4. Modelling Ambient Particle Concentrations
4.4. Incorporating Temporal Variability in the LUR Model
4.4.5. Step 3: Comparing Temporal Adjustments
In the following section the two versions of temporal adjustment A (direct adjustment) and B (adjustment by temporal patterns), as developed above, are compared in detail. The comparison is made by assessing how well each approach predicts variability of the set of prediction sites, using Pearson’s correlation. First, predictions for 2010 of London roadside stations are assessed, then concentrations for year 2011 are predicted for the same stations (dependent on data availability for year 2011) to test to what extent the adjustments would be temporally transferable. Subsequently predictions are made for concentrations at 10 monitoring stations outside London to test for spatial transferability of the approaches.
Predicting monitoring stations in London for 2010
Temporal adjustments A and B have been applied to LUR estimates of the set of prediction sites, as described above. The ability of the two adjustment versions to reflect concentration variability was examined and compared using Pearson’s correlation. Results are presented in Table 39 for PM2.5 and in Table 40 for PM10. For both PM2.5 and PM10 and for all levels of temporal aggregation the correlations were lower for adjustment B compared to adjustment A at every station. For hourly concentrations of PM2.5, correlations for adjustment A range from 0.576 to 0.880 compared to
162 between 0.063 and 0.274 for adjustment B. For hourly PM10 concentrations a similar difference was observed. Correlations range from 0.496 to 0.883 for adjustment A and from 0.130 to 0.317 for adjustment B. A similar difference was also observed between daily adjustments with the results of correlations for adjustment A higher than for adjustment B, for PM2.5 (adj. A: 0.629 < r > 0.942, adj. B: 0.170 < r > 0.355) and for PM10 (adj. A: 0.584 < r > 0.947, adj. B: 0.094 < r > 0.368). For weekly concentrations the same difference could as well be observed between adjustments A and B for PM2.5 (adj. A: 0.552 < r > 0.974, adj. B: 0.097 < r > 0.587), as well as for PM10 (adj. A: 0.488 < r > 0.939, adj. B: 0.048 < r > 0.625). The conclusion was drawn that temporally adjusted LUR predictions using adjustment A better reflected variability at other locations compared to adjustment B when applied to concentrations of the same time period and area.
2010 Data 2011 Data
ID Hourly Daily Weekly Hourly Daily Weekly
Adj. A Adj. B Adj. A Adj. B Adj. A Adj. B Adj. A Adj. B Adj. A Adj. B Adj. A Adj. B 1 0.659** 0.138** 0.732** 0.206** 0.750** 0.397** 0.037** 0.093** 0.098 0.106 0.006 0.098 2 0.790** 0.231** 0.843** 0.320** 0.824** 0.541** - - - - 3 0.844** 0.252** 0.942** 0.336** 0.941** 0.551** 0.082** 0.108** 0.146** 0.107* 0.101 0.162 4 0.860** 0.220** 0.940** 0.307** 0.944** 0.542** - - - - 5 0.576** 0.093** 0.629** 0.170** 0.552** 0.368** -0.011 0.049** 0.037 0.061 -0.131 0.027 6 0.869** 0.134** 0.923** 0.190** 0.832** 0.326** - - - - 7 0.686** 0.063** 0.726** 0.039** 0.532** 0.097 0.020 0.048** 0.074 0.014 -0.056 -0.007 8 0.870** 0.274** 0.959** 0.355** 0.974** 0.587** - - - - 9 0.880** 0.222** 0.942** 0.302** 0.928** 0.531** 0.048** 0.064** 0.104* 0.064 0.042 0.065
Table 39: Pearson’s correlations (r) between monitored concentrations and the temporally adjusted LUR model using adjustment A (direct adjustment by monitoring data) and B (adjustment based on temporal patterns) for hourly, daily and weekly PM2.5 predictions of London roadside stations for 2010 and 2011. * Significant on a 0.05 level, ** Significant on a 0.01 level
Testing temporal transferability
The construction of adjustment B was aimed to reflect transferable patterns in pollution which can be applied directly to predict different years or locations. Temporal transferability is tested here for 2011 concentrations for the same prediction sites. Data was however not available for all stations and only five out of nine PM2.5 stations and 20 out of 26 PM10 stations had sufficient data availability. The temporally adjusted LUR predictions based on 2010 data (adjustments A and B) were directly compared to 2011 concentrations. Adjustment A is added to this comparison mostly as a reference to show if in comparison adjustment B reflects a transferable temporal patterns. Direct adjustment with monitoring data as used for adjustment A would usually be calculated from 2011 background concentrations to predict locations for 2011.
