Carbon dioxide at an unpolluted site analysed with the smoothing kernel method and skewed distributions
详细信息 查看全文
- 作者：Isidro A. Pé ; rez ; iaperez@fa1.uva.es ; M. Luisa Sá ; nchez ; M. Á ; ngeles Garcí ; a ; Nuria Pardo
- 关键词：Rural CO2 ; Kernel regression ; Distribution fitting ; Triangular distribution
- 刊名：Science of the Total Environment
- 出版年：2013
- 出版时间：1 July, 2013
- 年：2013
- 卷：456-457
- 期：Complete
- 页码：239-245
- 全文大小：601 K
文摘
CO2 concentrations recorded for two years using a Picarro G1301 analyser at a rural site were studied applying two procedures. Firstly, the smoothing kernel method, which to date has been used with one linear and another circular variable, was used with pairs of circular variables: wind direction, time of day, and time of year, providing that the daily cycle was the prevailing cyclical evolution and that the highest concentrations were justified by the influence of one nearby city source, which was only revealed by directional analysis. Secondly, histograms were obtained, and these revealed most observations to be located between 380 and 410 ppm, and that there was a sharp contrast during the year. Finally, histograms were fitted to 14 distributions, the best known using analytical procedures, and the remainder using numerical procedures. RMSE was used as the goodness of fit indicator to compare and select distributions. Most functions provided similar RMSE values. However, the best fits were obtained using numerical procedures due to their greater flexibility, the triangular distribution being the simplest function of this kind. This distribution allowed us to identify directions and months of noticeable CO2 input (SSE and April-May, respectively) as well as the daily cycle of the distribution symmetry. Among the functions whose parameters were calculated using an analytical expression, Erlang distributions provided satisfactory fits for monthly analysis, and gamma for the rest. By contrast, the Rayleigh and Weibull distributions gave the worst RMSE values.