Cluster analysis applied to CO₂ concentrations at a rural site.

In rural environments, atmospheric CO2 is mainly controlled by natural processes such as respiration-photosynthesis or low atmosphere evolution. This paper considers atmospheric CO2 measurements obtained at a rural site during 2011 using the wavelength-scanned cavity ringdown spectroscopy technique and presents two clustering methods, the silhouette being calculated to evaluate procedure validity. In the first method, clusters were formed depending on the similarity of wind roses, with satisfactory silhouette values. An anticyclonic rotation of the wind direction was observed during the daily cycle and clusters were formed by consecutive directions following the mixing layer evolution. However, monthly roses revealed four quite different wind directions, mainly oriented in the E-W axis. Although CO2 was not used in this procedure, a successful link between clusters and CO2 was obtained. In the second procedure, clusters were formed by the similarity of CO2 histograms calculated in intervals of one or two ancillary variables, wind direction, time of day, or month. The influence of a nearby city, the daily evolution of the low atmosphere, and the growing season were highlighted. Finally, the usefulness of the method lies in its easy extension to other gases or variables.