To view PDF files, you need the Acrobat® Reader®.
8. Refined Calibrated Results
The results of the TNM Validation Study Phase 1 are now presented in terms of the same variables that were described in Section 7 (Section 7 results being for the initial calibrated data). This section focuses on refined calibrated results, where the data were calibrated using a reference microphone, as described in Section 6.2. The data were processed according to Section 5.1, the case of data captured during limited wind conditions [data captured during wind speeds exceeding ~11 mph (5 m/s) were removed - referred to as the strong-wind-removed data]. This refinement (from the all-wind data - data captured during all wind conditions were retained) was made to further increase the stability of the data (removing strong wind influences) and to eliminate any possible contamination at the microphone due to wind.
As in Section 7, a direct comparison is made between the TNM-predicted sound levels and the measured sound levels, and the remaining results are presented in terms of the difference or delta between the TNM-predicted sound levels and measured sound levels as a function of these variables: distance, height, wind speed and direction, and additionally, the percentage of heavy trucks. Again, in addition to presenting all the data from all the sites as a whole, the data are also divided into three categories: open area, acoustically soft ground sites {e.g., field grass [effective flow resistivity (F) . 150 cgs Rayls] or lawn [F . 300 cgs Rayls]}; open area, acoustically hard ground sites [e.g., pavement or water (F . 20,000 cgs Rayls)]; and barrier, soft ground sites. This is done in order to reveal possible site-specific influences on the results. (Measurements for barrier; acoustically hard ground sites were not performed as part of this evaluation; the barrier, acoustically soft ground sites will be referred to as just barrier sites for the remainder of this section.)
Plot and table descriptions that were detailed in Sections 6 and 7 will be repeated in this section for convenience.
8.1 Direct Comparison of TNM-Predicted and Measured Sound Levels
The first investigation of the results was simply to directly compare the TNM-predicted sound levels to the measured sound levels, as with the uncalibrated results (Section 6) and the calibrated, all-wind results (Section 7). For presentation, the set of graphs corresponding to the strong-wind-removed data results [data for winds exceeding ~11 mph (5 m/s) were removed] are seen in Figures G.2 through G.6 in Appendix G.
For a direct comparison, the data are plotted with the horizontal axis being the measured sound levels and the vertical axis being the TNM-predicted sound levels. Each 15-minute data block (15-min Leq) is represented as an orange X, where the number of data points is stated in the lower right corner of the figure. A dashed blue line represents the linear fit and solid green lines show the 95 percent confidence band. A solid black diagonal line symbolizes perfect agreement between TNM-predicted data and measured data. Data points that fall above (to the left of) this line indicate over-prediction and points that fall below (to the right of) this line indicate underprediction. The text at the top of the figure indicates the type of site for which the data correspond.
In addition to the graphs found in Appendix G, Table 10 in this section gives numerical values corresponding to the statistical elements of the graphs. In this table, the relation of the linear fit to the line of perfect agreement is examined along with the width of the 95 percent confidence band; values for five variables are stated across the columns. The first two variables concern the linear fit; values for both the average difference and the average of the absolute value of differences are stated. The first variable, the average difference, indicates how well TNM is performing over a broad range of sound levels, combining the over- and under-predictions. The second variable, the absolute value of differences, indicates how well TNM is performing as a function of the amplitude of the over- and under-predictions. This second variable can also indicate the consistency of over- or under-predictions for a range of sound levels. The third, fourth, and fifth variables in the table are the average, maximum, and minimum values of the 95 percent confidence band width, respectively. If all three values are small, and the maximum and minimum values are similar, this indicates that an average of the data shows little variation in amplitude over a broad range of sound levels; as such, a similar data set (sound levels measured and predicted under the same conditions) would provide similar results.
Results
The results for the data set where the strong wind data were removed will now be described in the order they are presented in Appendix G (Figures G.2 through G.6) and Table 10. The calibrated data for all sites (Figure G.2 and Table 10) show that TNM is in excellent agreement with the measured sound levels, the average difference being only -0.4 dB. There is only very slight over-prediction at the lower sound levels and slight under-prediction at the higher sound levels. The confidence band width is narrow over all sound levels, the average being 0.6 dB.
Note: positive values indicate over-prediction; negative values indicate under-prediction.
The results for the open area, acoustically soft ground sites (Figure G.3 and Table 10) indicate excellent agreement between predicted and measured data, with the average difference being 0.1 dB. There is some variation in the confidence band width, the average being 1.2 dB. Since the confidence band encompasses the perfect agreement line over the range of levels, it can be stated that there is no statistical difference between the measured and modeled results for open area, acoustically soft ground sites.