163 Results for Pearson’s correlations between monitored and modelled are presented in Table 39 (right side) for PM2.5 and in Table 40 (right side) for PM10. For PM2.5 all correlation results for 2011 are below 0.2 and only station 3 shows highly significant results (at a 0.01 level) above a correlation of 0.1 for hourly concentrations for adjustment B, and for daily concentrations both for adjustments A and B.
2010 Data 2011 Data
ID Hourly Daily Weekly Hourly Daily Weekly
Adj. A Adj. B Adj. A Adj. B Adj. A Adj. B Adj. A Adj. B Adj. A Adj. B Adj. A Adj. B
1 0.844** 0.139** 0.924** 0.134** 0.853** 0.214 - - - - 2 0.723** 0.305** 0.846** 0.313** 0.821** 0.461** 0.072** 0.266** 0.105 0.264** 0.046 0.298* 3 0.853** 0.169** 0.917** 0.183** 0.864** 0.254 0.082** 0.220** 0.117* 0.234** 0.102 0.280 4 0.675** 0.288** 0.851** 0.325** 0.875** 0.503** 0.063** 0.242** 0.085 0.248** 0.039 0.293* 5 0.712** 0.239** 0.815** 0.094 0.748** 0.108 0.075** 0.175** 0.092 0.072 -0.037 0.010 6 0.496** 0.295** 0.655** 0.297** 0.606** 0.347** 0.053** 0.274** 0.052 0.211** -0.068 0.185 7 0.883** 0.253** 0.943** 0.331** 0.921** 0.503** 0.039** 0.164** 0.074 0.145** -0.020 0.170 8 0.666** 0.185** 0.779** 0.124** 0.647** 0.048 0.045** 0.165** 0.070 0.146** 0.124 0.194 9 0.873** 0.218** 0.947** 0.256** 0.939** 0.425** 0.077** 0.156** 0.127* 0.154** 0.073 0.186 10 0.795** 0.226** 0.931** 0.312** 0.930** 0.521** 0.060** 0.176** 0.083 0.155** 0.036 0.171 11 0.639** 0.247** 0.705** 0.243** 0.677** 0.337** 0.030** 0.234** 0.029 0.209** -0.065 0.181 12 0.839** 0.183** 0.584** 0.180** 0.565** 0.412** 0.042** 0.123** 0.077 0.145** -0.003 0.172 13 0.798** 0.220** 0.902** 0.295** 0.885** 0.464** - - - - 14 0.837** 0.229** 0.895** 0.149** 0.766** 0.147 0.114** 0.267** 0.165** 0.249** 0.073 0.241 15 0.849** 0.268** 0.911** 0.242** 0.887** 0.361** 0.033** 0.257** 0.047 0.239** -0.025 0.262 16 0.792** 0.185** 0.851** 0.200** 0.760** 0.335* 0.052** 0.097** 0.072 0.034 -0.026 -0.019 17 0.749** 0.263** 0.846** 0.368** 0.754** 0.625** - - - - 18 0.821** 0.237** 0.894** 0.259** 0.893** 0.442** 0.080** 0.194** .131* 0.165** 0.104 0.198 19 0.221** 0.107** 0.478** 0.173** 0.469** 0.344* - - - - 19b 0.565** 0.248** 0.754** 0.288** 0.671** 0.444** - - - - 20 0.804** 0.179** 0.898** 0.206** 0.840** 0.320* 0.048** 0.220** .104* 0.316** 0.091 0.397** 21 0.788** 0.130** 0.843** 0.104** 0.698** 0.242 0.107** 0.402** .186** 0.512** 0.212 0.665** 22 0.842** 0.214** 0.944** 0.246** 0.937** 0.415** 0.050** 0.164** 0.090 0.153** 0.027 0.185 23 0.829** 0.188** 0.904** 0.218** 0.840** 0.300* - - - - 24 0.659** 0.247** 0.731** 0.169** 0.636** 0.270 0.078** 0.241** .111* 0.165** 0.059 0.167 25 0.805** 0.317** 0.898** 0.291** 0.883** 0.481** - - - - 26 0.611** 0.253** 0.629** 0.174** 0.488** 0.171 0.037** 0.335** 0.021 0.273** -0.056 0.305*
Table 40: Pearson’s correlations (r) between monitored concentrations and the temporally adjusted LUR model using adjustment A (direct adjustment by monitoring data) and B (adjustment based on temporal patterns) for PM10 predictions of London roadside stations for 2010 and 2011. * Significant on a 0.05 level, ** Significant on a 0.01 level
