SEARCH

SEARCH BY CITATION

Keywords:

  • nitrous oxide;
  • N2O;
  • methane;
  • CH4;
  • inversion modeling;
  • emission estimates

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Observations
  5. 3. NAME Dispersion Model
  6. 4. Baseline Concentration Analysis
  7. 5. Regional Emission Estimates
  8. 6. Discussion
  9. 7. Conclusions
  10. Acknowledgments
  11. References
  12. Supporting Information

[1] Methane (CH4) and nitrous oxide (N2O) have strong radiative properties in the Earth's atmosphere and both are regulated through the United Nations Framework Convention on Climate Change. Through this convention the United Kingdom is obliged to report an inventory of annual emission estimates from 1990. This paper describes a methodology that estimates emissions of CH4 and N2O completely independent of the inventory values. Emissions have been estimated for each year 1990–2007 for the United Kingdom and for NW Europe. The methodology combines high-frequency observations from Mace Head, a monitoring site on the west coast of Ireland, with an atmospheric dispersion model and an inversion system. The sensitivities of the inversion method to the modeling assumptions are reported. The 20 year Northern Hemisphere midlatitude baseline mixing ratios, growth rates, and seasonal cycles of both gases are also presented. The results indicate reasonable agreement between the inventory and inversion results for the United Kingdom for N2O over the entire period. For CH4 the agreement is poor in the 1990s but good in the 2000s. The UK CH4 inventory reported reduction from 1990–1992 to 2005–2007 (over 50%) is dominated by changes to landfill and coal mine emissions and is more than double the corresponding drop in the inversion estimated emissions (24%). The inversion results suggest that the United Kingdom has met its Kyoto commitment (−12.5%) but by a smaller margin (−14.3%) than reported (−17.3%). The results for NW Europe with the United Kingdom removed show reasonable agreement in trend, on average the inversion results for N2O are 25% lower and for CH4 21% higher.

1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Observations
  5. 3. NAME Dispersion Model
  6. 4. Baseline Concentration Analysis
  7. 5. Regional Emission Estimates
  8. 6. Discussion
  9. 7. Conclusions
  10. Acknowledgments
  11. References
  12. Supporting Information

[2] Methane (CH4) and nitrous oxide (N2O) are important greenhouse gases in the Earth's climate system. By 2008 the annual average mixing ratios of these gases in the midlatitudes of the Northern Hemisphere had reached 1864 ppb and 322 ppb, respectively. CH4 has a 100 year Global Warming Potential (GWP) of 25 based on an adjustment time (lifetime) of 12 years, whereas N2O has a 100 year GWP of 298 and an adjustment time of 114 years [Forster et al., 2007]. Globally the principle anthropogenic sources of CH4 in Tg(CH4) a−1 are agriculture (107–175), waste (35–69), energy (74–77), coal mining (30–46), oil and gas industry (52–68) and biomass and biofuel burning (14–88). For N2O the principal global anthropogenic sources, in Tg(N) a−1, are agriculture (1.7–4.8), industrial processes (0.2–1.8), rivers/coastal waters (0.5–2.9) and biomass burning (0.2–1.0). The values in brackets show the range of emission estimates from various studies as reported by Denman et al. [2007].

[3] In NW Europe (Ireland, United Kingdom, France, Germany, Belgium, Netherlands, Luxembourg, Denmark, hereafter referred to as NWEU), the principle anthropogenic sources of CH4 in 2006 were reported to the United Nations Framework Convention on Climate Change (UNFCCC) (http://unfccc.int/2860.php) to be agriculture (58%), waste (26%) and energy (16%). The natural sources of CH4, globally significant through wetland emissions, are thought to be small in comparison to the anthropogenic releases in the NWEU. For N2O in 2006, the principle anthropogenic sources in the NWEU as reported to the UNFCCC are agriculture (71%), industrial processes (14%), energy (12%) and waste (3%) and again the natural sources are considered to be small in comparison. The anthropogenic emissions of both gases in the NWEU are covered by the UNFCCC reporting process.

[4] Inverse modeling has been used to estimate global and regional surface emissions of CH4 [Hein et al., 1997; Dentener et al., 2003; Mikaloff Fletcher et al., 2004a, 2004b; Xiao et al., 2004; Frankenberg et al., 2006; Chen and Prinn, 2006; Bousquet et al., 2006] and N2O [Prinn et al., 1990; Hirsch et al., 2006; Huang et al., 2008; Bergamaschi et al., 2009]. Fewer inversion studies have attempted to estimate regional- to country-scale emissions [Stohl et al., 2009; Bergamaschi et al., 2005]. Stohl et al. [2009] estimate country, regional and global emissions for three greenhouse gases, HFC-134a, HFC-152a and HCFC-22 for 2005–2007. Their approach uses the FLEXPART Lagrangian model to describe the recent history of the air that reaches nine observation stations globally distributed. They use a priori emission maps and an inversion method that estimates both the baseline and the regional emissions that form part of the global estimates. Bergamaschi et al. [2005] present a global inversion of CH4 with a zoom component over Europe for the year 2001. The inversion system uses a priori data and many high-frequency and weekly observations distributed globally but focused in Europe to estimate annual emissions from the countries in Europe. The results for Germany, France and the United Kingdom suggest the inventories are significantly (50–90%) underreported. Germany has since recalculated its reported emissions for 2001. Subsequent inventory recalculations for the United Kingdom have lead to an increase of 38% for 2001 short of the 93% suggested by Bergamaschi et al. [2005]. Further recent work by Bergamaschi et al. [2010] shows they have revised their UK CH4 estimates downward and they are now much more aligned with the current increased inventory estimates. One of the issues with the 2005 work was the coarse resolution of the computational grid used in the analysis (1°) and this was particularly problematic for the Mace Head station in westerly airflows. As a result the estimated modeled Irish emissions were lower than the reported UNFCCC values and explain to some degree the modeled compensating elevated UK emissions.

[5] In the work presented here more than 20 years of high-frequency, in situ CH4 and N2O observations (1989 to April 2009) from Mace Head, a remote observation station on the west coast of Ireland, have been used. This work follows on from the work presented by Manning et al. [2003], where 5 year estimates of CH4 and N2O were presented for NWEU. A recent paper by Villani et al. [2010] confirmed that the Mace Head station is sensitive to emissions in Ireland, the United Kingdom and NW continental Europe, consistent with the region considered in this paper. The observations have been used to estimate, for both gases, the trends in background midlatitude Northern Hemisphere mixing ratios (baseline) and for the more limited period, 1990–2007, the UK and NWEU emissions. These estimates have been compared to those reported to the UNFCCC in 2009 which cover the period 1990–2007.

[6] The principle tool used in the analysis of the Mace Head observations is the NAME dispersion model (Numerical Atmospheric dispersion Modelling Environment) developed by the UK Met Office [Ryall and Maryon, 1998; Jones et al., 2007]. The NAME atmospheric dispersion tool has been used for a diverse range of applications, e.g., air quality [Redington et al., 2009; Witham and Manning, 2007], emergency response [Gloster et al., 2007; Ryall and Maryon, 1998] and inversion modeling [Manning et al., 2003; Reimann et al., 2005; Derwent et al., 2007; O'Doherty et al., 2004].

2. Observations

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Observations
  5. 3. NAME Dispersion Model
  6. 4. Baseline Concentration Analysis
  7. 5. Regional Emission Estimates
  8. 6. Discussion
  9. 7. Conclusions
  10. Acknowledgments
  11. References
  12. Supporting Information

[7] Since 1987, 40 min in situ CH4 measurements have been made as part of the Global Atmospheric Gases Experiment (GAGE) and since 1994 as part of the Advanced Global Atmospheric Gases Experiment (AGAGE) at Mace Head, Ireland (53.3°N, 9.9°W). CH4 is measured in air samples with the “AGAGE-MD” system with gas chromatograph (GC, Agilent 5880) and flame ionization detection. Samples are injected at 20 min intervals, with alternate ambient air and standard gas samples, giving a total of 36 calibrated air measurements per day. The calibration scale was established by Tohoku University based on gravimetric preparation of mixtures by Nippon Sanso. A detailed description of this calibration is reported by Cunnold et al. [2002] and Prinn et al. [2000]. The choice of calibration scale affects the absolute magnitude of the baseline concentrations but has no effect on the inversion results.

