Identification of spikes associated with local sources in continuous time series of atmospheric CO, CO 2 and CH 4

• Published in: Atmospheric Measurement Techniques
• Main Authors: A. El Yazidi, M. Ramonet, P. Ciais, G. Broquet, I. Pison, A. Abbaris, D. Brunner, S. Conil, M. Delmotte, F. Gheusi, F. Guerin, L. Hazan, N. Kachroudi, G. Kouvarakis, N. Mihalopoulos, L. Rivier, D. Serça
• Format: Article in Journal/Newspaper
• Language: English
• Published: Copernicus Publications 2018
• Subjects: Environmental engineering; TA170-171; Earthwork. Foundations; TA715-787; The Spike; Amsterdam Island
• Online Access: https://doi.org/10.5194/amt-11-1599-2018; https://doaj.org/article/769c7dc5bd314dfa9c9ae439ca0103a9

This study deals with the problem of identifying atmospheric data influenced by local emissions that can result in spikes in time series of greenhouse gases and long-lived tracer measurements. We considered three spike detection methods known as coefficient of variation (COV), robust extraction of baseline signal (REBS) and standard deviation of the background (SD) to detect and filter positive spikes in continuous greenhouse gas time series from four monitoring stations representative of the European ICOS (Integrated Carbon Observation System) Research Infrastructure network. The results of the different methods are compared to each other and against a manual detection performed by station managers. Four stations were selected as test cases to apply the spike detection methods: a continental rural tower of 100 m height in eastern France (OPE), a high-mountain observatory in the south-west of France (PDM), a regional marine background site in Crete (FKL) and a marine clean-air background site in the Southern Hemisphere on Amsterdam Island (AMS). This selection allows us to address spike detection problems in time series with different variability. Two years of continuous measurements of CO 2 , CH 4 and CO were analysed. All methods were found to be able to detect short-term spikes (lasting from a few seconds to a few minutes) in the time series. Analysis of the results of each method leads us to exclude the COV method due to the requirement to arbitrarily specify an a priori percentage of rejected data in the time series, which may over- or underestimate the actual number of spikes. The two other methods freely determine the number of spikes for a given set of parameters, and the values of these parameters were calibrated to provide the best match with spikes known to reflect local emissions episodes that are well documented by the station managers. More than 96 % of the spikes manually identified by station managers were successfully detected both in the SD and the REBS methods after the best adjustment of ...