This report describes work carried out over the period of the contract EPG 1/3/98 in four broad areas: model development; source attribution; scenarios and mapping; other outputs.

The aim of the contract was to develop the existing UK capability for modelling the transport, transformation and deposition of acidifying and/or ozone generating pollutants and the assessment of ecosystem damage using critical loads and levels and to use the model to inform UK policy development.

The period covered by this contract saw further work on HARM (Hull Acid Rain Model) and the development of a new model ELMO (Edinburgh-Lancaster Model for Ozone) for making estimates of peak ozone concentrations across the UK.

The development of a 10 km implementation of HARM was started under the previous contract (EPG 1/3/64); this was complemented by a major restructuring of the model code to reduce run times. The provisional 10 km model used existing emissions inventories (except 20 km UK NH₃ and 150 km EMEP area emissions), but ran to the 3064 receptors of the 10 km grid and used a 10 km rainfall field. Subsequently, EMEP 50 km emissions (for 1994) and a 10 km NH₃ inventory for GB were incorporated. The standard number of trajectories was increased from 24 to 72. A single-source (footprint) version of HARM was developed to run with 360 trajectories to overcome the 'spiders web' effect which had been noted previously. The DMS emissions which had been treated in the model as SO₂, were converted back into DMS to allow further exploration of the effects of DMS emissions. In summer 1997 the outputs of the new 10 km were compared with those from the 20 km code and are summarised in Tables 1 and 2 below. It should be noted that at this stage, the model still used 20 km deposition velocities (resampled at 10 km). The new vg values were not available until the summer of 1998 (HARM11.2).

Table 1. Comparison of grid cell deposition estimates in the original 10 km version of HARM with the previous 20 km version

As might be anticipated, the 10 km model yielded a wider range of values for both wet and dry deposition, although the means were similar. The budget figures for the UK (non-area weighted) were slightly different with the 10 km model giving more wet deposition and less dry deposition than the 20 km version (Table 2).

Table 2. UK deposition budgets for the original 10 km version of HARM and the previous 20 km version

For validation purposes, output from the 10 km model was compared with measured concentrations of SO₂ and NO₂ at the rural network sites and deposition compared with RGAR (RGAR, 1997) estimates for those grid cells containing precipitation composition monitoring sites. This validation process continued as the 10 km model evolved. Further changes in the 10 km model included: updated shipping emissions based on the Lloyds Register study; the incorporation of 10 km vgs for GB (supplied by ITE Bush) for HARM11.2; use of 1995 EMEP area emissions and 1995 emissions for the UK for all pollutants except HCl and NH₃. As specific values for deposition velocities for Northern Ireland were not available, the average of the GB values was (and still is) applied. A coding error relating to the reading of EMEP NH₃ emissions were corrected. These changes affected some of the model scenarios (see below). Some changes were also made to the definition of the edge of the UK grid. The model version at the end of this process was HARM11.4. Output from HARM11.4 was compared against the network data and the significance of the change in model scale in terms of CL exceedance considered. A paper covering these issues has been submitted for publication.

HARM11.4 continued to underestimate dry NH-N deposition by a significant amount. A range of strategies was adopted to address this and are discussed in more detail below.

The change from a 20 km version of HARM to a 10 km version, which increased the number of receptors in the standard model run from 843 to 3064, had a deleterious effect on the model's run time. On a Sun IPX the 20 km model ran in 2 hrs 30 mins, while the original 10 km code took 16 hrs. It was clear that this was unsatisfactory in the light of the fast turn round sometimes required for policy advice, and that the model code needed a thorough review. Two strategies were applied: compiler optimisation and the development of new source code. This work was carried out in collaboration with Dr. Mike Mineter (Department of Geography, University of Edinburgh) using the SunSPARC10 as the platform for development. A detailed report was compiled (Mineter, 1997) and only the key elements are summarised here. The original model used the default options when the executable code was compiled. Using the "-O2" global optimisation flag set in the SunPro Pascal SPARCompiler improved the run time by 30%. The second stage was to improve the efficiency of the code itself. An important element of this was to reduce the number of calculations required. In the original code, calculations were made through several nested loops. Moving calculations to outer loops, where possible, resulted in a significant reduction in the number of calculations, for example in calculating the trajectory paths to each of the receptors. Other changes included the use of Pascal records rather than multiple arrays, replacing a number of variables by constants and declaring variables locally within particular procedures, rather than in the module header. The improvements in runtime which resulted from these changes are summarised in Table 3.

