Similar cost scores are produced for the starting regional grids and the solution grid methods used. The regions in this section which are close to the observation sites (1, 2, 3, 5, 6, 7) all agree well with the 2012 NAEI. Overall, changing the starting regions does not appear to have a signicant impact on these emission totals, despite relatively dierent solution grids being produced. The county-based regions were chosen as part of the nal InTEM setup as these totals could aid policy makers, and are easier to visualise when communicating emission totals.
Table 5.3: InTEM coarse regional emission totals (kt yr-1) for the NAME-based and NAEI-based solution grids. Solution grids are based on the alternative 16 box starting region grid. Total does not include border regions (15 and 16).
1 2 3 4 5 6 7 8 9
2012
NAEI 11.3 12.9 8.0 6.5 14.2 8.9 12.0 7.2 13.4
NAME-method 9.7 6.9 8.5 1.5 9.0 12.1 12.5 11.5 13.2
sd 2.5 0.9 0.6 1.2 1.7 2.6 2.8 3.1 4.9
NAEI-method 10.3 7.6 7.2 8.3 9.0 15.9 7.2 17.9 7.3
sd 2.5 0.7 0.7 1.5 1.4 1.5 2.2 1.73 5.4
10 11 12 13 14 15 16 Total Cost
Score 2012
NAEI 8.8 14.9 46.9 49.59 0.3 152.9 1.7 369.4 -
NAME-method 3.6 3.7 58.6 55.22 2.0 580.0 189.8 206.1 12.5 sd 4.6 4.1 18.6 15.59 3.4 50.0 1.7 26.4 0.0
NAEI-method 0.8 12.0 60.2 58.08 0.2 560.0 107.2 242.2 12.7 sd 1.1 2.0 14.5 12.90 0.3 30.0 1.4 20.8 0.0
5.7 Observational uncertainty and variability
As mentioned in the methods chapter (Section 3.3), uncertainties associated with both the measurements and the modelling can be incorporated into InTEM to inuence the resulting emissions produced. This section describes two dierent approaches which can represent observational uncertainties, i.e. uncertainties associated with the instrumentation and the measured observations. Both approaches incorporate hourly instrument precision and calibration uncertainties.
These values range between 5 and 10 ppb on average. One method also includes
Chapter 5 Development of the inversion approach
the standard deviation within each hourly observation. This measure of the hourly observational variability is a method of paramertising `random' errors within the dispersion model (explained in greater detail in Section 5.8). These ppb values are larger than the instrument uncertainties, ranging from a few ppb to over 100 ppb in some instances. High, short lived methane spikes are inconsistently seen at two of the sites (Haddenham and Tilney) resulting from local methane sources. If a large methane concentration is observed, which has relatively low uncertainty associated with it, InTEM will try to recreate this magnitude. It was therefore considered that these instances could bias the inversion, and thus the hourly standard deviation was used as a measure of variability / uncertainty.
A) −0.5 0.0 0.5 1.0 1.5 2.0
51.552.052.553.0
Longitude (degrees)
Latitude (degrees)
●
●
●
●
HD
TN WY TY
0 24.1 42.8 76 135.2 240.3 427.1 759.4 1350 2400
Max Value = 11666.04 ng s−1 m−2 Total Emissions = 312 kt yr−1
B) −0.5 0.0 0.5 1.0 1.5 2.0
51.552.052.553.0
Longitude (degrees)
Latitude (degrees)
●
●
●
●
HD
TN WY TY
0 24.1 42.8 76 135.2 240.3 427.1 759.4 1350 2400
Max Value = 16729.68 ng s−1 m−2 Total Emissions = 350 kt yr−1
C) −0.5 0.0 0.5 1.0 1.5 2.0
51.552.052.553.0
Longitude (degrees)
Latitude (degrees)
●
●
●
●
HD
TN WY TY
0 24.1 42.8 76 135.2 240.3 427.1 759.4 1350 2400
Max Value = 19058.14 ng s−1 m−2 Total Emissions = 283 kt yr−1
Figure 5.9: InTEM emission elds comparing two setups assessing observational uncertainty. A) Observational uncertainty dened as hourly instrument precision and calibration uncertainties. B) Observational uncertainty dened as hourly instrument precision, calibration uncertainties and hourly concentration standard deviations. C) NAEI regridded onto 150 box solution grid. All uncertainties expressed in ppb.
