The Effects of Uncertainty in Deposition Data on Predicting Exceedances of Acidity Critical Loads for Sensitive UK Ecosystems Part 1

Introduction

Acidification can lead to harmful environmental effects, including depletion of fish stocks from lakes and streams, soil degradation and deforestation due to phytotoxi-city (Rodhe et al., 1995). The main sources of sulphur emissions, that in turn lead to sulphur deposition, are the burning of fossil fuels, for example, from coal-burning power stations and other stationary sources (e.g. industry). Oxidized nitrogen emissions can also originate from the same sources as sulphur, but additionally come from vehicle exhaust fumes. Both of these pollutants can contribute to acidification and their impact is commonly assessed relative to a critical load.

The critical load concept is based on the maximum load of pollutant deposition that an element of the environment can tolerate without harmful effects occurring. The amount of excess deposition above the critical load is called the exceedance. Within Europe, protocols controlling the emissions of sulphur and nitrogen are being implemented and further reductions in these pollutants have an associated cost. Consequently, it is imperative that uncertainties in the calculation of exceedances are quantified before further reductions in pollutants can be justified fully. The further development of methods to assess the uncertainties in critical load exceedance calculations is one of the priority tasks identified by the UN/ECE Convention on Long-range Transboundary Air Pollution Working Group for Effects. The Task Force on Integrated Assessment also requires this information for inclusion in their assessments and optimization studies on the European scale. Policy makers both within the UK and within Europe will use this uncertainty information.

Ecosystems differ in their sensitivity to acidification. The spatial distribution of ecosystems at risk from acidification is often derived using the critical loads approach. For the purposes of this topic a critical load may be defined as ‘a quantitative estimate of an exposure to one or more pollutants below which significant harmful effects on specified sensitive elements of the environment do not occur according to present knowledge’ (Nilsson and Grennfelt, 1988, p. 8). Critical loads have formed the basis for national and international deliberations on the control of atmospheric emissions and depositions of sulphur and nitrogen. The excess deposition above the critical load, the critical load exceedance, is used to quantify the potential harmful effects.

It should be noted that the empirical and mass balance methods on which national critical loads are based, define long-term critical loads for systems at steady-state. Therefore, exceedance is an indication of the potential for harmful effects to systems at steady-state. This means that current exceedance of critical loads does not necessarily equate with damage to the ecosystem.

Previously, critical load exceedances have been calculated as deterministic estimates. However, each variable in the calculation of critical load and exceedance has some uncertainty associated with it. Data inaccuracy may arise from numerous sources and may generally be divided into two main categories depending on their characteristics: precision or bias. Until recently exceedance calculations in the UK have not accounted for these inherent uncertainties in input parameter values.

The uncertainty of exceedance values is a product of many factors. This topic outlines the results of a preliminary study undertaken to quantify the variation in acidity critical load exceedance caused by the uncertainty in sulphur and nitrogen deposition values only. The additional effect of uncertainty in critical load values themselves has not been considered here. The uncertainty analysis involved a fixed value analysis where deposition was varied by a fixed amount and a Monte Carlo analysis where deposition was varied randomly. Both types of analysis used the same two baseline deposition scenarios. The response of the exceedance calculation to variations in sulphur alone, nitrogen alone and sulphur and nitrogen together was investigated (i.e. a sensitivity analysis). Nitrogen and sulphur were varied together to investigate any synergistic effects.

Exceedance predictions have been measured in two ways. Firstly, the total area of sensitive ecosystems for which the critical load was exceeded (i.e. exceeded ecosystem area) in each grid square. Sensitive ecosystems are defined as semi-natural and natural upland and lowland ecosystems that are sensitive to acidification and/or eutrophication. Secondly, the accumulated exceedance (AE). This takes account of both the exceeded ecosystem area and the magnitude of exceedance and is defined as:

The aims of the study reported in this topic were to quantify the effect of uncertainty in nitrogen and sulphur deposition on the exceeded area and AE of sensitive UK ecosystems. Additionally, the study sought to investigate whether there is a linear relationship, as suggested in a previous study, between deposition uncertainty and the area of critical load exceedance or accumulated exceedance. In doing this, consideration was given to the relative sensitivity of critical load exceedance estimation to sulphur or nitrogen deposition uncertainty. Two deposition scenarios were studied and compared; one for the period 1995-97 and a predicted one for 2010.