[8] For much of the time, the measurement station, which is situated on the Atlantic coast, monitors clean westerly air that has traveled across the North Atlantic Ocean (34–51% of the time). However, when the winds are easterly, Mace Head receives substantial regional-scale pollution in air that has traveled from the industrial regions of Europe (18–35% of the time). The site is therefore uniquely situated to record trace gas mixing ratios associated with both the midlatitude Northern Hemisphere background levels and with the more polluted air arising from Europe. The percentage average annual and monthly contributions from different geographical regions are shown in Tables 1 and 2, respectively. The definitions of the different categories are given in section 4. No long-term annual trend in the air mass classifications is discernible. There is a seasonal signal with more “Local” and “Mixed Origin” periods and fewer “European” periods in the summer (May-September) periods. The sum of “Mixed Origin,” “European” and “Baseline” periods, i.e., those used in the inversion process, shows no annual trend and only a modest seasonal cycle (9% amplitude) with a April-July minima.

Table 1. Percentage of 3 h Periods of Each Air Mass Classification per Yeara
YearMixed OriginBaselineSouthern LatitudesEuropeanLocalMixed, Baseline, European
  • a

    Mixed origin indicates 3 h periods that do not satisfy the criteria of the other classes.

19891951121890
199019512191089
199121401261187
19922050220891
19932144126891
19942247123891
199521391291089
19962136135793
199720341351189
19982047124891
19992047123990
200023451211188
20012043127990
200221411261188
20031844128891
20042049222791
200520473181285
20062244223989
200719462211385
20081849123891
Table 2. Percentage of 3 h Periods of Each Classification per Month Over All Yearsa
MonthMixed OriginBaselineSouthern LatitudesEuropeanLocalMixed, Baseline, European
  • a

    Mixed origin indicates 3 h periods that do not satisfy the criteria of the other classes.

Jan2049127397
Feb1851124393
Mar2145123689
Apr21362241081
May19323311983
Jun23432151481
Jul28430141785
Aug23491151587
Sep18462281392
Oct1648230594
Nov1950126596
Dec1645138399

3. NAME Dispersion Model

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Observations
  5. 3. NAME Dispersion Model
  6. 4. Baseline Concentration Analysis
  7. 5. Regional Emission Estimates
  8. 6. Discussion
  9. 7. Conclusions
  10. Acknowledgments
  11. References
  12. Supporting Information

[9] NAME is a Lagrangian particle model that can be driven by stored three-dimensional meteorological fields from a numerical weather prediction (NWP) model. In this work two independent sets of meteorology have been used to drive NAME: the Met Office's operational global NWP model, the Unified Model (referred to as UKMO) [Cullen, 1993], and the recently available reanalysis meteorology, ERA Interim (referred to as ERAI) [Berrisford et al., 2009], produced by the European Centre for Medium Range Weather Forecasts (ECMWF). The former is available from 1995 onward to current day and has undergone episodic improvements over the entire period but principally, with resolution upgrades, in 1999, 2002 and 2005. At the time of writing, ERAI meteorology was available from 1989 until 2008. It has the advantage of using a consistent model configuration throughout this time period. The meteorology from both models is a sequence of short-term forecasts that are produced from successive meteorological analyses. These analysis states use the available meteorological observations (land surface, ocean, satellite, etc.) to constrain the model state to the evolution of the real atmosphere. UKMO has a horizontal resolution of ∼60 km pre-December 2005 and then ∼40 km onward, the number of vertical levels used by NAME has increased from 11 (1995–1998) to 31 (from December 2005 to June 2009). ERAI is a spectral model with an equivalent horizontal resolution of 0.75° (∼80 km) and NAME uses its lowest 37 vertical levels.

[10] NAME was initially developed, more than 20 years ago, as an emergency response tool following the Chernobyl disaster but has since evolved into a general purpose dispersion model covering spatial scales from a few hundred meters to global scales. The model air parcels move in a three-dimensional space driven by either three-dimensional NWP meteorology or single-site observations. A random walk technique is used to simulate atmospheric turbulence. In the boundary layer, profiles of turbulence are estimated [Morrison and Webster, 2005], these converge at the top of the boundary layer to a small value in the free troposphere with a fixed standard deviation (σu = 0.25 m s−1 and σw = 0.1 m s−1). In the work presented here the model schemes for representing chemistry, dry deposition, wet deposition and radioactive decay were not used, i.e., the gases were considered to be inert tracers, which, given their long respective atmospheric lifetimes and the fact the model domain was limited to the European/North Atlantic region was considered to be appropriate. The most significant loss process to consider here is OH oxidation of methane, however, over a 12 d period, with typical OH levels, the depletion in methane from OH oxidation amounts to ∼0.7%. Even with ten times the amount of OH, such as would be seen in a strong photochemical ozone episode, the depletion would only be 7%. This is trivial in comparison with the other uncertainties. The depletions from surface uptake of methane would be even smaller than from OH depletion. The boundary layer depth (BLD) estimates are taken from the driving NWP meteorological model except for pre-August 2002 UKMO (unavailable). Pre-August 2002, NAME directly estimates the BLD as described by Manning et al. [2003]. The minimum BLD allowed within NAME was set to 100 m to be consistent with the maximum emission height and the height of the output grid. The impact of using different BLD estimates is further discussed in section 5.3.

[11] The NAME model is run in backward mode to estimate the recent history (12 d) of the air en route to Mace Head. An air history map, such as those shown in Figure 1, has been calculated for each 3 h period from 1995 until June 2009 using UKMO and from 1989 to 2008 using ERAI, amounting to more than 90,000 maps. The model output estimates the 12 d time-integrated air concentration (dosage) at each grid box (fixed at 40 km horizontal resolution and 0–100 m above ground level irrespective of meteorology used) from a release of 1 g s−1 at Mace Head (the receptor). The model is three-dimensional therefore it is not just surface to surface transport that is modeled. An air parcel can travel from the surface to a high altitude and then back to the surface but only those times when the air parcel is within the lowest 100 m above the ground will it be recorded in the maps. The computational domain covers 100.0°W to 45.125°E longitude and 10.0°N to 80.125°N latitude and extends to more than 10 km vertically. For each 3 h period 33,000 inert model particles were used to describe the dispersion, increasing the number of particles has been shown to have negligible impact on the noise within the inversion domain.

image

Figure 1. Examples of 3 h air history maps derived from the Numerical Atmospheric dispersion Modelling Environment (NAME) dispersion model (a) baseline period and (b) regionally polluted period. The air history maps describe which surface areas in the previous 12 d impact the observation point within a particular 3 h period. The black box indicates the geographical domain used for the inversion study.

Download figure to PowerPoint

[12] By dividing the dosage [g s m−3] by the total mass emitted [3600 s h−1 × 3 h × 1 g s−1] and multiplying by the geographical area of each grid box [m2], the model output is converted into a dilution matrixD [s m−1]. Each element of this matrix D dilutes a continuous emission (e) of 1 g m−2 s−1 from a given grid box over the previous 12 d to an air concentration [g m−3] at the receptor (m) during a 3 h period (equation (1)).

  • equation image

Running the model backward in this way, compared to a forward simulation, is very computationally efficient as every modeled particle has a direct impact on the solution. However, it should be noted that backward running is only a very close approximation to running forward, it is not exactly reversible.