5.7 Observational uncertainty and variability
Figure 5.9 shows the ne spatially-resolved emission maps when hourly standard deviations are not included (Figure 5.9.A) and are included (Figure 5.9.B) within InTEM. The NAEI at the same resolution is also included for reference (Figure 5.9.C). The county regional totals for all three maps are shown in Table 5.4. The overall totals for both methods are similar, however some of the regions dier signicantly. The region totals near to the observation sites (4, 10 and 15) increase when no standard deviation is included. Additionally, more methane is placed in region 3 (border box to the north east). There is signicantly less methane emitted in regions 5 and 8 however, which correspond to the Essex county region and the border box to the south east, respectively.
Table 5.4: InTEM county emission totals (kt yr-1) for experiments comparing observational uncertainty. Region label abbreviations: CBG Cambridge, SFK Suolk, NFK Norfolk, LDN London, ESX Essex, LCS Lincolnshire, BHS -Bedfordshire / Hertfordshire, SW - Region to the south west (L-shaped). Regions 3, 7, 8 and 13 make up the border. Regions 9, 12 and 14 divide the sea. NSC = Norfolk (10) + Suolk (4) + Cambridgeshire (15). Total does not include border regions.
CBG NFK SFK LDN ESX LCS Sea Regions
15 10 4 11 5 6 2 12 14
2012
NAEI 26.5 38.9 24.1 51.2 24.5 17.6 0.1 0.1 0.3
SD
Included 20.5 37.1 22.8 45.7 19.6 9.1 0.2 0.1 0.4
sd 2.1 1.7 1.9 18.0 8.1 4.1 0.6 0.1 1.2
SD Not
Included 27.7 42.6 29.3 55.0 7.7 9.9 0.2 0.1 0.1
sd 2.2 2.9 4.2 25.9 5.3 9.7 0.5 0.1 0.3
BHS SW Border Regions
Total NSC Cost Score
9 1 3 7 8 13
2012
NAEI 20.5 75.0 1.6 29.9 0.1 55.2 369.4 89.6
-SD
Included 30.6 124.5 116.1 234.0 53.6 294.9 305.5 80.4 12.5
sd 7.3 59.2 1.0 7.1 16.1 43.1 63.1 3.3 0.0
SD Not
Included 34.0 134.0 125.5 223.8 17.1 290.9 357.7 99.6 29.7
sd 5.1 57.9 1.5 17.2 9.1 38.7 64.8 5.6 0.0
In other comparisons explained in this chapter cost scores have been used to identify a more relevant experimental setup, for example Section 5.4. However, in this section, cost score cannot be used as a metric to justify one experimental setup over another. This is because the fundamental mechanics of the chosen cost
Chapter 5 Development of the inversion approach
score in this analysis is based upon uncertainty. The cost score is calculated by dividing the residual by the associated uncertainty at each timestep and thus the resulting scores would have to be normalised in order to be used as a comparison.
Previous analysis showed the elevated, short lived methane peaks to be from local point sources (Section 4.6). Therefore, an increase in methane emission totals in the regions close to the observation sites is to be expected when no hourly standard deviation is included in the InTEM setup. Less obvious reasons exist for the decrease in regions 5 and 8. This could be due to the distance these regions are from the observation sites as more distant regions seem to be consistently less resolved than regions which contain the observation sites. More importantly the inclusion of hourly variability acts as a pseudo-modelling uncertainty. This is described in further detail in the next section. For these reasons the hourly standard deviations were chosen to be used in the nal inversion setup.