Methods

Currently in the UK, critical loads data are calculated for a specified set of ecosystems. These are acid grassland, calcareous grassland, heathland, coniferous woodland and deciduous woodland. In addition, acidity critical loads have been defined for 1445 freshwater sites throughout Great Britain. The deposition data are total nitrogen and sulphur values mapped for every 5 x 5km grid square of the UK. The exceedance values are then calculated for every 1 x 1 km grid square of the UK. These data can be aggregated to a 5km spatial resolution or summed across grid squares to give national statistics. It should be noted here that the data are at two spatial resolutions, which itself introduces uncertainties. The comparison of data obtained with different spatial resolutions can be a major problem when attempting to measure spatial variation. This should be explored thoroughly in future work. However, Smith et al. (1995) have already simulated 1km deposition from 20km data (the scale of the deposition data current at the time of their work) and showed that the uncertainty introduced by using 20 km scale estimates of deposition is small. Smith et al. (1995) found that the major problem was the uncertainty in estimates of deposition, the issue addressed in this topic.

Input data

Critical loads data

To examine the acidifying effects of both sulphur and nitrogen deposition simultaneously, the critical loads function (CLF) was developed in Europe (Posch et al., 1999; Posch and Hettelingh, 1997; Posch et al., 1995; Hettelingh et al., 1995). The CLF defines separate acidity critical loads in terms of sulphur and nitrogen and compares them with sulphur and nitrogen deposition.

The intercepts of the CLF on the sulphur and nitrogen axes define the ‘maximum’ critical loads of sulphur and nitrogen (Figure 12.1). The maximum critical load of sulphur (CLmax(S)) is the critical load for acidity expressed in terms of sulphur only (i.e. when nitrogen deposition is zero). Similarly, the maximum critical load of nitrogen (CLmax(N)) is the critical load of acidity expressed in terms of nitrogen only (when sulphur deposition is zero). CLmin(N) is the deposition independent critical load of acidity solely due to nitrogen removal processes in the soil (nitrogen uptake and immobilization, Figure 12.1). These ‘minimum’ and ‘maximum’ critical loads are calculated for the sensitive ecosystems defined above (i.e. not for the whole of the UK) (Hall et al., 2001a). The total area of sensitive ecosystems in the UK is estimated as 95 782.4 km2. It is these critical load values for each ecosystem that are used for this study, which total 328 713 records.

Figure 12.1 The components of the critical load function (CLF). Note that where both sulphur and nitrogen deposition reductions are required, the exceedance must be calculated as the sum of both depositions and not simply the value equating to the length of the straight line from the deposition point to the CLF

Deposition data

Deposition input data are a combination of wet, dry and cloud nitrogen, non-marine sulphur and ammonia deposition. Note that marine sources of sulphur are not included in the calculations of exceedance, since the aim here is to assess the impact of acidifying deposition from anthropogenic sources only. Deposition estimates for moorland were applied in the calculations for acid grassland, calcareous grassland and heathland; woodland deposition estimates were used for coniferous and deciduous woodland and mean deposition estimates for freshwaters.

Two different baseline deposition scenarios have been used to investigate the effect of deposition uncertainty on exceedance within the UK. The first related to the period 1995-97. This scenario was based on 5 km spatial resolution data derived from annual measured mean deposition values for the three-year period 1995-97. Maps of total deposition for each 5 x 5km grid square of the UK were provided by the Centre for Ecology and Hydrology (CEH) Edinburgh for this work (Smith and Fowler, 2002; Smith et al., 2000). Figure 12.2a shows sulphur deposition and Figure 12.2c shows the total nitrogen deposition for this scenario. This scenario is referred to elsewhere in this topic as the 1995-97 baseline scenario.

