Improving AEDT Noise Modeling of Mixed Ground Surfaces (2017)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

6-1 CHAPTER 6. FINDINGS AND APPLICATIONS The findings and applications presented below were accomplished in response to the stated goals of this project. The Research Team’s approach to meeting these goals was presented to and accepted by The Panel. Aside from the gathering of airport data, the other goals of this project are restated here with clarifications:  Find a method that accurately quantifies the change in lateral attenuation when the noise propagation is modeled over mixed or hard impedance surfaces as compared to soft ground. o It was made clear in the Go Forward Plan of the Interim Report and at the Panel review meeting that the method used by AEDT to calculate lateral attenuation over soft ground as specified by SAE AIR 5662 should NOT be changed. The method found in this project should be used as an adjustment over ground that is not all soft.  The method used to study ground effect and mixed impedance surfaces will be straight ray theory. o Because straight ray theory is so well established and matches other methods’ predictions for quiescent atmospheres, using it to find only the DIFFERENCE in ground effect to be applied to the lateral attenuation predicted by SAE AIR 5662 negates the need for applying turbulence or additional complexities to the calculations. As stated in Section 4.10, two programs were used to exercise straight ray theory: EPD and EGA.  Sensitivity studies will be performed for the application of straight ray theory in the calculation of lateral attenuation to see if computational efficiencies through regression fits are warranted. Studies were to include: o Comparison of the full EPD model versus the approximation of one-third octave band levels in the EGA model. EPD exercises ray theory for pure tone sound. In order to calculate one-third octave band propagation, EPD must be run for the range of frequencies in a particular one-third octave band and the results must be averaged in order to get the result; whereas, the EGA model outputs result for one-third octave band spectra and need only be run once. o Calculating average flow resistivity in the Fresnel ellipse versus finding the average flow resistivity along the ground on a straight line between source and receiver. o Applying the full ray model by spectrum and geometry versus a simplified regression for a fixed spectrum and limiting curves based on specific geometries.  Study of the inherent error in applying what is a point-to-point propagation phenomenon to a track segment in an integrated noise model.  Finalize a list of improvements to AEDT. What follows is the work performed on each of these items.

6-2 6.1. Fresnel Zone Ellipse Resistivity versus 2D Profile Cut It is worth looking at the difference between calculating the average flow resistivity in the Fresnel ellipse versus only considering the average flow resistivity along the major axis of the ellipse. It is common to consider only the two-dimensional profile along the ground on the line below source and receiver to calculate the ground impedance between source and receiver (AAM, NORD2000, NMAP). In this case, the ‘active’ portion of this line to be used in determining the average flow resistivity and calculation of ground impedance is the extent of the major axis of the Fresnel zone ellipse (for each frequency considered). In the case of low elevation angles and frequencies, the line to be considered can extend beyond the source and receiver. Unlike elevation data, the assigned parameters to calculate ground impedance for grid points representing terrain data should not be interpolated. In the example of the National Land Cover Database, each grid point in the terrain file is classified as a ground type with an area of 10,000 square feet. The classification of the ground type in this area is an estimate of the average type of ground found within the area based on satellite imagery. While it may at first seem reasonable to expect a smooth transition from one grid point in the terrain file to the next, there is no realistic basis for this assumption. All that is known for any individual grid point is that the ground classification should apply to the edge of the area it represents. Interpolation of these properties between grid points can lead to unrealistic results. Consider a location near water’s edge with the noise source out over the water. The position of the terrain grid points relative to the land-water interface will influence the estimate of the average surface parameters in the Fresnel ellipse. In the two-dimensional case, the effect of allowing interpolation between the grid points to estimate the surface parameters is best demonstrated by example. Using the BASEOPS flow resistivity estimates associated with the NLCD ground cover classifications around Portland International Airport (PDX) shown in Section 5.2.3 above, the average flow resistivity in Fresnel ellipses for receivers on lines normal to the runway were calculated for the geometric parameters noted in Table 7. TABLE 7 Parameter Ranges Used to Calculate the Fresnel Ellipses around PDX Separation Distance (ft): 400, 1000, 4000, 10000, 25000 Source Height (ft): 10, 100, 1000 Frequency (Hz): 63, 125, 250, 500, 1000, 2000, 4000, 8000 Figure 40 shows the white ellipses drawn to either side of the beginning of runway 28L for a source height of 10 ft above ground level (agl) with a 4 ft high receiver 10,000 ft away representing the Fresnel zones for 63 Hz. It must be understood that the discussion herein assumes the terrain is flat. Calculating the area intersection of the ellipse above the runway and each of the 100 ft by 100 ft grid point squares and finding the weighted average of the flow resistivities is straight forward. Another calculation carried out simultaneously was the percent area of the ellipse occupied by water, because some noise models (e.g. NMAP) treat the ground’s flow resistivity in binary fashion where the ground is either hard or soft.