For hourly PM10 concentrations all correlation results are significant at a 0.01 level and adjustment B shows higher correlations compared to adjustment A between 0.097 and 0.402. For daily PM10 concentrations most results for adjustment A are not significant (13 out of 20). Apart from results for two stations, all correlation coefficients for adjustment B are however highly significant (at a 0.01 level) and range from 0.145 to 0.512. Both for hourly, as well as for daily PM10 concentrations correlations with adjustment B are similar and for some stations higher compared to correlations with concentrations for 2010. For weekly PM10 concentrations only correlations at two stations are
164 highly significant for adjustment B (stations 20 and 21) (No highly significant results for adjustment A). Both stations show higher correlations compared to correlations with data from 2010.
In summary, the application of adjustment B to 2011 concentrations shows for PM2.5 little temporal transferability. Results for correlations are very low (r < 0.2) and are below correlation results with 2010 data. For PM10 adjustment B shows similar and sometimes higher correlation results for 2011 compared to results for 2010 data, indicating good temporal transferability. Temporal adjustment B however only reflects a small part of variability at other stations either for 2010 or 2011 concentrations.
Testing spatial transferability
Both adjustment versions A and B were applied to predict hourly PM2.5 and PM10 concentrations for year 2010 at 10 locations outside London. The 10 monitoring stations and approximate distance to London are presented in Table 41 showing results for Pearson’s correlations between temporally adjusted LUR predictions and monitored concentrations for PM2.5 and PM10 for year 2010.
All correlations are significant at a 0.01 level. Results show that for both particle size fractions and all levels of temporal aggregation, adjustment A results in higher Pearson’s correlations compared to adjustment B. For hourly PM2.5 concentrations, correlations range from 0.466 to 0.783 for adjustment A and from 0.206 to 0.363 for adjustment B. For hourly PM10 concentrations, correlations range from 0.389 to 0.692 for adjustment A and from 0.156 to 0.236 for adjustment B. The range for daily correlations is a little higher for both adjustments: For daily PM2.5 concentrations, results range from 0.593 to 0.926 for adjustment A and from 0.254 to 0.515 for adjustments B. For daily PM10 concentrations, correlations range from 0.550 to 0.907 for adjustment A and from 0.227 to 0.348 for adjustment B. Weekly correlations show higher results compared to correlations at the hourly or daily level: For PM2.5, correlations range between 0.759 and 0.931 for adjustment A and between 0.454 and 0.711 for adjustment B. For weekly PM10 concentrations, correlations range between 0.646 and 0.887 for adjustment A and between 0.358 and 0.497 for adjustment B.
