time. This can affect studies examining links between extreme heat and outcomes such as violent crime, public health, electricity demand and crop losses.
Using more than 20 million crime records, the researchers compared estimates based on city-level temperature and crime data with those produced using data aggregated at the county level. They found that the estimated relationship between extreme heat and violent crime became smaller when the data were measured at the coarser geographic scale.
One commonly used temperature proxy examined in the research had an average measurement error of 3.3°F, or about 1.85°C. The researchers noted that this is larger than the 1.5°C warming threshold associated with the Paris Agreement, highlighting the potential significance of temperature measurement uncertainty when assessing climate impacts.
The researchers said the findings do not mean climate impacts are necessarily being underestimated in every study. Rather, the direction and size of the bias can vary depending on how temperature is measured and the location being studied.
The study recommends improving temperature monitoring by expanding and maintaining weather station networks and matching temperature observations more closely to the places where people, crops, businesses and ecosystems are actually exposed to heat.
The researchers also found that no single temperature dataset is optimal everywhere. In regions with dense weather station coverage, measurements from nearby stations can outperform complex gridded products, while gridded datasets may be more useful in areas where monitoring networks are sparse.
The findings suggest that improving the accuracy and geographic resolution of temperature measurements could help researchers produce more reliable estimates of the risks posed by extreme heat and provide policymakers with better information for climate adaptation planning.