4. Baseline Concentration Analysis

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Observations
  5. 3. NAME Dispersion Model
  6. 4. Baseline Concentration Analysis
  7. 5. Regional Emission Estimates
  8. 6. Discussion
  9. 7. Conclusions
  10. Acknowledgments
  11. References
  12. Supporting Information

[13] Baseline concentrations are defined as those that have not been influenced by significant emissions within the previous 12 d of travel en route to Mace Head, i.e., those that are well mixed and are representative of the midlatitude Northern Hemisphere background concentrations. The observations at Mace Head from 1989 to June 2009 have been analyzed for each gas measured. The analysis considers the long-term trend of the monthly and annual baseline mixing ratios, their rate of growth and their seasonal cycle.

4.1. Defining Baseline Observations

[14] A 3 h period is classed as “baseline” if it meets certain criteria with respect to a dilution sensitivity limit. The dilution sensitivity limit is attempting to define a threshold above which an emission source would generate a concentration at Mace Head that would be discernible above the baseline noise. As this limit is based on the NAME dilution calculations it is independent of the gas species and therefore the same limit value can be used for both of the gases analyzed. The criteria are:

[15] 1. The total air concentration from the nine grid boxes centered on and surrounding Mace Head is less than ten times the dilution sensitivity limit, i.e., local emissions do not significantly contribute.

[16] 2. The total air concentration contribution from the European landmass is less than five times the dilution sensitivity limit, i.e., European emissions do not significantly contribute.

[17] 3. The contribution from the southerly latitudes is less than twice the dilution sensitivity limit indicating that southerly latitude air, with a different baseline concentration, is not significantly present. For both gases there is a strong hemispheric gradient because of disproportionately higher emissions in the Northern Hemisphere.

[18] In previous work the noise in the CH4 observations classed as baseline was ∼10 ppb. For a CH4 emission to be discernible at Mace Head, its impact there must be >∼10 ppb. In this work we have arbitrarily chosen 100 kt CH4 a−1 gridbox−1 (∼4% of the total UK CH4 emission) to be the minimum discernible emission, this equates to emissions of ∼2 μg m−2 s−1, 0.5 μg m−2 s−1 and 0.125 μg m−2 s−1 at 40 km, 80 km and 160 km grid box resolution, respectively, and delivers a dilution sensitivity limit of 3.4 s m−1. Choosing a larger (smaller) minimum discernible emission reduces (increases) the dilution sensitivity limit and thus increases (decreases) the resolution of the gridded output, but will reduce (increase) the emission sensitivity of the grid boxes. The chosen limit is arbitrary but sensitivity analysis, discussed in section 4.4, shows that the impact of this choice is small.

[19] Figure 2 shows a 3 month extract of the CH4 observations measured at Mace Head. The observations have been color coded to indicate whether, using the above classification, the air mass they were sampled from was considered baseline. For this baseline analysis all nonbaseline observations are removed.

image

Figure 2. Three month time series of Mace Head CH4 observations showing the impact of the baseline and nonbaseline classification. The baseline observations are shown in red.

Download figure to PowerPoint

4.2. Removing Outliers and Smoothing the Baseline

[20] The points defined as baseline using the above methodology still have a certain level of noise. The principle reasons for this are: emissions from the populated regions of the USA and Canada; unexpected emissions (for instance, forest fires in Canada or shipping); local emissions that are not identified using the above criteria; incorrectly modeled meteorology or transport (i.e., European air defined as baseline by error). Irrespective of the methodology used to identify these events some will inevitably be classed as baseline when it is inappropriate to do so. To capture such events the baseline data are statistically filtered, within a moving 40 d window, to isolate and remove these outliers that are actually nonbaseline observations. The observations removed through applying this statistical filter are shown in green in Figure 2.

[21] For each hour in the time series the remaining baseline points in a running 40 d window are fitted using a quadratic function and the value extracted for the current hour in question. The process is then advanced by an hour and repeated. Finally, the baseline time series is smoothed using the median value within a moving 20 d time window.

[22] The choice of the length of the time windows and number of points is arbitrary. It is a balance between using sufficient baseline points while still capturing the seasonal signal. Modest (±50%) changes to these values have little impact on the resulting baseline concentration.

4.3. Baseline Noise

[23] The noise or potential error in the smoothed baseline mixing ratio is estimated to be the standard deviation of the difference between the observations classed as baseline and the smoothed baseline mixing ratios at the corresponding times. Figure 3 shows, on a much expanded y axis compared to Figure 2, the typical spread of baseline observations about the smoothed continuous baseline estimate. For CH4 the baseline noise over 1989–2008 using ERAI is 8.7 ppb (assuming average 1860 ppb baseline for CH4 this is 0.47%) and over 1995–2009 using UKMO is 8.3 ppb (0.45%), the values for N2O are 0.25 ppb (0.08% assuming average baseline of 320 ppb) and 0.19 ppb (0.06%), respectively. If the same period (1995–2008) is analyzed, the noise values for ERAI and UKMO are closer still, with a 0.3 ppb difference for CH4 and a 0.002 ppb difference for N2O. This shows the reduction in magnitude from using ERAI over the entire period 1989–2008 to using UKMO is partially a result of the improved accuracy of the observations from 1994 onward rather than because of the different meteorology.

image

Figure 3. Observations of CH4 at Mace Head within a three month period classed as baseline (open squares) with the estimated daily baseline mixing ratios for the same period (crosses). Note: the y axis has been expanded compared to Figure 2.

Download figure to PowerPoint

4.4. Baseline Analysis

[24] Monthly baseline mixing ratios are estimated by averaging all of the hourly baseline values within the appropriate time window. A monthly value is estimated provided hourly values span at least two thirds of the month, this ensures the value is representative of the baseline concentration across the whole month. Using this methodology all months, from February 1989, are assigned baseline mixing ratios.

[25] The baseline analysis has been conducted with both sets of meteorology. The differences in the estimated monthly baseline mixing ratios are small, the mean difference in monthly baseline for CH4 is 0.1 ppb (σ = 0.93 ppb) and for N2O is 0.01 ppb (σ = 0.01 ppb), in both cases the baseline derived using ERAI is, on average, marginally higher. The monthly baseline mixing ratios from both analyses are shown in Figure 4.

image

Figure 4. Northern Hemisphere monthly baseline mixing ratios for CH4 (left axis) (solid line for ERA Interim (ERAI) and open squares for the Met Office's operational global NWP model, the Unified Model (UKMO)) and N2O (right axis) (dashed line ERAI and open circles UKMO).

Download figure to PowerPoint

[26] As a sensitivity test, the impact of adding 50% to the dilution sensitivity limit was investigated using the ERAI. The mean difference in the monthly baseline using the two different limit values was found to be negligible, 0.1 ppb (σ = 1.1 ppb) for CH4 and 0.01 ppb (σ = 0.02 ppb) for N2O. This is comparable to the difference because of the choice of meteorology.

[27] The annual growth rate on a particular day is defined as the current daily baseline value minus the previous year's value on the same day, e.g., growth rate for 14 January 1996 = (daily baseline for 14 January 1996) – (daily baseline for 14 January 1995). By averaging the annual growth rate values (one per day) within a running 12 month period (6 months either side of the day), a smoothed annual growth rate per day is estimated. Monthly averages of these growth rates are shown in Figure 5. Over the 20 year period the average CH4 growth rate is 3.9 ppb a−1 but with significant variability year to year (peaks in 1991, 1998–1999, 2003, 2007–2008). The N2O growth rate has been much more consistent at 0.7 ppb a−1 but still with variability, with peaks in 1990, 1994, 1996, 2000, 2003, 2005, 2007–2008 and troughs in between.

image

Figure 5. Northern Hemisphere monthly growth rates for CH4 (solid line ERAI and open squares UKMO) (left axis) and N2O (dashed line ERAI and open circles UKMO) (right axis). The solid horizontal line and dashed horizontal line indicate the zero growth level for CH4 and N2O, respectively.