For open area, acoustically hard ground sites, the data are first presented as a group (Figure G.4 and Table 10) then divided into far distance (lower sound levels) and near distance (higher sound levels) data (Figure G.5 and Table 10). As a group, a substantial skew is seen in the linear fit, where the average difference from perfect agreement is 1.2 dB. Dividing the data into two categories allows a better evaluation of TNM's performance at this type of site. For near distances, the average difference for the linear fit is only -0.4 dB, with the average confidence band width being 1.8 dB; this indicates excellent agreement. For far distances, the average difference is 2.4 dB, with the average confidence band width being 0.7 dB; this indicates some over-prediction.
Lastly, for barrier sites (Figure G.6 and Table 10), results show excellent agreement, indicating a consistent average under-prediction of only -0.6 dB. The confidence band width is narrow, an average of 0.7 dB, over all sound levels.
Discussion
Overall, TNM is performing very well in a direct comparison to the measured data, the average difference from perfect agreement being less than half a decibel. In examining the performance by site type, TNM is performing very well for open area, acoustically soft ground sites; open area, acoustically hard ground sites at near distances; and barrier sites - all within 0.1 to 0.6 dB of perfect agreement, some cases showing no statistical difference between the measured and modeled results. The only difference of concern arises for open area, acoustically hard ground sites at far distances [in these cases, beyond 900 ft (~275 m) from the roadway], where TNM is over-predicting an average of 2.4 dB.
At far distances for open area, acoustically hard ground sites, the measured spectra and predicted spectra were examined to help understand the differences in levels. Spectra for the 900-ft (~275- m) position at site 13CA and the 1273-ft (~390-m) position at site 17CT were observed for both the measured and TNM-predicted data (a special diagnostic tool for TNM allows the extraction of spectra, not just the overall sound levels). Examination of a limited set of data points revealed that the overall sound levels at the far distances were dominated by frequencies between 200 and 2000 Hz. For a majority of the frequencies in this range, TNM under-attenuated the sound, causing the overall sound levels to be higher than the measured sound levels. Since TNM theoretically accounts for the hard ground by reflecting most of the energy, these barely attenuated reflected sound waves also reach the receiver. The sound waves at the measurement site most likely did not achieve this near perfect reflection upon impact with the hard ground; therefore, less energy was reflected at each encounter, and at far distances, the sound levels would be lower than those predicted using near perfect reflection. This will be investigated further in future development of TNM.
The strong-wind-removed data in this section and all-wind data in Section 7.1 show similar results. One of the noticeable differences is that the confidence band widths are narrower for the all-wind cases than the strong-wind-removed cases; this is most likely because of the greater number of data points in the all-wind cases - it is more certain that the average of that type of data set would be in that range.
Another noticeable difference between the two data sets concerns the average difference of the linear fit from perfect agreement. One would expect the linear fit difference to decrease when removing the strong wind data, as this eliminates some of influences of wind on the measured data which are not accounted for by TNM. The linear fit difference does in fact decrease for the following groups: open area, soft ground sites; near distance open area, hard ground sites; and barrier, soft ground sites. The difference from perfect agreement also decreases for all sites as a whole, from -0.8 dB to -0.4 dB. The biggest improvement is for barrier sites (from -1.2 to -0.6 dB), most likely due to the strong winds, since they would have the largest effect at barrier sites. This is further discussed in Section 8.3.
The only increase in the difference between the linear fit and perfect agreement is for the far distance open area, hard ground sites (from 2.2 dB to 2.4 dB). For the data points that were eliminated due to strong winds, it appears that the influence of the wind on the measured sound levels brought them closer to the TNM-predicted sound levels. Although the wind conditions are discussed in detail in Section 8.3, a brief insight into the linear fit differences for the far distance open area, hard ground sites is explained here. It is known that in downwind conditions (when the wind is blowing in the direction from the roadway to the receiver), one would measure higher sound levels at a receiver (especially noticeable at receivers placed a far distance from the roadway) than for either upwind or calm conditions. Ignoring other variables, TNM should under-predict in strong downwind cases since it does not account for wind. If, however, measurements are taken at far distances over acoustically hard ground (where TNM is overpredicting), this downwind effect would be seen as a decrease in over-prediction. Two of the open, hard ground sites had receivers placed a far distance from the roadway, Site 17CT and Site 13CA. No strong wind data were removed for Site 17CT, but Site 13CA had several strong wind data points removed; the removed data were measured during strong downwind conditions. This explains the small decrease in accuracy when removing the strong wind data points for the far distance open area, hard ground sites. Again, the influences of wind will be further discussed in Section 8.3.