The second scenario was the Gothenburg Protocol deposition scenario modelled from the Hull Acid Rain Model (HARM) (Metcalfe et al., 2001) and the Fine Resolution Ammonia Model (FRAME) (Singles et al., 1998; Sutton et al., 1995) atmospheric deposition models.

Figure 12.2 Baseline deposition scenarios at a 5 km spatial resolution. (a) 1995-97 sulphur, (b) 2010 non-marine sulphur, (c) 1995-97 total nitrogen (nitrogen oxide + ammonia) and (d) 2010 total nitrogen (nitrogen oxide + ammonia). Note: there is currently no data available for the Isle of Man for this scenario. Nitrogen oxide is the collective term for nitric oxide and nitrogen dioxide.

This is a deposition scenario for 2010, assuming the implementation of the recent United Nations Economic Committee for Europe Gothenburg Protocol (UN/ECE, 1999). HARM deposition was calculated at 10 km spatial resolution and FRAME at 5km spatial resolution. HARM was converted to 5km spatial resolution by sub-dividing the 10 km grid cells so, like the first scenario, it also had a 5 km cell size. The HARM outputs are used for sulphur and nitrogen oxide deposition while FRAME outputs are used for ammonia deposition. The deposition scenario is illustrated in Figures 12.2b for sulphur and 12.2d for nitrogen. This scenario is referred to elsewhere in this topic as the 2010 baseline scenario.

The exceedance calculation

The acidity CLF can be used to calculate the total acid deposition reduction necessary to obtain ecosystem protection. Five different regions are defined on the CLF (Figure 12.1); one being the envelope of protection below the CLF and the other four defining combinations of sulphur and nitrogen deposition, where reductions are required to reach the CLF (Figure 12.1). Note that where both sulphur and nitrogen deposition reductions are required, the exceedance must be calculated as the sum of both depositions and not simply the value equating to the length of the straight line from the deposition point to the CLF. Elaboration on the calculation of the exceedance function for acidity critical loads as defined above can be found in Posch et al. (1999).

The total area of ecosystems exceeded (in km2) in each 5km square can be calculated and mapped. The AE values can be summed to give total AE values for each 5 km grid square. These values can then be summed for regions of interest. For this study, statistics for the whole UK were used to give an estimate of the effect of deposition uncertainty at the national level. Maps at a 5km spatial resolution of the ecosystem areas exceeded and AE for the 1995-97 and 2010 baseline deposition scenarios are shown in Figure 12.3. The existing national methodology used to produce Figure 12.3 (Hall et al., 2001b) is referred to as the deterministic approach throughout this topic.

Uncertainty analyses

Uncertainty analysis involves the computation of uncertainty in the output induced by the quantified uncertainty in data inputs and model parameters. Two methods for examining uncertainty in deposition values were used: fixed value analysis and Monte Carlo analysis. The fixed value analysis, although an unrealistic method of simulating uncertainty, was used as an initial attempt to create bounds of uncertainty of the exceedance calculation.

The fixed value analysis

The estimates of total sulphur to a 20 x 20 km area were speculated to have an uncertainty of ± 40 % in central England increasing to ± 80 % in the west of Scotland and in Wales.

Figure 12.3 Maps at a 5 km spatial resolution showing where acidity loads are exceeded by acid deposition for (a) area of ecosystems exceeded by 1995-97 baseline deposition, (b) area of ecosystems exceeded by 2010 baseline deposition, (c) accumulated exceedances for 1995-97 baseline deposition and (d) accumulated exceedances for 2010 baseline deposition.

These values were derived from the work of Smith et al. (1995) who also noted that, as there are many uncertainties in the system, the estimates may be inaccurate. For the purposes of this study, the estimated total sulphur and nitrogen deposition to any 5 x 5 km area in the UK was assumed to have an uncertainty of ± 40 % and to be uniform across the country. It is recognized that further investigation into this limit is necessary but it was deemed a reasonable estimate.