Download figure to PowerPoint

[28] To illustrate the seasonal cycles of CH4 and N2O, the monthly baseline mixing ratios are detrended and presented in Figure 6 for the baseline derived from ERAI. The spread shows the extent of the year to year variability. CH4 has a peak mixing ratio in February-April (+9 ppb) and a minimum in July (−18 ppb). A similar cycle is shown for N2O, peak in March-April (+0.26 ppb) and trough in August (−0.41 ppb). However, the reasons for these seasonal cycles are different. For CH4, it is the loss through OH chemistry that drives the cycle with OH production at a maximum in August. For N2O, the seasonal cycle is due to the seasonality in; surface sources, transport between the Northern and Southern Hemispheres and transport of N2O-poor air from the stratosphere to the troposphere. The observed summertime minimum in tropospheric N2O in the Northern Hemisphere is reported to result from the propagation of springtime lower stratospheric minima down to the troposphere with a ∼4 month lag time [Liao et al., 2004; Nevison et al., 2004; Jiang et al., 2007].

image

Figure 6. Northern Hemisphere seasonal cycles of CH4 (dashed line, left axis) and N2O (solid line, right axis).

Download figure to PowerPoint

5. Regional Emission Estimates

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Observations
  5. 3. NAME Dispersion Model
  6. 4. Baseline Concentration Analysis
  7. 5. Regional Emission Estimates
  8. 6. Discussion
  9. 7. Conclusions
  10. Acknowledgments
  11. References
  12. Supporting Information

[29] By removing the time-varying baseline mixing ratios from the raw observation data, a time series of excursions from the baseline for each observed gas has been generated (1995–2009 using UKMO and 1989–2008 using ERAI). The observed deviations from baseline are averaged over each 3 h period. Negative perturbations are set to zero. For CH4 (N2O), the average percentage of negative perturbations per year was 38% (35%), but the values were generally small, median = −5.9 ppb (−0.1 ppb), and below the noise in the baseline estimate, 8.7 ppb (0.25 ppb). Because of the way the baseline is constructed, it is not surprising that around half of the baseline observations (∼45% of observations, see Table 1) lead to small negative perturbations. These observed deviations are assumed to be driven by emissions on regional scales that have yet to be fully mixed on the hemisphere scale. Henceforth these baseline-removed observations are referred to as simply the observations.

[30] Figure 7 shows the number of observations per year classed as European that are above 200 ppb for CH4 and above 4.5 ppb for N2O. Over the 20 year period the number of events for N2O has declined (linear regression: −0.14 a−1) but a similar trend is not seen in the CH4 observations (−0.01 a−1). From this basic analysis it would tentatively suggest that the regional N2O emissions have declined over time, or at least some significant sources have been removed, whereas the regional CH4 emissions have seen no observable change in their frequency.

image

Figure 7. Number of European pollution events per year above baseline for CH4 (gray) and N2O (black) that have exceeded a certain threshold (200 ppb and 4.5 ppb, respectively).

Download figure to PowerPoint

5.1. Inversion Methodology

[31] The observation time series, together with the NAME model output predicting the recent history of the air, were used to estimate the emission distribution of each gas over NWEU. The iterative best fit technique, simulated annealing [Press et al., 1992], was used to derive these regional emission estimates based on a statistical skill score (cost function) comparing the observed and modeled time series at Mace Head. The technique starts from a set of randomly generated emission maps, it then searches for the emission map that leads to a modeled time series at Mace Head that most accurately mimics the observations.

[32] The aim of the inversion method is to estimate the spatial distribution of emissions across a defined geographical area (e) using equation (1). The emissions are assumed to be constant in time over the analysis period. The observations are converted from volume mixing ratio [ppb] to air concentration [g m−3] using the modeled temperature and pressure at the observation point.

5.1.1. Inversion Domain and Grid Resolution

[33] The inversion domain is chosen to be a smaller subset of the full domain used for the air history maps. It covers 30°W–42°E longitude and 29.3°N–77.3°N latitude and is shown as the black box in Figure 1. The smaller domain covers all of Europe and extends a reasonable distance into the Atlantic. The inversion domain needs to be smaller to ensure recirculating air masses are fully represented but also because emission sources very distant from Mace Head have little discernible impact on the concentration at the station, i.e., the signal would be too weak to be seen. The inversion method assumes baseline concentration air enters the inversion domain regardless of direction. For the eastern and southern edges in particular this assumption will be incorrect. Emissions in Russia and around the Black Sea would be expected to elevate the atmospheric concentrations along the eastern edge, and because of the latitudinal gradient it would be reasonable to assume below midlatitudinal baseline concentration air enters from the south. This issue is overcome in the inversion by solving for but not analyzing the estimated emissions in any grid box on the edge of the inversion domain. It is assumed that the error of above or below baseline concentration air entering the domain will be absorbed into the solutions in these edge grid boxes. This assumption has been tested by comparing the solutions for the United Kingdom with this domain to those obtained when a significantly larger inversion domain is used. The results of this comparison are discussed in section 5.3.

[34] In order for the best fit algorithm to provide robust solutions for every area within the domain, each region needs to significantly contribute to the air concentration at Mace Head on a reasonable number of time periods. If the signal from an area is only rarely or poorly seen at Mace Head, then its impact on the cost function is minimal and the inversion method has little skill at determining its true emission.

[35] The contribution that different grid boxes make to the air concentration observed varies from grid box to grid box. Grid boxes that are distant from the observation site contribute relatively little to the observation, whereas those that are close can have a large impact. In order to balance the contribution from different grid boxes, those that are more distant are grouped together into increasingly larger blocks. The grouping varies for each time period considered and between the different gases because of varying meteorology and the impact of missing observations, respectively. The underlying horizontal grid resolution is 40 km (= x), which is equal to the resolution of the NAME output. The grouping creates blocks that have a resolution of x, 2x, 4x, 8x, 16x and 32x.

[36] An emission sensitivity level [g m−2 s−1] has been estimated for each gas. Below this level the impact of the emission at the receptor is assumed lost within the baseline noise of the observation. The dilution sensitivity limit threshold (3.4 s m−1), as derived in the baseline analysis, is used again. Baseline noise is defined as the standard deviation of baseline observations about the defined smoothed baseline value, as discussed in section 4. The values are 8.7 ppb (6 μg m−3) for CH4 and 0.25 ppb (0.48 μg m−3) for N2O using ERAI at standard temperature and pressure. The emission sensitivity for CH4 is thus defined to be 1.8 μg m−2 s−1 (∼90 kt a−1 for a 40 km grid box) and for N2O to be 0.14 μg m−2 s−1 (∼7 kt a−1 for a 40 km grid box).

[37] For each grid box the number of times it provides a contribution to Mace Head above the dilution sensitivity limit threshold is calculated from the dilution matrix. If this is below the minimum number required (arbitrarily defined as 240 3 h time periods, i.e., 30 d) within the 3 year period, then a grid resolution twice as coarse is considered. This process is repeated until the condition is satisfied or until the grid resolution is 32 times the original (32x) (Figure 8), where x = ∼40 km.

image

Figure 8. Example of the distribution of the different sized regions used by the inversion method to estimate regional emissions (1x by 1x through to 32x by 32x, x = ∼40 km).

Download figure to PowerPoint

5.1.2. Inversion Process

[38] The modeled time series at the measurement station is calculated by applying the current hypothesized emission map to the dilution matrix (equation (1)).

[39] The inversion process works by iteratively choosing different emissions, varying the emission magnitudes and distributions, with the aim of minimizing the mismatch between the observations and the modeled concentrations. No a priori conditions are set, with the exception that no emissions are allowed to be negative. The relative skill of a derived emission map is tested by comparing the modeled and observed time series by using a cost function that combines four different statistics. Other cost functions were investigated, for example; root mean square error, varying the weights applied to the different statistics; and, removing one of the four statistics, using the pseudo data experiment (see section 5.1.4) but were found to be less accurate at recreating a prescribed (known) emission distribution.

  • equation image

where

[40] r = Pearson correlation coefficient

[41] NMSE = Normalized Mean Square Error: equation image

[42] fac2 = Fraction within a factor of 2 of observations

[43] facNoise = Fraction of model values within Noise of the observations

[44] Noise = Standard deviation of baseline observations about the defined smoothed baseline value. Note observations below the Noise level are considered to have a magnitude equal to the Noise level with respect to the fac2 calculation.