Table 3. Model run times on a SPARC10

This optimisation of the HARM code represents the most significant change to the code since it was originally developed by Dr. Dick Derwent in the mid-1980s. The model now comprises two distinct elements: a core of PASCAL code running the chemistry and a C+ shell controlling model input and output options. At the end of the process, the 10 km model was running as fast as the 20 km version, which was felt to be a significant achievement. Apart from the additional speed, the main advantage of the new structure is that it allows many model inputs to be modified and output options selected without editing the main body of the code, thus minimising the risk of inadvertant changes.

As well as the code changes outlined above, the platforms on which HARM and ELMO are run have changed. SunULTRA 1 and ULTRA10 workstations have been purchased through the contract. The change in hardware has also brought about significant improvements in model run time.

It is apparent that models, such as HARM, which assume instantaneous mixing through an air parcel will have difficulties in estimating ground level concentrations of pollutants from low level sources. The majority of UK NH₃ emissions come from cattle and other livestock (RGAR, 1997) which HARM then mixes instantaneously through a boundary layer depth of 800 m. As a result, HARM has been unable to reproduce NH₃ concentrations measured by the old diffusion tube network or estimated using more specialised ammonia models such as FRAME (Singles et al., 1998) (Figure 1). Low modelled gas concentrations resulted in low estimates of dry deposition of NH-N, although HARM was able to provide a reasonable representation of wet NH-N deposition. This issue was discussed in the last RGAR report (RGAR, 1997). Model underestimation of NH₃ may be exacerbated by the failure of currently adopted NH₃ emissions inventories to reflect the true magnitude of emissions. As N forms an increasing proportion of deposited acidity (reflecting the significant reductions in SO₂ emissions over the last decade), so it became imperative to address HARM's representation of NH₃.

The initial approach to changing HARM was to try scaling deposition velocities for NH₃ to increase the efficiency of removal from the air parcel. In addition to changing the deposition velocities, GB emissions were also varied over a range believed to be feasible given previous estimates. A report on this work was supplied to DETR. Taking the current GB NH₃ emission as correct, the results that came closest to those based on FRAME and reported in RGAR (1997) were found by scaling UK vgs by 16.

Following this exercise it was apparent that the model was still underestimating dry NH-N deposition and that the spatial distribution of the deposition field did not provide a good match against that estimated by FRAME. Alpha factors (in square deposition) were used to enhance dry deposition of a number of different pollutants in early versions of both the EMEP model and of HARM which had large grid cells, but had been abandoned as grid size was reduced. Dr Mark Sutton (ITE Bush) used FRAME to estimate alpha factors across the 10 km grid which were then implemented in HARM11.5. Changes in the HARM modelled reduced N budget between the original model and the latest version (11.5) are summarised in Table 4.

Table 4 UK reduced N budgets in k tonnes. HARM figures based on a GB emission of 279 k tonnes NH₃ and using 1992-95 rainfall

The effect of implementing the alpha factors on HARM's ability to reproduce our best estimate of NH-N dry deposition is illustrated in Figure 2, where it is compared with output from FRAME and HARM using the scaled deposition velocities.

HARM11.5 has been used for scenario work since the summer of 1999 (see below). The emissions currently employed in HARM11.5 are 1995 for UK SO₂ and NOx, 1996 for UK NH₃ and 1992 for UK HCl. All EMEP area emissions are for 1995. Mapped outputs from HARM10.4, the original 10 km version of HARM and HARM11.5 for total S, oxidised and reduced N deposition are compared in Figure 3a) to c).

The development of a month-by-month version of HARM described in the report from EPG 1/3/64 has continued, with a comparison of modelled gas concentrations against data from the rural SO₂ and NO₂ networks. Unfortunately, it has still not been possible to assess model performance against monthly deposition data as these have not been available. At present, wet deposition is not calculated on a monthly basis. Dry deposition could be estimated in this way and has been requested from ITE Bush. The model output suggests significant differences in the seasonal behaviour of S and oxidised N and it is felt that getting data to test the model against would still be valuable.

