This question is important in order to assess the robustness of the trends and variability in the surface temperature records. A new paper was recently accepted by the AMS Journal of Atmospheric and Ocean Technology that addresses this issue. It is titled: “Comparison of Co-Located Automated (NCECONet) and Manual (COOP) Climate Observations in North Carolina” by Christopher Holder, Ryan Boyles, Ameenulla Syed, Dev Niyogi, and Sethu Raman. A draft copy is available at the link above.
Even though the study is focused over North Carolina, we believe the findings can be considered generic enough and provide a feel for the uncertainty in the datasets. The abstract states,
“The National Weather Service’s cooperative observer network (COOP) is a valuable climate data resource that provides manually observed information on temperature and precipitation across the nation. These data are part of the climate dataset and continue to be used in evaluating weather and climate models. Increasingly, weather and climate information is also available from automated weather stations. A comparison between these two observing methods is performed in North Carolina, where thirteen of these stations are collocated. Results indicate that, without correcting the data for differing observation times, daily temperature observations are generally in good agreement (0.96 Pearson product-moment correlation for minimum temperature, 0.89 for maximum temperature). Daily rainfall values recorded by the two different systems correlate poorly (0.44), but the correlations are improved (to 0.91) when corrections are made for the differences in observation times between the COOP and automated stations. Daily rainfall correlations especially improve with rainfall amounts less than 50 mm per day. Temperature and rainfall have high correlation (nearly 1.00 for maximum and minimum temperatures, 0.97 for rainfall) when monthly averages are used. Differences of the data between the two platforms consistently indicate that COOP instruments may be recording warmer maximum temperatures, cooler minimum temperatures, and larger amounts of rainfall, especially with higher rainfall rates. Root mean square errors are reduced by up to 71% with the day-shift and hourly corrections. This study shows that COOP and automated data (such as from NCECONet) can, with simple corrections, be used in conjunction for various climate analysis applications such as climate change and site-to-site comparisons. This allows a higher spatial density of data and a larger density of environmental parameters, thus potentially improving the accuracy of the data that are relayed to the public and used in climate studies.”
Some interesting findings from our study suggest:
It is not possible to state which is correct but does provide insights into the uncertainty and differences one could expect just from differences in the measurements from collocated stations. The errors or uncertainty that will be caused by station location is another issue we could not address but is an important issue for data representativeness.
This brings back the old adage “Everyone but the observer believes in the observations; nobody except the modeler believes in his/her model results!” , which for the climate scenarios seem to be flipped over and people often may start looking at model results as “reality”.