[45] The iteration process is repeated until any potential improvement in skill in the emission map is estimated to be negligible.

[46] To simulate uncertainties in the meteorology, dispersion and observations a time series of random noise is applied to the observations. The random element is multiplicative and taken from a lognormal distribution with mean one and variance, arbitrarily, set to one fifth of the Noise. Any observations that are negative are considered to be zero.

[47] For each time period solved for, the whole inversion process is repeated multiple times (arbitrarily set to 52) to give a robust indication of the potential uncertainty in the emission solution, each time with a different random starting point and a different time series of random noise. Solutions are calculated for 3 year periods covering the period when observations are available. After solutions have been estimated for a particular 3 year period, the period is moved forward by 1 month and the process repeated, e.g., January 1995 to December 1997, February 1995 to January 1998.

[48] An annual estimate of emissions is calculated by averaging all of the solutions that contain a complete calendar year (1300 solutions) within the solved-for time period. The range for each year for each geographical region is calculated from the same sample of 1300 solutions and is taken as the 5th and 95th percentile.

5.1.3. Observations Used in the Inversion

[49] Any periods that were classed as baseline but were removed by the statistical filtering are removed from the analysis as these are considered to be unrepresentative of air from that sector. Times when the air is classed as “local” are likewise removed from the analysis. A 3 h period is classed as local if the contribution from the 9 grid boxes surrounding Mace Head is above fifteen times the dilution sensitivity limit. The local times represent periods when the emissions from the local area (120 km × 120 km area centered on Mace Head) would have a dominant effect on the observations. These are typically characterized by low wind speeds, low boundary layer heights and thus poor dispersion conditions. During such times the meteorological models used, with horizontal resolutions of between 40 and 80 km, are poor at resolving the local flows as they are dominated by subgrid scale processes, e.g., land-sea breezes. In all 86% (87%) of the CH4 (N2O) observations were retained for analysis using ERAI. A similar percentage is retained when using UKMO.

5.1.4. Pseudo Data Experiment

[50] To assess the ability of the inversion system to estimate correctly emissions on the regional scale, it was first applied to model derived pseudo observations. The carbon monoxide (CO) emissions from the European Monitoring and Evaluation Programme (EMEP) (www.emep.int) were used to calculate a model time series at Mace Head. Time series were derived using both ERAI and UKMO. The inversion system was tested using the ERAI dilution matrix applied to, first, the pseudo observations derived using ERAI and then to the pseudo observations derived using UKMO. The latter test investigates the ability of the inversion system to estimate emissions with a system that has errors. The impact of applying random noise to the system is investigated by solving with and without noise. Figure 9 shows the time series of emission estimates and true emissions for NWEU in this idealized case study.

image

Figure 9. Time series of annual emission estimates using model-derived pseudo observations. The gray columns are the European Monitoring and Evaluation Programme inventory values. The 5th, 25th, median, 75th and 95th percentiles is the case with no noise and the same meteorology. The solid line is the case with noise and the same meteorology. The dashed line is with noise and different meteorology (UKMO used to create pseudo observations and ERAI dilution matrix used within the inversion). The 5th and 95th estimates provide the uncertainty ranges in each case and are shown as thinner lines with the same style.

Download figure to PowerPoint

[51] The agreement between the median emission total and truth when no noise is added to the pseudo observations is excellent and the uncertainty bars are small. When noise is added to the measurements the median fit is still good but the uncertainty range is larger (solid lines). When a different meteorology is used to derive the pseudo observations and noise is added (dashed line) the fit is still good but the uncertainty range is at its largest. However, in every case the uncertainty range completely encompasses the true solution and gives confidence that the methodology is able to recreate the correct emission total within the estimated uncertainty range.

5.2. Results

[52] Figure 10 is an example of the observed and modeled time series of the air mixing ratio for CH4 for the first 3 months of 2006 at Mace Head. The modeled time series (gray) is a composite of all of the inversion solutions that cover this 3 month period. The magnitudes and patterns are similar and demonstrate that the inversion process is able to derive an emission map that produces a good match to the observations. Table 3 shows the average (minimum and maximum) statistical fit between the observed and modeled time series at Mace Head for each year. Figures 11 and 12 show examples of the average inversion emission maps for CH4 and N2O, respectively, for January 1990 to December 1992 (Figures 11, left, and 12, left) and January 2006 to December 2008 (Figures 11, right, and 12, right). The vast majority of the estimated emissions are land-based as expected, although for N2O, there are significant emissions estimated in the North Sea. The application of fertilizer onto crops can leach into rivers and thereby reach the sea, therefore, it is entirely plausible that a reasonable amount of N2O could be released from the sea rather than the land. The relatively large-sized grid boxes on the east of the United Kingdom also, in effect, smudge emissions over sea areas. This is why the geographical areas used to define UK and NWEU emissions extend out into the North Sea (Figure 13).

image

Figure 10. Time series of observed and best fit modeled CH4 mixing ratios (deviation from baseline) at Mace Head for the first three months of 2006 (gray equals composite of all NAME inversion solutions that cover this period, and black squares equal observations).

Download figure to PowerPoint

image

Figure 11. Average NAME inversion CH4 emission map estimates (ng m−2 s−1) for (left) January 1990 to December 1992 and (right) January 2006 to December 2008.

Download figure to PowerPoint

image

Figure 12. Average NAME inversion N2O emission map estimates (ng m−2 s−1) for (left) January 1990 to December 1992 and (right) January 2006 to December 2008.

Download figure to PowerPoint

image

Figure 13. Geographical areas used to define regional totals.

Download figure to PowerPoint

Table 3. Average (With Minimum and Maximum) Statistical Match Between Observations and Best Fit Modeled Time Seriesa
YearCorrelationNMSEbfac2cfacNoised
  • a

    Any 3 year period that covers the entire year in question is included in the analysis for that year.

  • b

    Normalized mean square error.

  • c

    Fraction within a factor of 2 of observations.

  • d

    Fraction of model values within Noise of the observations.

19900.81 (0.76–0.84)1.51 (1.17–2.19)0.79 (0.76–0.82)0.63 (0.60–0.67)
19910.81 (0.75–0.84)1.56 (1.17–2.50)0.80 (0.76–0.83)0.65 (0.60–0.69)
19920.81 (0.75–0.84)1.63 (1.21–2.50)0.82 (0.78–0.85)0.67 (0.64–0.72)
19930.82 (0.77–0.86)1.58 (1.04–2.31)0.84 (0.79–0.88)0.70 (0.65–0.74)
19940.83 (0.77–0.87)1.35 (0.94–2.20)0.84 (0.81–0.88)0.69 (0.64–0.74)
19950.84 (0.79–0.87)1.13 (0.87–1.68)0.83 (0.79–0.86)0.65 (0.61–0.71)
19960.83 (0.78–0.86)1.06 (0.83–1.52)0.81 (0.78–0.85)0.63 (0.59–0.67)
19970.82 (0.76–0.86)1.14 (0.83–1.70)0.81 (0.78–0.84)0.63 (0.59–0.67)
19980.82 (0.76–0.85)1.33 (0.91–2.01)0.83 (0.79–0.86)0.66 (0.61–0.72)
19990.82 (0.78–0.86)1.44 (1.10–2.06)0.84 (0.80–0.86)0.68 (0.64–0.72)
20000.83 (0.77–0.86)1.49 (1.10–2.12)0.85 (0.81–0.87)0.69 (0.67–0.72)
20010.83 (0.77–0.87)1.33 (0.84–2.12)0.84 (0.81–0.87)0.68 (0.63–0.72)
20020.84 (0.78–0.87)1.18 (0.84–1.92)0.84 (0.81–0.87)0.67 (0.63–0.70)
20030.84 (0.78–0.87)1.20 (0.84–1.74)0.84 (0.82–0.87)0.68 (0.64–0.72)
20040.83 (0.78–0.87)1.33 (0.94–2.10)0.85 (0.82–0.87)0.70 (0.66–0.73)
20050.81 (0.76–0.85)1.47 (1.07–2.10)0.85 (0.83–0.87)0.71 (0.68–0.73)
20060.82 (0.76–0.86)1.43 (1.02–2.04)0.85 (0.82–0.87)0.71 (0.67–0.73)
20070.83 (0.79–0.86)1.34 (1.02–1.94)0.85 (0.82–0.87)0.70 (0.67–0.72)