8.2 Differences in Sound Levels as a Function of Distance and Height
The second investigation of the results examined the average differences (TNM minus measured) and standard deviation as a function of the distance of the receiver from the roadway or noise barrier and height of the receiver above the ground. As was stated in Section 7.2 (the all-wind data), it is important to investigate these variables: multiple distances can help determine how far from the road TNM is valid; short and tall heights above the ground can help validate ground effects (the microphone closer to the ground should be more affected by the ground surface) and can help in examining a noise barrier's shadow zone. For presentation, the set of graphs corresponding to the strong-wind-removed data results [data for winds exceeding ~11 mph (5 m/s) were removed] as a function of distance and height are seen in Figures G.7 through G.12 and Tables G.1 through G.3 in Appendix G.
For these sets of graphs, the data are plotted with the horizontal axis being the distance from either the center of the near travel lane of the roadway or the barrier and the vertical axis being the average difference (TNM minus measured) in sound levels. Also shown vertically is the standard deviation of the data from the average values. A solid black horizontal line at a value of 0 dB for the average difference symbolizes perfect agreement between TNM-predicted data and measured data. Data above this line indicate over-prediction and data below this line indicate under-prediction. The text at the top of the figure indicates the type of site for which the data correspond, with the specific sites listed in the legend. The text also indicates if the data presented are for the 5-ft height position or the 15-ft height position. For the tables in Appendix G, values for the average difference in sound levels are presented along with the standard deviation for each microphone location at each site.
In addition to the graphs and tables found in Appendix G, the table in this section, Table 11, gives the values for the average difference in sound levels for each type of site (open area, soft ground; open area, hard ground; and barrier). The averages are given for ranges of distances from the highway or noise barrier; note that only some ranges of distances are covered for each type of site. The data are also divided by the two different heights (5 ft and 15 ft), where averages over all distances are given in the right hand column.
Note: positive values indicate over-prediction; negative values indicate under-prediction.
Results
The results are now presented for Appendix G (Figures G.7 through G.12 and Tables G.1 through G.3) and Table 11, the data set where the strong wind data are removed. The data for the open area, acoustically soft ground sites at the 5-ft (1.5-m) height location (Figure G.7 and Table G.1) show that the average differences between the TNM-predicted and measured sound levels for each position at each site are within about 2.0 dB, except for Site 02MA, where TNM is over-predicting by 2.6 dB at the 200-ft (~60-m) position. The average difference of all these sites is 0.4 dB (Table 11) and the standard deviations range from 0.1 to 0.9 dB. For the 15-ft (4.5-m) height locations (Figure G.8 and Table G.1) the average differences for each position at each site are within about 1.5 dB, except for Site 02MA, where TNM is under-predicting by 2.5 dB at the 600-ft (~180-m) position, and Site 10CA-open, where it is under-predicting by 3.3 and 3.4 dB. The average difference for all these sites is -0.7 dB (Table 11) and the standard deviations range from 0.1 to 0.5 dB. In examining the different ranges of distances (in Table 11), it is seen that there is no overall trend in variation as a function of distance; with height, the true values of the differences for 5 ft (1.5 m) above the ground are always slightly greater than the 15-ft (4.5-m) height differences, but the magnitudes of the differences reveal no trend.
The data for the open area, acoustically hard ground sites at the 5-ft (1.5-m) height location (Figure G.9 and Table G.2) show that the average differences between the TNM-predicted and measured sound levels for each position at each site range from 0.0 to 4.0 dB, the larger differences generally tending to be at farther distances. The average difference for all these sites is 0.6 dB (Table 11) and the standard deviations range from 0.1 to 0.9 dB. For the 15-ft (4.5-m) height locations (Figure G.10 and Table G.2), the average differences for each position at each site are within about 1.5 dB, except for Site 17CT, where TNM is over-predicting by 2.8 dB. The average difference for all these sites is 0.0 dB (Table 11) and the standard deviations range from 0.0 to 0.8 dB. In examining the different ranges of distances (in Table 11), it is seen that there is a trend in variation as a function of distance; at far distances the differences are greater than at the near distances. With height, the differences reveal no trend.
The data for the barrier sites at the 5-ft (1.5-m) height location (Figure G.11 and Table G.3) show that the average differences between the TNM-predicted and measured sound levels for each position at each site are within about 2.0 dB, except for Site 09CA, where it is, in general, under-predicting by 2.3 to 3.6 dB, and Site 11CA, where TNM is over-predicting by 3.1 dB at the 300-ft (~90-m) position. The average difference for all these barrier sites is -0.4 dB (Table 11) and the standard deviations range from 0.0 to 1.4 dB. For the 15-ft (4.5-m) height locations (Figure G.12 and Table G.3), the average differences for each position at each site are within about 2.0 dB, except for Site 09CA, where TNM is generally under-predicting by 3.4 dB at all locations, and Site 10CA-berm, where it its over-predicting by 2.4 dB at the 70-ft (~20-m) position. The average difference for all these sites is 0.1 dB (Table 11) and the standard deviations range from 0.1 to 1.5 dB. In examining the different ranges of distances (in Table 11), it is seen that there are no strong trends in variation as a function of distance or height for these sites, except that there is under-prediction in the 201- to 300-ft (~60- to ~90-m) range and over-prediction in the 301- to 500-ft (~90- to ~150-m) range for both heights.
