2.3. Methane
2.3.1. In Situ Methane Measurements
Quasi-continuous measurements of atmospheric methane continued at MLO and BRW at a frequency of four ambient measurements each hour. The relative precision is 0.07%. Details of the measurement techniques and analysis of the in situ data through early 1994 were published in 1995 [Dlugokencky et al., 1995]. Daily averaged methane mole fractions (in nanomol mol-1) are plotted in Figure 2.10 for BRW (a) and MLO (b). The data have been edited for instrument malfunction using a rule-based expert system [Masarie et al., 1991], but they were not selected for meteorological conditions. High CH4 values at BRW are due to emissions from local sources. Limitations of the unselected data sets have been discussed previously by Dlugokencky et al. [1995].
Fig. 2.10. Daily mean CH4 mole fractions in nanomol mol-1 for (a) BRW and (b) MLO for 1996 and 1997. The data are unconstrained but have undergone a quality control step to ensure that the analytical instrument was working optimally when they were obtained [Masarie et al., 1991].
The quasi-continuous measurements from BRW and MLO are an invaluable quality control tool when they are compared with weekly, discrete samples from these sites. They are used to illuminate sampling errors, to determine biases associated with weekly sampling, and to estimate reasonable uncertainties for monthly means, trends, and other parameters determined from the weekly samples.
Histograms of the differences between in situ and flask sample measurements are plotted in Figure 2.11. Differences were determined as the difference between the average of all discrete samples collected with a unique sample date and a value interpolated from hourly averages at the time the discrete sample was collected. A smooth, normal distribution is also plotted based on the mean and standard deviation of the population of differences. The mean differences (one sigma) are -0.9 ± 4.8 ppb at MLO and 0.4 ± 6.0 ppb at BRW, where outliers greater than 3 sigma were excluded. This result gives us confidence that our sampling methods are not resulting in an offset in the CH4 discrete sample measurements.
Fig. 2.11. Histograms of differences between discrete samples and in situ measurements interpolated from hourly averages for MLO (a) and BRW (b) from samples collected through 1996. The smooth curve is a normal distribution calculated from the population of differences.
If we assume that high frequency sampling results in time series that accurately define the parameters of interest (e.g., trend and seasonal cycle), then a qualitative comparison of in situ and discrete sampling time series should reveal how accurately weekly sampling captures the main features in the data. In Figure 2.12 smooth curves fitted to discrete and in situ time series are plotted. Agreement is excellent. For example, trends for BRW are 6.9 ppb yr-1 for discrete sample and 7.4 ppb yr-1 for in situ, and at MLO 7.7 ppb yr-1 for discrete sample and 7.7 ppb yr-1 for in situ. Additionally, uncertainties in trends, changes in trends, and monthly averages can be calculated for the discrete sample measurements from in situ measurements using a nonparametric statistical technique called the bootstrap method, similar to what we have done for CO2 [Tans et al, 1990b]. To do this, 50 "bootstrap" time series are created by randomly selecting hourly averages, one per week, from BRW and MLO in situ records. For BRW data are used only if they were obtained when the wind was from the clean air sector, and for MLO the in situ records were constrained to downslope times. Next, the parameters of interest are determined for each "bootstrap" time series. These parameters are averaged and their standard deviation calculated; the standard deviation is used as the uncertainty for that parameter for the discrete sample data. For MLO from April 1987 through 1996, the trend is 7.7 ± 0.3 ppb yr-1, changing at -1.4 ± 0.1 ppb yr-2. At BRW from 1986 through 1996, the trend is 6.9 ± 0.3 ppb yr-1, decreasing at -1.0 ± 0.1 ppb yr-2. Uncertainties can also be calculated for mean seasonal cycles and monthly mean values.
Fig. 2.12. Smooth curves fitted to discrete samples (dashed line) and in situ measurements (solid line) for BRW and MLO.
In summary we find good agreement between discrete samples and in situ measurements at BRW and MLO suggesting that storing the discrete samples has little effect on the integrity of the samples, at least with respect to methane. Weekly samples reproduce the main features in the in situ records reasonably well at BRW and MLO. This likely holds true for most of the sites in our sampling network. Finally, using nonparametric statistical methods, in situ measurements are useful for calculating uncertainties for discrete sample records.