[53] Emission totals from specific geographical areas are calculated by summing the emissions from each 40 km grid box in that region (Figure 13). Emission estimates for N2O for the United Kingdom using both sets of meteorology are presented in time series graphs and compared to the reported UNFCCC values (Figure 14). Similar graphs (Figure 15) are shown for NW Europe with the United Kingdom removed (NWEU-noUK). Figures 16 and 17 show emission time series graphs for CH4. The NAME inversion annual emission estimates with uncertainties for the United Kingdom using ERAI for N2O and CH4 are given in Tables 4 and 5, respectively, and are compared to the UNFCCC inventory estimates.

image

Figure 14. Emission estimates for N2O for the United Kingdom. The annual NAME inversion results using ERAI (UKMO) are shown as a solid (dashed) line with uncertainty bars showing the median, 5th, 25th, 75th, and 95th percentiles. The United Nations Framework Convention on Climate Change (UNFCCC) inventory estimates for different sectors are shown as cumulative columns (from the bottom: energy, industry, agriculture, and waste).

Download figure to PowerPoint

image

Figure 15. Emission estimates for N2O for NW Europe with United Kingdom removed (NWEU-noUK). The annual NAME inversion results using ERAI (UKMO) are shown as a solid (dashed) line with uncertainty bars showing the median, 5th, 25th, 75th, and 95th (5th and 95th) percentiles. The UNFCCC inventory total estimates are shown as columns.

Download figure to PowerPoint

image

Figure 16. Emission estimates for CH4 for the United Kingdom. The annual NAME inversion results using ERAI (UKMO) are shown as a solid (dashed) line with uncertainty bars showing the median, 5th, 25th, 75th, and 95th percentiles. The UNFCCC inventory estimates for different sectors are shown as cumulative columns (from the bottom: energy, industry (very small), agriculture, and waste).

Download figure to PowerPoint

image

Figure 17. Emission estimates for CH4 for NW Europe with United Kingdom removed (NWEU-noUK). The annual NAME inversion results using ERAI (UKMO) are shown as a solid (dashed) line with uncertainty bars showing the median, 5th, 25th, 75th, and 95th (5th and 95th) percentiles. The UNFCCC inventory total estimates are shown as columns.

Download figure to PowerPoint

Table 4. N2O UK NAME Inversion Emission Estimates Using ERAI Compared to the UNFCCC Inventory Values Submitted in 2009a
Year5th25thMedian75th95thInventory
  • a

    Percentile values are rounded to nearest 5 kt a−1. NAME, Numerical Atmospheric Dispersion Modeling Environment; ERAI, ERA Interim; UNFCCC, United Nations Framework Convention on Climate Change.

1990125145160175200210
1991120140160175200210
1992110135150170195185
1993110130145160180175
1994110125135150165175
1995115125135140155175
1996115125135145155170
1997115130135145160175
1998110125135145165175
199995110125140160140
200090100110120135135
20018090100110130130
2002758595100115120
20038090100105115120
20048095100110125120
2005708090100120120
20065570758090115
20075565708090110
Table 5. CH4 UK NAME Inversion Emission Estimates Using ERAI Compared to the UNFCCC Inventory Values Submitted in 2009a
Year5th25thMedian75th95thInventory
  • a

    Percentile values are rounded to nearest 10 kt a−1.

1990151024603180401053504980
1991118022002980378051304940
199293016802480330046104870
199380014502140296041504720
199476014302090280038404390
199593015602180283038004340
1996111016702270290038304230
1997106017302300295038804000
1998106018202410308040503790
1999114019202620330044503540
2000118019702640332044703320
2001103017802450310040603040
200297016602270286037802900
2003104017702400303040702610
2004113020502750347046002520
2005122020202620329045402430
200693015602070265034502400
200780013901880240033002330

5.3. Sensitivities and Uncertainties in the Inversion Methodology

[54] There is no absolute method of defining the uncertainties associated with these best fit estimates. They occur because of errors in the inversion process, in the input meteorology, in the transport modeling in NAME and in the observations. This section discusses the sensitivities of the inversion results to (1) choice of meteorology, (2) domain size, and (3) a bias in the baseline concentration, and discusses other areas of potential uncertainty and how these have been minimized.

[55] The NAME inversion results using ERAI and UKMO are presented (Figures 1417). Although there are differences, within the uncertainty ranges there is good agreement between the solutions using different meteorologies. For CH4, the differences are all within the 25th–75th uncertainty range. For N2O there is a fairly consistent difference up until 2005, but always within the 5th–95th uncertainty range. In the early years of the comparison the UKMO results are lower than the ERAI results, in the latter years the reverse is the case. The reasons for this reversal are linked to the definition of the boundary layer height in the two models and how these have evolved over the years within the UKMO. Figure 18 compares the average monthly BLD modeled at Mace Head by the two meteorologies. The UKMO estimated BLD at Mace Head is on average lower than the ERAI estimated BLD however over the years the average difference between the two estimates has narrowed considerably. If this is generally the case over the whole domain the model with the lower BLD will enhance the modeled surface concentration for a given surface emission. This would then lead the inversions using UKMO to require lower emissions to model a given surface concentration and hence potentially explain some of the differences between the emission estimates using the two meteorologies. The periodic upgrades to the UKMO have meant that the location of the Mace Head station with respect to the meteorological grid has varied over the years which will mean that the amount of land to sea within the grid box will have changed and will alter the BLD estimates interpolated to the Mace Head station. Therefore without further analysis it is not possible to be definitive about the impact of the results shown in Figure 18.

image

Figure 18. Compares the average monthly boundary layer depth at Mace Head as calculated by ERAI with those calculated by UKMO (1995–1998 plus signs, 1999–2002 open squares, 2003–2005 solid squares, and 2006–2008 crosses).

Download figure to PowerPoint

[56] The inversion method assumes that only baseline concentration air enters the domain. This does not always hold true. It is assumed that above (below) baseline air entering the domain will lead to elevated (depressed) emission estimates only at the edge of the domain. Therefore the inversion domain solved for is significantly larger than the NWEU domain and it is assumed that the majority of these edge effects will not impact on the analysis of NWEU and UK emission totals. Figures 19 and 20 show the sensitivity of the NAME inversion UK estimates to a systematic bias in the baseline estimate, both positive and negative, and to using a larger domain for the inversion calculation. The baseline mixing ratio was first increased and then decreased by one standard deviation (σ) of the baseline noise. Each 3 year period (within 1989–2008) using ERAI was solved three times with no additional noise and the median annual estimate calculated as before. The red dash-dot line shows the impact of lowering the baseline and the red dashed line the impact of increasing the baseline. The estimates as presented in Figures 14 and 16 are also shown for comparison. For CH4 the changes in the baseline are comfortably within the 25–75th percentile uncertainty ranges, showing that a bias of this magnitude is captured by the uncertainty analysis. For N2O the addition or subtraction of a bias in the baseline has an impact that is sometimes larger than the 5–95th percentile uncertainty ranges. Therefore the N2O totals are susceptible to a bias in the baseline mixing ratio and this could lead to an increase of up to 30% in the UK uncertainty estimates. The colored lines in Figures 19 and 20 show the impact of increasing the domain (blue lines) as well as having a bias in the baseline mixing ratio (dashed and dash-dot blue lines). The results are similar to those when the standard domain is used (black) and demonstrate that the inversion solution is largely insensitive to the domain size chosen. For N2O the impact is to generally lower (on average by 10%) the UK estimates, there is no impact for CH4 outside the uncertainty ranges already presented.

image

Figure 19. Sensitivity of UK N2O NAME inversion emissions. Whisker plot equals same data as plotted in Figure 14. Blue (same domain, no noise) solid line: normal baseline; dash-dot line: low baseline (normal baseline −1σ); dash line: high baseline (normal baseline +1σ). Red (larger domain, no noise) solid line: normal baseline; dash-dot line: low baseline (normal baseline −1σ); dash line: high baseline (normal baseline +1σ).