Discussion
Where the strong wind data were removed, the results (as a function of distance and height) indicate that the average difference between the TNM-predicted sound levels and the measured data is mostly within 1.5 to 2.0 dB, with several sites' differences being within 1.0 dB. The exceptions are few and occur only at some microphone positions for some sites; discussions regarding these sites will follow. Also, in examining the sites by type, the results do not show any strong trends due to the height of the receiver (microphone) or distance from the roadway, except for the open area, hard ground sites, where the tendency is toward larger differences between TNM-predicted data and measured data at the farther distances [greater than 500 ft (~150 m)].
These discussions by site type apply to both the all-wind (Section 7.2) and strong-wind-removed data (this section).
For the open area, acoustically soft ground sites 02MA and 10CA-open, some under- and overpredictions occurred. Site 02MA was the only measurement site in Phase 1 of this validation study to have an undulating ground surface, and it also had a grass median. The site undulations most likely contributed to the difficulty in achieving good predicted results at some of the positions. Section 8.5 will discuss alternate methods for modeling this site, with some differences in the results. Phase 2 of the validation study will incorporate more of the undulating sites to further analyze TNM's performance. Site 10CA-open was the only site in Phase 1 to have a plowed soft dirt ground surface. At this site the 15-ft (4.5-m) positions were underpredicted. The ground was modeled as loose soil (F = 500 cgs Rayls); in addition, other ground types were implemented, but with no improvements. It is known that rough surfaces attenuate sound differently than smooth surfaces [Attenborough 2000] [Chambers 1997]. Because TNM can account only for the ground type and not the surface type, it is likely that the rough surface of plowed soft dirt may have contributed to the differences in the predicted and measured levels. More sites with unusual ground surfaces should be examined to better evaluate TNM's performance in such situations.
For the open area, acoustically hard ground sites 15CA and 17CT, the over-predictions seem to be distance dependent (greater with greater distance). TNM propagates sound over acoustically hard ground in a theoretical sense, where the ground reflections may not properly capture the energy loss experienced in a real outdoor situation. Please refer back to the spectral discussion in Section 8.1 for more details.
For the barrier sites, under- and over-predictions varied depending on the removal of strong winds. Where all wind data were included, Sites 04CT, 08CA, and 09CA all show underpredictions. Where the strong wind data were removed, Site 09CA still shows under-predictions and Sites 10CA-berm and 11CA show some over-predictions. The over-predictions occurred at the closer high microphone position at Site 10CA-berm and the farther low microphone position at Site 11CA; similar locations at other noise barrier sites show good results. There is no immediately apparent reason for the over-predictions at these two sites; further investigation is needed. As for the under-predictions, some are expected at noise barrier sites under certain wind conditions. In downwind situations (wind blowing in the direction from the roadway to the receiver), noise barriers can become less effective as the wind pushes the sound down into the shadow zone (the area behind a noise barrier where the sound is strongly attenuated under calm wind conditions). There will be further discussions of wind in Section 8.3. With Site 09CA, TNM seems to be consistently under-predicting, more so for the all-wind case. Further thought about this site possibly reveals the cause. A 5- to 6-ft (1.5- to 1.8-m) wall surrounded the back of the site, along with a community of relatively dense houses, forming a triangular shape, with the noise barrier being one side of an almost equilateral triangle. It is possible that with the elevated roadway and barrier, the sound could have gotten trapped in this huge "pit," causing the reflected sound levels to be added; in such a case, the measured levels would be consistently higher than the TNM-predicted levels, which the results indicate. Refer to Figures B.8(a) and (b) in Appendix B for a picture and TNM views of Site 09CA.
8.3 Differences in Sound Levels as a Function of Wind Speed and Direction
The third investigation of the results for the strong-wind-removed case examined the differences (TNM minus measured) as a function of wind speed and direction. As was stated in Section 7.3 (the all-wind data), it is important to investigate TNM's performance in terms of wind variables, since these are not accounted for in the model; under certain conditions, measured sound levels are affected by the wind, influencing the differences between TNM-predicted and measured sound levels. For presentation, the set of graphs corresponding to the strong-wind-removed data results [data for winds exceeding ~11 mph (5 m/s) were removed] as a function of wind speed and direction are seen in Figures G.13 through G.15 in Appendix G.