Eight simulations were investigated, for this fixed values analysis, D1 to D8. These were designed to investigate variation in sulphur only, variation in nitrogen only and variation in sulphur and nitrogen together:

whereis the total (oxidized + reduced) nitrogen deposition value as estimated from a selected baseline deposition scenario (1995-97 or 2010) for nitrogen andthe corresponding value for sulphur whilerepresent the generated deposition scenario used in the simulation for nitrogen and sulphur respectively.

Scenarios D5 and D6 were run as worst and best case scenarios calculated by incrvsing (worst case) and reducing (best case) both nitrogen and sulphur deposition.

The Monte Carlo analysis

Monte Carlo simulation is a well-established technique for assessing the effects of uncertainty in inputs and model parameters. ‘Monte Carlo simulations start with the sampling of parameter values from a known or suspected distribution. A set of distributions for each parameter is called a scenario. The scenario is used as input to the model which results in a distribution of output values. The chosen number of samples, and subsequent runs, depends on the accuracy required for the resulting output distribution’ (Barkman et al., 1995, p. 20).

The majority of previous studieshave recognized that the underlying distribution of deposition values is largely unknown due to limited data availability (Jonsson et al., 1995; Barkman et al., 1995). Hence, previous studies have often assumed a uniform distribution between the upper and lower quoted deposition values. Since the shape of the deposition distributions was unknown it was decided to sample from both uniform (or rectangular) (D9-D11) and triangular (D12-D14) distributions and see how the different distributions affected the result. One of the attractive features of a triangular distribution appears to be that it exhibits a rough similarity to a normal distribution without tails, a seeming advantage if little is known about the distribution.

The following Monte Carlo simulations were used:

 

 

 

 

whereis the generated nitrogen deposition scenario used in the simulation, is the generated sulphur deposition scenario used in the simulation,is the lower limit for nitrogen depositionis the lower limit for sulphur depositionis the higher limit for nitrogen depositionis the higher limit for nitrogen deposition

is a random value between 0 and 1 and rs is a random value between 0 and 1.

Monte Carlo simulations involve assumptions about the mutual independence of the input variables. It has, therefore, been assumed here that the nitrogen and sulphur depositions are independent. The validity of this assumption may be questioned as oxidized nitrogen emissions can also originate from the same sources as sulphur, from, for example, coal-burning power stations and other stationary sources, but additionally from vehicle exhaust fumes. These simulations have also ignored the spatial correlation of the model input data as a single value of rs and rN has been used for the whole map.

Calculated outputs

The outputs analysed are the exceeded area and AE. In the next section, these are mapped to give the spatial distribution of these variables in the UK using the deterministic approach. Frequency distributions of the exceedance values for each critical load value (totalling 328 713 for all ecosystems) calculated using the deterministic approach are displayed to show how areas of ecosystems are distributed around the exceeded/non-exceeded cut-off level. These areused to help explain how perturbations to the deposition scenarios affect the area exceeded and AE outputs. For the fixed value analysis the variation in uncertainty was quantified as upper and lower bounds of exceeded area and AE. For the Monte Carlo trial runs the results were presented in the form of frequency diagrams. The uncertainty is quantified both as ranges and confidence intervals for these trial runs.

Results

Deterministic

This section gives the exceedance results using the baseline exceedance calculations defined in Hall et al. (2001b), and summarized above, as well as the 1995-97 and 2010 baseline deposition scenarios. The critical loads data used were the data described in Hall et al. (2001a). The total area of each ecosystem for which the critical load was exceeded and the AE across the UK, not accounting for uncertainty, were calculated. The ecosystem area exceeded in the UK for 1995-97 was estimated to be 68 265 km2 and for 2010 as 29 330 km2, 71 % and 31 % of sensitive ecosystems respectively. The AE in the UK for 1995-97 was estimated to be 7 236 915 keq year-1 and for 2010 as 1 436 411 keq year-1. These estimates are the deterministic baseline values to which the effect of uncertainty is compared in the next sections. The exceeded areas and AE have been mapped at a 5 km spatial resolution and are shown in Figure 12.3.