A average fl routine re equally-sp be colloca requested every foo Because t respect th two grid p kPa s/m2. percent o resistivity A profile cu 10 ft grid created fr the area it then the 1 of 100, 00 flow resis simple prof ow resistivit turns the int aced distanc ted at grid p ; thus, reques t; whereas, he profile cu e boundary t oints with fl For calcula f hard surfac even when th s a check o t, 3 different spacing, and om the NLCD represents. 0 ft by 10 ft 0 kPa s/m2 a tivity of 225 FIGURE receivers ile cut routin y along the l erpolated va es along the oints. The s ting 101 prof a separation t routine int hat each terra ow resistiviti ting the ave e along the e profile cut f the accurac grid resolutio a 1 ft by 1 grid by rep For the exam grid point su nd the grid kPa s/m2. 40. Fresnel 10,000 ft aw e (Plotkin e ength of the lue from a r line connecti pacing along ile points for of 10,000 fe erpolates betw in grid poin es of 225 and rage flow re profile cut, is entirely ab y of the ave ns were used ft grid spacin eating the flo ple above, i rrounding the points surrou ellipses for a ay - ellipse m t al., 2013) major axis o egularly spa ng two locat the line is two location et will resu een grid po t represents. 100,000 kPa sistivity this this will enti ove water ne rage flow r : the NLCD b g. For both w resistivity f the water g water grid p nding the lan 10 ft source arking a pa was used to f the ellipse ced grid of ions on the g determined b s separated b lt in 100 ft ints to find So, a profile s/m2 will lik is reasonabl rely miss th xt to the wat esistivity est ase resolutio of the finer value of the g rid point is 1 oint out to 5 d grid point height on R th length di calculate the in Figure 40 values along rid. The poi y the numbe y 100 ft resul spacing alon a value on th point equal ely have an e; however, e contributio er-land interf imate for the n of 100 ft b grid spacing rid to all the 00 ft from th 0 ft will hav out to 50 ft unway 28L fference of 1 two-dimens . The profi a straight li nts do not ha r of profile p ts in profile p g the profil e line, it wi ly spaced bet estimate of 5 in calculatin n of water’s ace. two-dimens y 100 ft, a 10 s, a new grid grid points w e land grid e a flow resis will be assig with 4 ft hig /3 for 63 Hz 6-3 ional le cut ne at ve to oints oints e cut. ll not ween 0,112 g the flow ional ft by was ithin point, tivity ned a h .

6-4 The values of the calculated average flow resistivities in the ellipse using the different grid resolutions are shown in Table 8. It can be seen that increasing the grid resolution before calculating the average flow resistivity along the profile cut for this ellipse does not significantly change the difference with the weighted area (exact) method until the area of the ellipse became within two orders of magnitude of the grid area resolution. The last two frequencies with the source height at 1000 ft show how the routine for the profile cut interpolation of nearby grid points can be in error. The error is greater for smaller ellipse areas. In order to determine the variance of using different resolution grids for estimating the average flow resistivities across the parameters listed Table 7, calculations like the above were made along the runway every 500 feet out to a distance of 15,000 feet from the runway start. Both sides of the runway were considered. Graphically, the area under consideration is shown by the ellipses out to the furthest distance of 25,000 feet on either side of runway 28L in Figure 41. For clarity, only the 8 kHz results are plotted in Figure 42 which represents the percent absolute difference of the average flow resistivity estimate of a profile cut from the average flow resistivity inside the Fresnel ellipse for the geometric parameters listed in Table 7 as a function of the elliptical area. Confirming the discussion above, the largest differences occur for the lowest resolution grid spacing and smaller areas. One important note: The data points that were equal to zero were shifted to a value of 10-2 in order to plot using the log scale.