For these sets of graphs, the data are plotted with the horizontal axis being the wind speed and the vertical axis being the difference (TNM minus measured) in sound levels. Each data point represents a 15-minute data block (15-min Leq) and is further categorized by wind direction. For characterization of wind direction, the wind component perpendicular to the roadway is specified; the three wind direction categories are up, down, and calm. "Up" signifies an upwind condition (wind blowing in the direction from the receiver to the roadway) at a speed greater than or equal to 2.2 mph (1 m/s); "Down" signifies a downwind condition (wind blowing in the direction from the roadway to the receiver) at a speed greater than or equal to 2.2 mph (1 m/s); and "Calm" signifies that the perpendicular wind component is less than 2.2 mph (1 m/s) in either direction. A solid black horizontal line at a value of 0 dB for the difference symbolizes perfect agreement between TNM-predicted data and measured data. Data above this line indicate over-prediction and data below this line indicate under-prediction. The text at the top of the figure indicates the type of site for which the data correspond, with the specific sites listed in the legend. It should be noted that fewer data points are available for this analysis than for that in Section 8.1 (also presenting 15-minute data blocks) because the wind had to be directionally consistent throughout the 15 minutes; otherwise, the data point was discarded.
In addition to the graphs found in Appendix G, Tables 12 through 14 in this section give numerical values corresponding to the graphs. In these tables grouped by site type, averages for the wind speed are presented for each site, along with the corresponding values for the average difference in sound levels categorized by wind direction. Also, overall averages are given at the bottom of each table for all sites combined. For the wind study, results are not presented as a function of microphone height above the ground; investigation is planned for later phases of the study.
Results
The results will now be described in the order they are presented graphically in Appendix G and in tables in this section. The data for the open area, acoustically soft ground sites (Figure G.13 and Table 12) show that, for the data as a group, there is no strong trend indicated. There are only two sites that have anything but calm wind conditions in the direction perpendicular to the highway; Site 10CA indicates nothing, and Site 02MA indicates that TNM is under-predicting in downwind conditions. Averages over each wind condition for all open area, soft ground sites show a -1.0-dB under-prediction for upwind conditions, a -1.9-dB under-prediction for downwind conditions, and a 0.3-dB over-prediction for calm conditions.
The data for the open area, acoustically hard ground sites (Figure G.14 and Table 13) show that, for the data as a group, there is no strong trend indicated. There are two sites that have other than calm wind conditions in the direction perpendicular to the highway; Site 15CA indicates that TNM is over-predicting in downwind conditions, and Site 16MA indicates that TNM is under-predicting in downwind conditions. Averages over each wind condition for all open area, hard ground sites show nothing for upwind conditions, a 0.1-dB under-prediction for downwind conditions, and a 1.0-dB over-prediction for calm conditions.
Note: positive values indicate over-prediction; negative values indicate under-prediction.
Note: positive values indicate over-prediction; negative values indicate under-prediction.
Note: positive values indicate over-prediction; negative values indicate under-prediction.
The data for the barrier sites (Figure G.15 and Table 14) show that, for the data as a group, there is a trend indicating that upwind conditions may cause over-prediction by TNM and downwind conditions may cause under-prediction by TNM. Sites 06CA, 10CA, and 11CA readily show over-prediction in upwind conditions. Site 12CA leans toward under-prediction in downwind conditions. Site 09CA, the only site with upwind, downwind, and calm conditions needs to be examined closely. As stated earlier (in Section 8.2), Site 09CA shows differences offset in the negative direction. If the predicted 09CA data were to be shifted up 3.0 dB as an approximation to account for additive reflections (as suggested in Section 7.3; explained in 8.2), the data set would show the calm data differences being distributed around the zero line, the upwind differences indicating mostly over-predictions, and the downwind differences indicating under predictions. Averages over each wind condition for all barrier sites show a 0.3-dB overprediction for upwind conditions, a -1.9-dB under-prediction for downwind conditions, and a - 0.4-dB under-prediction for calm conditions. Upon shifting the 09CA averages by positive 3.0 dB, the averages over each wind condition would show a 0.9-dB over-prediction for upwind conditions, a -0.9-dB under-prediction for downwind conditions, and a -0.1-dB under-prediction for calm conditions.
Discussion
Where the strong wind data were removed, TNM's accuracy is relatively unaffected by the wind (on average less than 0.5 dB) for the open area, soft ground sites; TNM-predicted sound levels are closer to the measured levels in downwind conditions (increasing the accuracy) because of other overpredictions for the open area, hard ground sites; and TNM's accuracy is dependent on the wind (affected on average up to 1.0 dB) for barrier, soft ground sites.
As was stated in Section 7.3, TNM's accuracy for certain cases should be dependent on the wind since the model calculates sound levels for a windless environment. In general, upwind conditions can lower the measured sound levels at the receiver position, and downwind conditions can raise the measured sound levels at the receiver position, the effects greater with higher wind speeds. Refraction caused by the wind affects both soft-ground attenuation and barrier insertion loss [Beranek 1992]. In addition, over hard ground sites, the sound can be channeled in downwind conditions (raising the received sound levels).
