Skip to main content
ARS Home » Pacific West Area » Tucson, Arizona » SWRC » Research » Publications at this Location » Publication #193085


item Emmerich, William
item LACA, E.

Submitted to: Journal of the Royal Statistical Society, Series A (Statistics in Society)
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 8/4/2006
Publication Date: 1/15/2007
Citation: Perez-Quezada, J.F., Saliendra, N.Z., Emmerich, W.E., Laca, E.A. 2007. Evaluation of statistical protocols for quality control of ecosystem co2 fluxes. Journal of the Royal Statistical Society Series A. 170(Part 1): 213-230.

Interpretive Summary: Micrometeorological and CO2 flux data are subject to interpretation by individual scientist and is a very labor intensive. Because of the individual interpretation there are differences in the final results produced. This work evaluated the differences between scientists and models to do the interpretation. The models were able to produce results similar to a group of scientist with less variability. Incorporating model into processing this type of scientific data will result in less labor and produce more consistent quality control of the data

Technical Abstract: The process of quality control of micrometeorological and CO2 flux data can be subjective and may lack repeatability, which would undermine the results of many studies. Multivariate statistical methods and time series analysis (TSA) were used together and independently to detect and replace outliers in CO2 flux data derived from a Bowen ratio-energy balance (BREB) system. The results were compared with those produced by five experts who applied the current and potentially subjective protocol. All protocols were tested on the same set of three five-day periods, when measurements were conducted in an abandoned agricultural field in northern Kazakhstan during the growing season of 2002. The performance of the protocols was defined by the closeness to the average of the experts. Average fluxes after correction by individual experts differed from the mean for all experts by as much as 4.4 g CO2 m-2 d-1. The statistical protocol that combined multivariate distance, multiple linear regression and TSA differed from the experts’ average by 1.15, 1.74 and 0.09 g CO2 m-2 d-1 in each period. A sequential fitting of a TSA model (second-order autoregressive) over variables used to calculate CO2 flux performed well in two datasets (showed differences of 0.07 and 0.11 g CO2 m-2 d-1), but showed a high difference on the other one (5.52 g CO2 m-2 d-1). An automated version of one of the protocols presented here could be used as a standard way of filling gaps and processing data from BREB and other techniques (e.g. eddy covariance). This would enforce objectivity in comparisons of CO2 flux data generated by different research groups and streamline the protocols for quality control.