6-5 TABLE 8. Average Flow Resistivity Calculations with 4 ft High Receiver Method Grid Resolution (ft x ft) Average Flow Resistivity (kPa s/m2) Difference (%) Water (%) Frequency = 63 Hz; Ellipse Area = 2,709,251 ft2; Source Height = 10 ft Weighted Area 100 x 100 56,736 - 41.1 2D Profile Cut 100 x 100 51,809 -8.7 49.2 2D Profile Cut 10 x 10 51,637 -9.0 43.1 2D Profile Cut 1 x 1 51,666 -8.9 43.3 Frequency = 1 kHz; Ellipse Area = 668,450 ft2; Source Height = 10 ft Weighted Area 100 x 100 56,882 - 49.5 2D Profile Cut 100 x 100 52,397 -7.9 41.7 2D Profile Cut 10 x 10 52,231 -8.2 43.8 2D Profile Cut 1 x 1 52,278 -8.1 44.0 Frequency = 8 kHz; Ellipse Area = 211,794 ft2; Source Height = 10 ft Weighted Area 100 x 100 60,677 - 54.5 2D Profile Cut 100 x 100 56,023 -7.7 46.0 2D Profile Cut 10 x 10 55,829 -8.7 48.2 2D Profile Cut 1 x 1 55,859 -8.6 48.4 Frequency = 63 Hz; Ellipse Area = 2,412,965 ft2; Source Height = 100 ft Weighted Area 100 x 100 60,242 - 53.9 2D Profile Cut 100 x 100 54,675 -9.2 44.5 2D Profile Cut 10 x 10 54,473 -10.6 46.7 2D Profile Cut 1 x 1 54,488 -10.6 46.9 Frequency = 1 kHz; Ellipse Area = 216,360 ft2; Source Height = 100 ft Weighted Area 100 x 100 74,312 - 72.8 2D Profile Cut 100 x 100 67,447 -9.2 61.6 2D Profile Cut 10 x 10 66,953 -9.9 64.8 2D Profile Cut 1 x 1 67,107 -9.7 65.2 Frequency = 8 kHz; Ellipse Area = 14,670 ft2; Source Height = 100 ft Weighted Area 100 x 100 41,192 - 35.7 2D Profile Cut 100 x 100 44,353 7.7 34.5 2D Profile Cut 10 x 10 42,919 4.2 37.8 2D Profile Cut 1 x 1 43,272 5.0 38.9 Frequency = 63 Hz; Ellipse Area = 107,153 ft2; Source Height = 1000 ft Weighted Area 100 x 100 22,568 - 15.9 2D Profile Cut 100 x 100 27,173 20.4 15.8 2D Profile Cut 10 x 10 25,454 12.8 19.5 2D Profile Cut 1 x 1 25,865 14.6 20.4 Frequency = 1 kHz; Ellipse Area = 1,369 ft2; Source Height = 1000 ft Weighted Area 100 x 100 911 - 0.0 2D Profile Cut 100 x 100 1,807 98.4 0.0 2D Profile Cut 10 x 10 1,175 29.0 0.0 2D Profile Cut 1 x 1 843 -7.5 0.0 Frequency = 8 kHz; Ellipse Area = 124 ft2; Source Height = 1000 ft Weighted Area 100 x 100 225 - 0.0 2D Profile Cut 100 x 100 496 120.4 0.0 2D Profile Cut 10 x 10 225 0.0 0.0 2D Profile Cut 1 x 1 225 0.0 0.0 A comparison of the percent of the profile cut crossing water with the percent area of the Fresnel ellipses occupied by water is shown in Figure 43. The difference is greater for smaller areas and courser grid resolution. The graph only shows the comparison when the percent of the ellipse occupied by water is at or above 10 %. This is justified in that the calculation of the effective attenuation in a two- impedance area consisting of water and (soft) ground will not change by more than 1 dB when the percent of the Fresnel ellipse is occupied by 10 % water. As shown by the data points on the abscissa, the lower resolution grid can result in instances where the profile cut misses intersecting any areas of water in the ellipse. The apparent bounding curve evident in Figure 43 is a result of the inherent error between the percent the major axis is occupied by water versus the area of the ellipse occupied by water when the

water-land dependen U example s by 1 ft. showed th area in the percent h resolution the averag interface is t, no further a sing the prof hown here pr The compari at there was Fresnel ellip ard result w . Thus, a 1 f e flow resisti normal to th nalysis was p FIG ile cut metho ovide values son of estim an increase i se decreased hen there w t x 1 ft grid r vity within 1 e major axis erformed be URE 41. Ex d to estimate within 10 % ating the per n the differen . The only g as water ins esolution is r 0 % of the ex as is the case yond identify tent of cove the average of the exact cent of wate ce between rid resolution ide the Fresn equired whe act method. North of ru ing the origin rage for all e flow resistiv method for th r in the Fres the two meth for which th el ellipse a n using the p nway 28L. A of the trend llipses. ity in the Fre e highest gri nel ellipse w ods as the pe e profile cut rea was the rofile cut tec s this is situ . snel ellipse f d resolution o ith the profi rcent water method had a 100 ft by 1 hnique to est 6-6 ation or the f 1 ft le cut in the zero 00 ft imate