In examining the results as a function of wind, the open area, soft ground data overall indicate under-predictions for both upwind and downwind conditions, more so for the downwind conditions. However, data exist for only one site under upwind conditions and one site for downwind conditions. The site with downwind conditions, Site 02MA, does indicate some influence from wind since the downwind data are vertically offset from the calm wind data; as was stated in Section 7.3, there is the possibly that this site may behave more like a barrier site when it comes to wind effects because of the large ground undulations. Overall, it is difficult to make any firm conclusions about the effects of wind on the accuracy of TNM predictions at open area, soft ground sites, although it seems that TNM's accuracy is relatively independent.
In examining the results as a function of wind, the open area, hard ground data overall indicate neutrality for downwind conditions; no data exist for upwind conditions. Under downwind conditions, the sound can be refracted downward then reflected upward, channeling the sound (hence raising the sound levels at the microphones, especially over long distances). The results do not readily support this, although some explanation can be offered. As was stated in Section 7.3, for the downwind results, even though Table 13 indicates a 0.1 dB overall over-prediction and not an under-prediction as one would expect, it is important to recall results presented earlier in this section. TNM is tending to over-predict for farther positions at hard ground sites; these over-predictions may overpower the under-predictions from downwind effects, thereby merely lowering the over-predictions. For hard ground sites, results do not indicate that the accuracy of TNM is very affected by the wind, but part of the non-effect is due to general over-prediction at hard ground sites; this is noted for investigation in later phases of the TNM Validation Study.
In examining the results as a function of wind, the barrier data overall indicate some overprediction for upwind conditions and some under-prediction for downwind conditions. Many of the sites indicate this trend, especially the ones where upwind or downwind data, when compared to calm wind data, show a definite increase or decrease, respectively, from the calm wind results (Sites 06CA, 09CA, and 14CA in Figure G.15). At noise barrier sites, under upwind conditions, the sound is being refracted upward, making the barrier more effective (hence reducing the sound levels at the microphones). Under downwind conditions, the sound is being refracted downward behind the barrier, making the barrier less effective (hence raising the sound levels at the microphones). Results from the barrier sites indicate that wind is a factor in TNM's ability to predict precisely accurate results. It is seen, however, that the average wind influence is less than 1.0 dB (less than 2.0 dB without adjusting Site 09CA); this is noted for investigation in later phases of the TNM Validation Study.
Examining the differences between TNM-predicted sound levels and measured sound levels as a function of wind, similar overall results are seen for the all-wind data and the strong-windremoved data. The only notable difference is the greater influence of wind at barrier sites. The higher wind speeds included in the all-wind data influenced the differences between TNMpredicted and measured data more than the lower wind speeds for the strong-wind-removed data. If one were to assume that the all-wind data were uncontaminated by the high wind speeds, which in most cases was probably true, then results indicate that TNM's accuracy is dependent on the wind environment, where the wind causes differences from measured levels of 2.0 dB or more at higher wind speeds [>11 mph (5 m/s)] and 1.0 dB at lower wind speeds.
8.4 Differences in Sound Levels as a Function of Percentage of Heavy Trucks
The third investigation of the results for the strong-wind-removed case examined the differences (TNM minus measured) as a function of percentage of heavy trucks. This was not investigated for the all-wind data. It is important to investigate TNM's performance in terms of percentage of heavy trucks in order to verify the implementation of this type of vehicle in the model. Heavy trucks are modeled differently from other vehicle types because of the added noise emission source for the truck stack exhaust. At highway speeds, 95 percent of the acoustical energy is apportioned to the tire/pavement interaction noise and 5 percent to the truck stack exhaust noise [Coulson 1996]. There are also other differences from other vehicle types, for example, relative levels in the emission spectra.
Any issues related to the implementation of heavy trucks would be more apparent with a greater percentage. For presentation, the set of graphs corresponding to the strong-wind-removed data results [data for winds exceeding ~11 mph (5 m/s) were removed] as a function of percentage of heavy trucks are seen in Figures G.16 through G.18 in Appendix G.
For these sets of graphs, the data are plotted with the horizontal axis being the percentage of heavy trucks and the vertical axis being the difference (TNM minus measured) in sound levels. Each data point represents a 15-minute data block (15-min Leq). A solid black horizontal line at a value of 0 dB for the difference symbolizes perfect agreement between TNM-predicted data and measured data. Data above this line indicate over-prediction and data below this line indicate under-prediction. The text at the top of the figure indicates the type of site for which the data correspond, with the specific sites listed in the legend.
Results
The results will now be described in the order they are presented graphically in Appendix G. The data for the open area, acoustically soft ground sites (Figure G.16) show no overall trend; the percentage of heavy trucks ranges from about 2 to 13 percent. Only Site 03MA shows a trend of slightly more over-prediction with a higher percentage of heavy trucks.