In summary, results for PM2.5 show for adjustment A (direct adjustment with monitoring concentrations) a similar range of results as previously presented for stations in London, with only a few stations, such as ‘Newcastle Centre’ or ‘Bristol St Paul’s’, showing in comparison lower correlations for hourly and daily concentrations. For adjustment B correlations for PM2.5 concentrations are generally at least as high and in some cases higher compared to correlations for stations in London. For PM10 concentrations, correlation results of adjustment A are for several stations lower than results for stations in London, especially for hourly and daily correlations. For
165 adjustments B correlations between PM10 concentrations with stations outside London are within the range of results for correlation results of stations inside London for all levels of temporal aggregation.
PM2.5 PM10
Hourly Daily Weekly Hourly Daily Weekly
Station Dist. Adj. A Adj. B Adj. A Adj. B Adj. A Adj. B Adj. A Adj. B Adj. A Adj. B Adj. A Adj. B Thurrock Stanford-le-H. 30 - - - 0.784 0.199 0.907 0.276 0.887 0.459 Mole Valley Dorking 35 - - - 0.781 0.202 0.907 0.231 0.886 0.422 Reading New Town 55 0.835 0.433 0.926 0.515 0.928 0.711 - - - - Harwell 75 0.720 0.293 0.845 0.348 0.901 0.525 0.598 0.154 0.769 0.255 0.753 0.392 Oxford St Ebbes 80 0.783 0.363 0.886 0.426 0.931 0.639 0.692 0.236 0.820 0.348 0.877 0.475 Norwich Lakenfields 150 0.655 0.312 0.779 0.387 0.854 0.613 0.566 0.168 0.739 0.272 0.781 0.382 Bristol St Paul's 170 0.479 0.278 0.634 0.356 0.769 0.605 0.389 0.156 0.550 0.237 0.646 0.453 Nottingham Centre 200 0.651 0.212 0.747 0.260 0.828 0.490 0.498 0.206 0.708 0.321 0.779 0.497 Leeds Centre 300 0.585 0.272 0.699 0.342 0.792 0.565 0.508 0.185 0.643 0.227 0.654 0.358 Newcastle Centre 450 0.466 0.206 0.593 0.254 0.759 0.454 0.436 0.207 0.560 0.260 0.697 0.395 Table 41: Pearson’s correlations of temporally adjusted LUR predictions for adjustments A (direct adjustment by monitoring data) and B (adjustment based on temporal patterns) at 10 stations outside London for hourly, daily, and weekly PM2.5 and PM10 with approximate distance to Central London given in km. Dist. = Approximate distance from central London. All results are significant on a 0.01 level.
The results show very good spatial transferability for both adjustment approaches with correlations generally similar for locations outside London compared to locations in London. No clear indication of lower correlation coefficients with increasing distance from London could be observed. As for correlations with 2010 concentrations for locations in London, adjustment A was able to explain a much larger part of the concentration variability compared to adjustment B.
Summary
Temporally adjusted LUR concentrations using adjustment version A reflect variability of 2010 concentrations at different locations much better, compared to adjustment B. The variability included in adjustment B is limited to repeated temporal patterns, which only reflect a small part of actual variability. Adjustment B shows generally good spatial transferability and for PM10 as well good temporal transferability. Yet Pearson’s correlations remain below 0.5 for all predictions. Adjustment A has shown limited direct temporal transferability. The adjustment can however be
166 calculated for every year using concentrations from London background monitoring stations (available online, for years since mid-1990s). Using adjustment A therefore provides potential for applications for many years, despite low direct temporal transferability. Spatial transferability was tested and provided good correlation results between modelled and monitored for locations outside London. Considering above results, adjustment version A was selected for further analyses. The LUR model using adjustment version A is referred to as LUR-adj in the following sections.