The frequency distribution of exceedance values for all ecosystems was formed for both 1995-97 and 2010 baseline deposition scenarios (Figure 12.4). The peak of the 1995-97 frequency distribution (Figure 12.4a) fell into the positive range indicating that deposition generally exceeded the critical load while the peak of the frequency distribution for 2010 was in the negative range (Figure 12.4b). This is to be expected, as the deposition values for 2010 were less than those for 1995-97 (Figure 12.2), so the critical loads were less likely to be exceeded, resulting in smaller areas of ecosystems exceeded.

Fixed value uncertainty analysis

The effects of fixed value uncertainty in deposition values on the total area of ecosystems exceeded and AE are displayed in Table 12.1 for all eight simulations (D1 to D8). Changes relative to the deterministic values are calculated using 68 265 km2 and 7 236 915 keq year for area exceeded and AE respectively for the 1995-97 simulations and 29 330 km2 and 1436411 keq year for the 2010 simulations.

Area exceeded

The worst case scenario for the 1995-97 period resulted in an increase in the area of critical load exceedance of 12 676 km2, whereas the best case scenario resulted in a decrease of 30448 km2. For the 2010 deposition the worst case scenario resulted in an increase in exceeded area of 14 065 km2 and the best case scenario a decrease in area exceeded of 20 767 km2.

Table 12.1 Change in area of critical load exceedance and AE from the deterministic case for 1995-97 and 2010

Simulation		Change in ecosystem area exceeded (km2)		Change in AE (keq year 1)
		1995-97	2010	1995-97	2010
D1 =	base N, high S	+ 8 224	+ 3 165	+2 246 559	+ 309 284
D2 =	base N, low S	-10388	-3 572	-2 044 765	-277 427
D3 =	high N, base S	+10 080	+11 071	+4 276 278	+1 867 068
D4 =	low N, base S	-16 897	-17170	-3 745202	-1157104
D5 =	high N, high S	+12 676	+14 065	+6 660 613	+2 277 535
D6 =	low N, low S	-30 448	-20767	-5 299 371	-1 264 338
D7 =	high N, low S	+ 3 810	+ 8 548	+1 975 553	+1 484 453
D8 =	low N, high S	- 4957	-16031	-1 851 343	-1007 360

Figure 12.4 Frequency distributions showing the number of individual areas of ecosystem which comprise the UK that fall into 0.5keq exceedance bins. (a) 1995-97 simulation and (b) 2010 simulation

Across the range of deposition uncertainty (- 40% to + 40% of the baseline deposition estimate) there was a non-linear relationship between deposition and area of critical load exceedance for both the 1995-97 and 2010 deposition scenarios. It was also apparent that a larger change in exceeded area was associated with nitrogen rather than sulphur deposition (Table 12.1).

Accumulated exceedance

The 1995-97 worst case scenario resulted in an increase in AE of 6660 613 keq year-1. The best case scenario resulted in a decrease in AE of 5 299 371 keq year 1. The 2010 deposition worst case scenario resulted in an increase of AE of 2277 535 keq year-1. The best case scenario resulted in a decrease in AE of 1 264 338 keq year-1.

Across the range of deposition uncertainty (- 40% to + 40% of the baseline deposition estimate) there was a non-linear relationship between deposition and AE for both the 1995-97 and 2010 deposition scenarios. As with the area exceeded, a larger change in AE was apparent for nitrogen rather than sulphur deposition (Table 12.1).

Although using the above worst and best case scenarios may not realistically simulate deposition uncertainty under field conditions, the results indicate the possibility of a substantial under-estimation of areas of critical load exceedance at the national scale.