FIGURE 42. Difference of average flow resistivity calculations for 8 kHz. 6-7

6.2. Stra A mentioned octave ban one-third of the jet Model Par Flow Resis Elevation A Distance (f *Ranges of an E normalize be measur humidity, ight Ray M study was co above, the d. By comp octave band. spectral class TAB ameter tivity (kPa s/m ngle (degrees t) gles presented in ach of the o d to be 70 dB ed at a dista and 1 atmosp FIGURE 43 odeling nducted to d EGA model arison, the E The ground es in the AED LE 9 Range 2) ) brackets [ ]. Ang ne-third oct at 1 kHz. F nce of 1,000 here of press . Compariso etermine the utilizes a sin PD model’s p effects were T database f s of Parame les presented as a ave band sp or the sake ft from an ai ure. n of calculat difference be gle calculat ure tone app computed usi or the geome ters used wit Range 150 and 2 [0.1:0.1:1 656, 1312 range with increm ectral classe of this exerci rcraft with at ion of perce tween using ion of the gr roach require ng the EPD a tries listed in h EPD and 0,000 .0]; [2:1:20]; , 2067, 3281, ents between colo s for jet ap se, the spectr mospheric c nt hard (wat the EPD and ound effect s multiple ca nd EGA mo Table 9. EGA Model [30:10:80]; 8 6562, 13123, 1 ns. proaches and a represent t onditions of er). EGA model for each one lculations for dels based on s 5* 9685 departures he levels as w 77 F, 70% re 6-8 s. As -third each each were ould lative

In propagatio Propagati two effect for the ran conditions These effe A line of deviation model ap AEDT. F 6.3. Gro T accuracy burden lev can be us order to com n were app on effects inc s are frequen ge of parame and distanc cts were add unity slope of the differ proximate th IGURE 44. und Imped he considerat of the model el expected ed to estimat pare how th lied to the lude lateral a cy dependen ters shown in e propagated ed to the spe is plotted fo ence level of e full EPD c Plot of A-w parameters s ance Mode ions that shou and the avail for the AEDT e ground imp e two models spectra and ttenuation, at t. Lateral att Table 9. A . Spherical ctra as decibe r comparison the two mo alculations w eighted level hown in Tab l ld be balanc ability of inf user. As ex edance, but estimate the the A-weig mospheric ab enuation was tmospheric ab spreading w l corrections in the figur dels’ estimat ell for the estimates ca le 9 for all j ed when sele ormation to u plained in Se the availabil propagation hted level sorption, an calculated u sorption wa as calculate . A plot of th e. Also sho es. As can expected sou lculated by et spectral c cting a groun se it. The la ction 3, ther ity of the ne of jet noise, of the resul d spherical sp sing the EPD s calculated f d for the dis e results is sh wn are the m be seen in th rce spectra EPD vs EGA lasses in AE d impedance tter is import e are very co cessary infor the correctio t was calcu reading. Th and EGA m or the atmosp tance propag own in Figu ean and sta e figure the and geometr for the ran DT. model includ ant because mplex model mation to ex 6-9 ns for lated. e first odels heric ated. re 44. ndard EGA ies in ge of e the of the s that ercise

6-10 the models is limited. The former is also important in that small improvements in accuracy might be forgone if they require too large an effort. The ANSI standard (ANSI/ASA-S1.18, 2010) for measuring ground impedance details two models of ground impedance. The measurement uses the familiar two-microphone method with the microphones set at different heights above the ground equidistant from a sound source. The models require one or two parameters to calculate the impedance. The models work well for hard surfaces and grasslands but not as well for a variety of other surfaces. By programming the option of using the one or two-parameter ground model into the EGA code, it can be shown that the two treatments of the ground can give comparable results. It is important to note that this is a result of using the ground parameters that best match the type of ground; thus, it is something of a circular argument. It has been shown (Section 6.2) that the EGA model provides comparable results to the EPD model which is essentially the model used in the ANSI standard for checking the overall error when finding the ground types. Consider the case of industrial grass. The one-parameter model of the ground can characterize the ground with a flow resistivity of 250 kPa s/m2. The two-parameter model characterizes the same surface with an effective flow resistivity of 100 kPa s/m2 and a porosity parameter of 10 m-1. Figure 45 shows the match of the models to measured data using the NORDTEST method from Sohlmann’s study (2004). Using a flow resistivity of 250 kPa s/m2 the EPD model was exercised every 10 Hz across the frequency range shown in Figure 45. The average ground effect for the one-third octave bands with center frequencies from 200 to 2500 Hz using the EPD model is also shown. The EGA model was exercised for this geometry and flow resistivity for the one-parameter model of the ground. As can be seen, the EPD and EGA models are closely matched; furthermore, the measurements are close to the estimates of the two models. Exercising the EGA model for the two-parameter model of the ground with the above parameters shows a similar match to the measurements and other models. Using these same parameters in the EGA model, the propagation of a typical jet spectrum was estimated. The surface was assumed to be flat with the above parameters. The height of the source was set to 304.8 ft. The receiver height was set to 4 ft. The distance along the ground between source and receiver was set to be 3,281 ft. The estimate was made by adding spherical spreading, atmospheric absorption, and ground attenuation to the original spectrum. The original spectrum was arbitrarily assumed to represent levels 1,000 ft from the source. The meteorological conditions modeled were 1 atmosphere pressure, 14.5 C, and 70 % relative humidity. Figure 46 shows the spectra and levels. The atmospheric absorption decreases the high-frequency bands. As can be seen in the figure, the resulting A-weighted levels are different by 1 dB, so the difference in ground models did not significantly affect the estimated level. The one- and two-parameter models of the ground match the measurements for a number of different surfaces in Sohlman’s study (2004) with the correct input parameters.