Download figure to PowerPoint

image

Figure 20. Sensitivity of UK CH4 NAME inversion emissions. Whisker plot equals same data as plotted in Figure 16. Blue (same domain, no noise) solid line: normal baseline; dash-dot line: low baseline (normal baseline −1σ); dash line: high baseline (normal baseline +1σ). Red (larger domain, no noise) solid line: normal baseline; dash-dot line: low baseline (normal baseline −1σ); dash line: high baseline (normal baseline +1σ).

Download figure to PowerPoint

[57] All of the emissions are assumed constant in time within a given 3 year period and are geographically static. This is clearly a significant simplification. A sudden, but subsequently maintained, change in emissions will be picked up by solving multiple 3 year periods covering slightly different time periods, i.e., solving for a 3 year period and then advancing by 1 month. Enhanced emissions in any particular season, e.g., increased N2O emissions in spring following fertilizer application, will not be resolved. Intermittent or sporadic emissions will be averaged and their “true” contribution maybe lost.

[58] All areas of the domain are assumed to impact reasonably equally on Mace Head given the dispersion and transport processes modeled by NAME. The grouping of grid boxes together, so that each area contributes approximately equally to the observations, attempts to ensure this but clearly there will be some variability. Also large grid boxes could have significant variability actually within the grid box itself especially if there are significant orographic features within the grid box, e.g., the Alps, or land-sea boundaries. This may lead to errors if certain parts of the grid box are more frequently sampled than others. However, because of the large travel distances and therefore time elapsed between emission in these large grid boxes and measurement at Mace Head the impact of this will be small. Also by only reporting emissions within NWEU this issue is considered to be small.

[59] The inversion method makes no distinction between anthropogenic and natural sources and thus its estimates are for the combined total, making direct comparisons with the UNFCCC inventory difficult. From other studies the natural emissions in NWEU are estimated to be small in comparison to the anthropogenic emissions. For CH4 the natural emissions in NWEU are estimated to be 240 kt a−1 [Bergamaschi et al., 2005], less than 10% of the total emission.

[60] It is also important to recognize that the atmospheric release of N2O from agricultural practices, such as nitrogen fertilizer application, may occur at some considerable distances from the point of application because of transportation by rivers. This adds to the uncertainty of using the maps to attribute emissions to particular regions. The area considered to be the United Kingdom includes the waters directly surrounding the United Kingdom (Figure 13) and so the impact of this is considered to be small for the United Kingdom. This would be problematic if the individual contributions of Belgium or The Netherlands for example were presented and is the reason why only the NWEU total is considered. The most significant region in relation to this issue is the border between Northern Ireland and The Republic of Ireland, however because of the proximity to Mace Head and the corresponding high resolution of the output in that area, the impact is small.

[61] The transport modeling and thus the inversion algorithm also assume that the loss processes associated with each gas are negligible within the regional domain. Given the atmospheric lifetimes of both of the gases studied here, 8.7 years (CH4) and 114 years (N2O) [Denman et al., 2007], this is considered to be a robust assumption.

6. Discussion

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Observations
  5. 3. NAME Dispersion Model
  6. 4. Baseline Concentration Analysis
  7. 5. Regional Emission Estimates
  8. 6. Discussion
  9. 7. Conclusions
  10. Acknowledgments
  11. References
  12. Supporting Information

[62] Considering the N2O results, Figure 14 and Table 4 show the inventory and inversion estimates for the United Kingdom throughout the whole 18 year period. The inventory estimates are reported to have an uncertainty of 252% (www.unfccc.int, UK National Inventory Report, 2010) and estimate there is a 95% probability the UK N2O emissions in 2008 were between 33% and 73% below the level in 1990. The inversion results are consistently lower than the inventory estimates (20–50 kt a−1). However, the trend comparison is excellent, the inversion results estimate there is a 95% probability that the UK emissions in 2007 are between 34% and 70% below the level in 1990, and are estimated to have fallen by 5.1 kt a−1 compared to the inventory rate of fall of 5.8 kt a−1. The step change in the emission total in 1999 appears in both estimates although smoothed in the inversion estimates because of the 3 year averaging period. In the inventory this step is related to the application of an abatement technology at an adipic acid plant in the northeast of England, a similar technology was also fitted to an adipic acid plant in Germany in 1997. The agreement between the inversion estimates and the inventory would imply that the reported reduction in emissions because of the new UK technology was achieved. The results for NWEU with the United Kingdom removed (NWEU-noUK) (Figure 15) show the inversion estimates are again consistently lower than the inventory estimates (60–240 kt a−1). Both estimates show an overall reduction in emissions, but the inversion results show a significant rise in emissions over the period 1990–1995 that is not seen in the inventory.

[63] The results for CH4 for the United Kingdom (Figure 16 and Table 5) show a more significant difference. The NAME inversion estimates show a modest decline 1990–1993 but since have remained approximately constant, within the uncertainty ranges. The NAME inversion UK emissions in 2007 are estimated to be 41% (median) below the level in 1990, but with a very large spread (25th to 75th percentile range is 16% to 60%, and the 95% probability range is 83% below to 78% above). The UNFCCC inventory estimates are reported to have an uncertainty of 23% (www.unfccc.int, UK National Inventory Report, 2010), estimate there is a 95% probability the UK CH4 emissions in 2008 were between 50% and 56% below the level in 1990 and show a marked and monotonic decline over the whole period. There is a significant difference in the trend in the emission estimates, the UNFCCC inventory estimates have decreased by 170 kt a−1 compared to a modest decrease of 22 kt a−1 in the NAME inversion estimates. The agreement over the period 2000–2008 is good, the inventory estimates are within the 25th–75th NAME inversion uncertainty range for each year. The estimates for NWEU-noUK show reasonable agreement across the whole 20 years. The median NWEU-noUK results are higher (380–3770 kt a−1) than the UNFCCC estimates but the overall trend agreement is good. It is possible that in the inversion estimates some of the actual UK emissions are incorrectly assigned to a grid box outside of the UK domain (Figure 13) and highlights the challenge of estimating emissions from just one site. A similar site, or sites, on the east coast of the United Kingdom would enable a much clearer distinction between UK and non-UK sources. This may partially explain the rise in the NWEU-noUK total in 1991–1993 and the corresponding dip in the UK total in the same years. It is also notable that the uncertainty ranges for CH4 are significantly larger than those for N2O reflecting the larger baseline noise for CH4.

[64] The inversion results are consistent with the initial and crude analysis of the observations (Figure 7), i.e., a relatively flat UK CH4 emission trend and a declining emission trend for UK N2O.

[65] The comparison in the trend in emissions for the United Kingdom for CH4 and N2O are strikingly different. However, for both gases the inversion methodology was identical. Also because the agreement with the independent NOAA observational record for both gases at Mace Head has shown strong consistency over the entire period, it is unlikely that the comparisons are significantly affected by a difference in the accuracy of the observations. Other possible explanations could be a significant difference in their respective emission distributions or the relative impact of emission intermittency. However, both gases are primarily released from diffuse sources, e.g., agriculture, landfill. The intermittency of significant N2O emissions following fertilizer application will certainly impact on the observations and hence the inversion results and it is difficult to directly quantify this affect. The long (3 year) inversion periods have been chosen to minimize this impact.