In the light of the negotiations which finally led to the signing within the UNECE of the 2^nd NOx protocol (Gothenburg protocol) in December 1999 and the development of strategies within the EC for acidification, eutrophication and ozone, it was agreed that HARM should be developed to include O₃ precursors. The standard version of HARM fixes the initial concentration of ozone at 30 ppb. Ozone is consumed at each model time step in the oxidised N chemistry, it is also replenished to reflect transfer from the stratosphere. HARM does not, however, represent the mechanisms of formation of low level ozone.

The first stage (carried out in the autumn of 1997) was to integrate a section of code from the BACKIT model (Johnson and Derwent) in to the HARM code. This model included major NM-VOCs and CO and was called OZ-HARM. Early in 1998 the name ELMO (Edinburgh-Lancaster Model for Ozone) was adopted. All emissions (except UK NH₃ and HCl) were updated to 1995, with 1989 isoprene emissions across the EMEP area supplied at 150 km scale by Dave Simpson (these were re-sampled for a 50 km grid). Isoprene across the UK was allocated to those grid squares where forest dominated. In common with HARM, emissions from shipping sources were scaled up following the Lloyds Register report. UK NM-VOC emissions were scaled up to address apparent underestimation reflected by comparison with data from the monitoring networks. The NM-VOC emissions supplied are not speciated, but within ELMO the total is allocated to nine major VOCs: methane; ethane; propane; butane; ethylene; propylene; isoprene; toluene and o-xylene. These represent a mix of high, medium and low reacting species whose degradation, through oxidation, can be taken to represent ozone formation. The oxidation is driven by modelled concentrations of the free radicals OH and HO₂. ELMO is set up to estimate peak ozone concentrations by setting photolysis rates for cloud free conditions in July at the latitude of the UK. The values selected are realistic for surface albedo, stratospheric ozone and background aerosol loading. The arrival time for the ELMO trajectories is set to mid-afternoon.

In early 1999 a national wind-rose weighting was adopted in ELMO that was based on data from Heathrow for days when O₃ concentrations exceeded 50 ppb. This was felt to provide a reasonable representation of wind direction under high ozone conditions. This model (ELMO2.4) was run to the UK's rural ozone monitoring sites for validation purposes and a series of model runs were carried out cutting UK and EMEP area NOx and VOC emissions to assess their impact on ozone concentrations. In the summer of 1999 S emissions (including DMS) in ELMO were coupled in to the ozone chemistry to allow the model to be used to estimate concentration of secondary SO₄ aerosol (PM₁₀) (ELMO2.5). ELMO has been compared against peak ozone concentrations from the monitoring networks in 1995 and shows that although the model underestimates, the general pattern of ozone concentrations across the UK is reasonable. A series of model runs has also been carried out to identify the threshold between NOx and VOC limitation (following Sillman et al., 1997). This is based on plotting O₃/NOy-NOx (where NOy-NOx = NOz) on the x-axis, against the change in O₃ concentration between the base case and a scenario run, on the y-axis. Based on Sillman et al's analysis, the boundary between NOx and VOCs sensitive chemistries lay at an O₃/NOz ratio of between 8 and 10.

A Sillman plot for an easterly trajectory across the UK is shown in Figure 4 and shows a similar threshold for NOx vs VOC sensitivity as that identified by Sillman et al.

As ELMO estimates peak ozone concentration, it has not been appropriate to compare it against STOCHEM (Collins et al., 1997), but work is in progress to compare ELMO running on an east-west trajectory to a receptor site in Hampshire, against UK-Photochemical Trajectory Model (PTM) (Derwent and Davies, 1994; Derwent, undated).

By establishing empirical relationships between ozone concentrations at the rural monitoring sites and EPAQS and WHO standard exceedances, AOT40 and AOT60, it has been possible to use ELMO to estimate these values under a range of future emissions scenarios.

A detailed description of ELMO, its validation, the Sillman plots and scenario analysis is being prepared for publication in Atmospheric Environment.