Discussion
When examining the strong-wind-removed data as a function of the percentage of heavy trucks, it is seen that there are some slight site-specific trends. Also, a slight overall trend is seen for the acoustically hard ground sites; the previously discussed over-predictions (Section 8.2) for far distances at acoustically hard ground sites most likely influenced these results. Overall results show no distinct trends. This indicates no apparent influence of the percentage of heavy trucks on the performance of TNM, suggesting that TNM implements heavy trucks correctly.
8.5 Some Alternate TNM Runs and Recommendations
All TNM runs for the results described above were modeled according to the standard practice for this study. This standard practice included modeling all substantial terrain features; it also involved choosing the default ground type to be either lawn or field grass, unless the entire site consisted of a different ground type (e.g., pavement). Some other practices were implemented for research purposes, to determine if their effects would be useful or not. The results for a few of the Phase 1 sites, applying the alternate modeling techniques, will be presented in this section, along with recommendations based on these investigations. The strong-wind-removed data [data for winds exceeding ~11 mph (5 m/s) were removed] were used in these investigations.
Undulations versus Flat Terrain
The first investigation involved terrain features. Site 02MA was examined in this case, the only site with an undulating ground surface. See Figures B.2(a) and (b) in Appendix B for site specifics as it was originally modeled; although only one cross section is shown, it should be noted that the actual site consisted of undulations in multiple directions. The elevation of the undulations ranged from +5 to -20 ft (+1.5 to -6.0 m). All terrain lines were removed for the investigation of this case, modeling the site as flat, not undulating.
The investigation of strong-wind-removed results examined the average differences (TNM minus measured) and standard deviation as a function of the distance of the receiver from the roadway and height of the receiver above the ground, as was done in Section 8.2. The results are presented graphically in Figures G.19 and G.20 in Appendix G. These plots include all the open area, soft ground sites. In directly comparing the flat Site 02MA results (Figures G.19 and G.20) to the undulating Site 02MA results (G.7 and G.8), it is seen that flattening the terrain in the TNM run definitely affects the resulting sound levels. At the 5-ft (1.5-m) height (Figures G.19 and G.7), TNM's predictions are improved (on average about 1.5 dB) at 200 ft (~60 m) and impaired (on average about 0.5 dB) at 400 and 600 ft (~120 and ~60 m). At the 15-ft (4.5-m) height (Figures G.20 and G.8), TNM's predictions are improved (on average about 0.5 dB) at 50 ft (~15 m), impaired (on average about 0.5 dB) at 200 ft (~60 m), and are relatively unaffected at 400 and 600 ft (~120 and ~60 m).
The results as a whole indicate that this undulating site should not be simplified by flattening it (overall, nothing is gained), and that undulations of this size [+5 to -20 ft (+1.5 to -6.0 m)] cannot be ignored. Of course, more undulating ground sites (and sites with elevation changes) need to be evaluated. Measurements have already been performed at several of these sites in Pennsylvania; analysis for these sites will be part of Phase 2 of the TNM Validation Study.
Modeling a Grass Median
The second investigation involves grass medians. Through the Volpe's interaction with TNM users, it was found that TNM was calculating counter-intuitive results with grass medians, where the medians were defined solely by the default ground type between roadways. Although the counter-intuitive behavior only appears at certain distances from the roadway, an investigation involving the addition of a grass ground zone to define the median was executed. Sites 01MA and 02MA were examined in this investigation, both sites having grass medians. See Figures B.1(a) and (b) and B.2(a) and (b) in Appendix B for site specifics as they were originally modeled.
For the strong-wind-removed data, results were examined for the average differences (TNM minus measured) and standard deviation as a function of the distance of the receiver from the roadway and height of the receiver above the ground, as was done in Section 8.2. The results are presented graphically in Figures G.21 and G.22 in Appendix G. These plots include all the open area, soft ground sites. In directly comparing the grass ground zone median results (Figures G.21 and G.22) to the default grass median results (G.7 and G.8), it is seen that adding the grass ground zone to represent the median in the TNM runs (instead of allowing the median to be set based on a default ground type of grass) definitely affects the resulting sound levels. For Site 01MA, there is very little effect, but Site 02MA shows differences. For Site 02MA, there is some improvement and some impairment, the greatest improvement closer to the roadway. Because very little is shown here, the results are inconclusive. In Volpe's TNM testing and interactions with TNM users, however, some improvements were seen when adding a grass ground zone to represent the median.
It is recommended to add a grass ground zone as the median when the actual site possesses one, unless its width is small [less than 10 ft (~3 m)]; in smallwidth cases, the roadways in each direction should be extended to just overlap. Further investigation needs to be performed in order to fix this counterintuitive behavior.