[66] The UK UNFCCC CH4 estimates show a significant decline (2364 kt) between 1990 and 2003. This is dominated by a reduction of emissions from solid waste disposal on land (landfill) of 1320 kt (waste sector) and from fugitive emissions from solid fuel (coal mining) of 610 kt (energy sector). These two source categories account for 82% of the reduction in the reported emissions. The reduction in the median UK CH4 inversion estimates between 1990 and 2003 amount to 780 kt, albeit with significant uncertainty, a third of the reduction declared to the UNFCCC. Therefore the difference in the reduction in UK CH4 emissions between the two methods is 1590 kt.

[67] It is important to understand the reason for this mismatch in the UK CH4 estimates. Could a significant proportion (66%) of the reported emission reduction in the UK landfill and coal mining have already been achieved by 1990? This would fit with the inversion estimates. Or could the reduction in the emissions from landfill and coal mining have been partially compensated for by an increase in a different unreported source(s)? This is unlikely because these significant emissions would still have to be present today and would not fit with the present-day good match between the two emission estimates. Was the reduction in landfill emissions not achieved? This could be either that the landfill emission reduction methods were not as effective or that the landfill were not emitting as much as estimated in the early 1990s. The former would imply that the current inventory is still underreporting which is not supported by the inversion estimates. The latter explanation would be more consistent with the inversion results and support the argument that pre-1990 emission reductions in waste/coal were more effective than estimated in the inventory. The magnitude of the difference between the inventory and the inversion estimates is more than twice the reduction reported for coal mining alone hence it is unlikely that this is the main reason for the difference although it could be a significant contributing factor if it was in error.

[68] Another possibility is that the UK CH4 inversion estimates and more importantly the emission trend is significantly in error. However, the same methodology was applied equally to CH4 and to N2O so it is unlikely that only one of the trends is correct. The meteorology used was consistent throughout the time period so this is unlikely to be the cause of an error in the trend. The measurement instrument was changed in February 1994 but the difference between the inventory and inversion results were at their peak in 1993–1995 and remained very significant (outside 75th percentile range) until 2000 and so this is unlikely to be a significant cause for the mismatch. Other inversion estimates for the United Kingdom for 2001 [Bergamaschi et al., 2005] show higher UK emissions (4200 kt a−1) than found in this study. However, more recent work by this group [Bergamaschi et al., 2010] revised their emissions for the United Kingdom down to a value more consistent with both this study and the current inventory value for 2001.

[69] All of these estimates were performed using data from just one measurement station on the west coast of Ireland. The inversion estimates would be improved if more stations, each making comparable quality observations, were included in the analysis. Not only would this add more data, but crucially, enable a better distinction between the United Kingdom and Continental European emissions, especially if some of the additional sites were on the east coast of the United Kingdom. Unfortunately additional observations at other stations in Europe were only started, at the earliest, in the late 1990s. Including other stations midway through the analysis period would undermine the consistent approach used throughout the time series in this study, but would be a worthy exercise to undertake in future work.

7. Conclusions

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Observations
  5. 3. NAME Dispersion Model
  6. 4. Baseline Concentration Analysis
  7. 5. Regional Emission Estimates
  8. 6. Discussion
  9. 7. Conclusions
  10. Acknowledgments
  11. References
  12. Supporting Information

[70] This paper describes a method for estimating the midlatitude Northern Hemisphere baseline mixing ratio from observations recorded at the remote Mace Head, Ireland measurement station. The baseline trends of CH4 and N2O from 1989 until 2009 are presented. They show that the abundance of N2O has consistently increased in the atmosphere with an average growth of 0.7 ppb a−1. The CH4 mixing ratio has risen overall and an average growth rate of 3.9 ppb a−1 is reported. However, with both gases there is significant year-to-year variability in the growth rates, with CH4 having seen several periods of negative growth. Both gases have significant seasonal cycles with sharp summer minima and broad spring maxima. The CH4 minimum and maximum occur a month earlier than that seen in the N2O record.

[71] An inversion method is described that is shown to accurately reproduce regional emission totals given, so-called, modeled pseudo observations, even when an independent meteorology is used. Annual inversion emission estimates for the United Kingdom and NWEU (United Kingdom removed) for CH4 and N2O are presented for each year, 1990–2007, using, where available, two different meteorological data sets. The comparison between the estimates derived using both meteorologies is good and the results using UKMO consistently fall within the estimated uncertainty ranges of the ERAI results. The inversion results have been compared to the country values reported to the UNFCCC process.

[72] The results indicate reasonable agreement for the United Kingdom for N2O over the entire period. The median inversion estimates are on average 22% lower than the inventory estimates. For CH4 for the United Kingdom the median NAME inversion estimates show a decline of 24% in emissions from the 3 year period 1990–1992 to the period 2005–2007, compared to a drop of over 50% in the reported inventory values. The uncertainties in both methods are significant but the different trends are striking. The agreement in the trend for NW Europe (United Kingdom removed) for both gases is reasonable, on average the inversion results for N2O are 25% lower than inventory estimates and for CH4 21% higher. The drop in the UK CH4 inventory values is reported to be dominated by a reduction in landfill and, to a lesser extent, coal mine emissions. The UK CH4 estimates agree, within the 25th–75th percentile model uncertainties, from 2000 onward. The agreement of the N2O estimates and the later agreement of the CH4 estimates points to the inversion methodology being capable of estimating UK emissions. It would suggest that the landfill and maybe the coal mine emissions in the early 1990s may have been overestimated in the reported inventory values. This is particularly relevant as 1990 is used as the base year on which to calculate UK reductions in emissions to comply with the UK Kyoto commitments, set at 12.5% within the compliance period 2008–2012. If the median UK inversion estimates for 1990 and 2007 for N2O and CH4 are used to replace the inventory values in calculating the total GHG reduction in emissions from 1990 then the United Kingdom has achieved a drop of 14.3% by 2007 compared to the reported value of 17.3%. Therefore the United Kingdom would still have complied with its commitment to the Kyoto Protocol but, if the NAME inversion estimates are assumed, by a smaller margin.

Acknowledgments

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Observations
  5. 3. NAME Dispersion Model
  6. 4. Baseline Concentration Analysis
  7. 5. Regional Emission Estimates
  8. 6. Discussion
  9. 7. Conclusions
  10. Acknowledgments
  11. References
  12. Supporting Information

[73] We gratefully acknowledge the efforts of Gerry Spain at the Mace Head observing station and the Physics Department, National University of Ireland, Galway, for making the research facilities at Mace Head available. The operation of Mace Head, an AGAGE station, was supported by NASA (grants NAGW-732, NAG1-1805, NAG5-3974, NAG-12099 to MIT, and NAGW-2304, NAG5-4023 to SIO), the Department of Energy and Climate Change (United Kingdom) (contracts PECD 1/1/130 and 7/10/154, EPG 1/1/82, EPG 1/1/130 to International Science Consultants, CPEA 27, GA0201, CESA 0702 to the Met Office, United Kingdom, and GA01081 to the University of Bristol). The authors would also like to thank the referees for their helpful and constructive comments.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Observations
  5. 3. NAME Dispersion Model
  6. 4. Baseline Concentration Analysis
  7. 5. Regional Emission Estimates
  8. 6. Discussion
  9. 7. Conclusions
  10. Acknowledgments
  11. References
  12. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Observations
  5. 3. NAME Dispersion Model
  6. 4. Baseline Concentration Analysis
  7. 5. Regional Emission Estimates
  8. 6. Discussion
  9. 7. Conclusions
  10. Acknowledgments
  11. References
  12. Supporting Information
FilenameFormatSizeDescription
jgrd16751-sup-0001-t01.txtplain text document1KTab-delimited Table 1.
jgrd16751-sup-0002-t02.txtplain text document1KTab-delimited Table 2.
jgrd16751-sup-0003-t03.txtplain text document2KTab-delimited Table 3.
jgrd16751-sup-0004-t04.txtplain text document1KTab-delimited Table 4.
jgrd16751-sup-0005-t05.txtplain text document1KTab-delimited Table 5.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.