Hard Ground Zone on Soft or Soft Ground Zone on Hard
The third investigation involved ground zones. Through the Volpe's interaction with TNM users and modeling for this study, it was found that TNM was potentially having difficulty with its predictions when placing a soft ground zone on default hard ground (e.g., a lawn ground zone on default pavement). Since Site 16MA, seen in Figures B.15(a) and (b) in Appendix B, contains mixed ground surfaces, it was used to test placing a soft ground zone on default hard ground; it originally was modeled as a pavement ground zone placed on default field grass. This was done informally, and graphical representation of the results are not presented here. However, cursory results showed that TNM predicts more accurate results when placing the pavement ground zone of default field grass than when placing a field grass ground zone on default pavement. Other TNM users have seen similar results. This too needs further investigation.
For now, it is recommended that mixed ground sites be modeled with the default ground type being soft ground and the ground zones being hard ground.
Discussion and Recommendations
In running alternate TNM configurations of particular sites, it is seen that substantial ground undulations cannot be ignored (i.e., the site cannot be modeled with flat ground), as is expected. For the site examined, the undulations ranged from +5 to -20 ft (+1.5 to -6.0 m); more sites with undulations and changes in elevation will be analyzed in Phase 2 of the study. Investigations involving grass ground zone medians were inconclusive here, but it is recommended to add a grass ground zone as the median when the actual site possesses one, unless its width is small [less than 10 ft (~3 m)]; in small-width cases, the roadways in each direction should be extended to just overlap. Investigations involving ground zones at mixed ground [soft (e.g., field grass) and hard (e.g., pavement)] sites indicate that this type of site should be modeled with the default ground type being soft ground and the ground zones being hard ground.
8.6 Summary of Refined Calibrated Data Results
(Excerpts from Discussions in each of the subsections of Section 8)
Direct Comparison of TNM-Predicted and Measured Sound Levels.Overall, TNM is performing very well, the average difference from perfect agreement being less than half a decibel. In examining the performance by site type, TNM is performing very well for open area, acoustically soft ground sites; open area, acoustically hard ground sites at near distances; and barrier sites - all within 0.1 to 0.6 dB of perfect agreement, some cases showing no statistical difference between the measured and modeled results. The only difference of concern arises for open area, acoustically hard ground sites at far distances [in these cases, beyond 900 ft (~275 m) from the roadway], where TNM is over-predicting an average of 2.4 dB.
Differences in Sound Levels as a Function of Distance and Height.Where the strong wind data were removed, the results indicate that the average difference between the TNM-predicted sound levels and the measured data is mostly within 1.5 to 2.0 dB, with several sites' differences being within 1.0 dB. The exceptions are few and occur only at some microphone positions for some sites. Also, in examining the sites by type, the results do not show any strong trends due to the height of the receiver (microphone) or distance from the roadway, except for the open area, hard ground sites, where the tendency is toward larger differences between TNM-predicted data and measured data at the farther distances [greater than 500 ft (~150 m)].
Differences in Sound Levels as a Function of Wind Speed and Direction.Where the strong wind data were removed, TNM's accuracy is relatively unaffected by the wind (on average less than 0.5 dB) for the open area, soft ground sites; TNM-predicted sound levels are closer to the measured levels in downwind conditions (increasing the accuracy) because of other overpredictions for the open area, hard ground sites; and TNM's accuracy is dependent on the wind (affected on average up to 1.0 dB) for barrier, soft ground sites.
Differences in Sound Levels as a Function of Percentage of Heavy Trucks.When examining the strong-wind-removed data as a function of the percentage of heavy trucks, some slight sitespecific trends can be seen. Also, there is a slight overall trend for the acoustically hard ground sites; the previously discussed over-predictions (Section 8.2) for far distances at acoustically hard ground sites most likely influenced these results. Overall results show no distinct trends. This indicates no apparent influence of the percentage of heavy trucks on the performance of TNM, suggesting that TNM implements heavy trucks correctly.
Some Alternate TNM Runs and Recommendations.In running alternate TNM configurations of particular sites, it is seen that substantial ground undulations cannot be ignored (i.e., the site cannot be modeled with flat ground), as is expected. For the site examined, the undulations ranged from +5 to -20 ft (+1.5 to -6.0 m); more sites with undulations and changes in elevation will be analyzed in Phase 2 of the study. Investigations involving grass ground zone medians were inconclusive here, but adding a grass ground zone as the median when the actual site possesses one is recommended, unless its width is small [less than 10 ft (~3 m)]; in small-width cases, the roadways in each direction should just be extended to overlap. Investigations involving ground zones at mixed ground [soft (e.g., field grass) and hard (e.g., pavement)] sites indicate that this type of site should be modeled with the default ground type being soft ground and the ground zones being hard ground.
| Previous | Contents | Next |