FF IGURE 45. groun IGURE 46. 1,000 ft t Comparison d. Error ba Estimates f o 3,281 ft for of model es rs represent or propagati 1 atmosphe timates with the standar on of a repr re pressure, measureme d deviation esentative sp 14.5 C, and nts made ov of the measu ectrum of a 70% relati er grass-cov rements. jet aircraft f ve humidity. 6-11 ered rom

6-12 While there are more complex models to estimate ground impedance from a variety of parameters, the most commonly used approach for calculating ground impedance is the one-parameter model which requires an estimate of the ground flow resistivity. Because estimates of flow resistivity have been associated with the ground cover in the NLCD, the one-parameter model can be applied over large areas anywhere in the United States. The ground classifications presented by the ANSI standard or the NordTest method would serve equally as well for the confidence level one can expect in classifying large areas of land around an airport. A recent publication (Dragna et al., 2015), describes an aspect of the one-parameter model that is aphysical, namely that it does not satisfy the condition that the real part of complex density of the ground must be greater than zero. The authors point out that there are better models that require more information than what is reasonable to expect an AEDT user to know or have available. Until such time that the required parameters for those models become readily available and wide spread in coverage, it is suggested that the one-parameter model of ground impedance be used, where average flow resistivity is defined by the Fresnel model (see Section 6.1). 6.4. Model Sensitivity Studies The statistical methodology contained in the publication Design and Analysis of Experiments (DOE) (Montgomery, 1997) can be used to assist with a sensitivity study of the ground attenuation calculation method used in the EGA algorithm. The goals of the DOE and sensitivity study are to determine global response trends of the system being examined and to identify which of the input parameters are significant contributors to the output of the system. Since EGA is a deterministic computer simulation model, repeated observations from running the code with the same inputs will be identical. It is this lack of random error that makes computer experiments different from physical experiments, which leads to the use of distinct analysis techniques known as the Design and Analysis of Computer Experiments (DACE) (Sacks et al., 1989). DACE analysis is performed on deterministic systems, i.e. programs with repeatable results, without any randomization elements. The DACE techniques are used to explore the parameter space of a computer simulation (for example to perform global sensitivity analysis) with a primary goal of generating good coverage of the input parameter space. In summary, the primary goal of the current sensitivity study is the identification of significant parameters and main model effects for EGA’s ground attenuation calculation through the use of DOE/DACE techniques. Inputs to the EGA model are horizontal distance (200-25,000 feet), elevation angle (0.1-45 degrees), and ground flow resistivity (10-100,000 kPa s/m2, sampled logarithmically). Source height is a function of distance and elevation angle, and receiver height is held at a constant four feet above the ground. These continuous parameters were sampled using the Dakota (2016) program and Latin Hypercube Sampling (LHS) and were fed into the EGA routine by way of a Matlab wrapper function. A total of 150 samples were taken, equating to 150 unique EGA runs with 150 different sets of input parameters. For a single horizontal distance, elevation angle, and flow resistivity, EGA outputs the corresponding ground resistivity attenuation (in dB) for one-third octave band frequencies from 50 Hz to 10kHz. Air absorption for each one-third octave band center frequency was computed using the SAE- ARP-866B standard (SAE, 2012). Atmospheric conditions were held constant with relative humidity at 70%, temperature at 15 degrees Celsius, and air pressure at 1013.25 mbar. Example attenuation scenarios were validated against the tables presented in the SAE standard. Losses due to spherical spreading were also calculated as a function of distance between the source and receiver. For a generic fixed-wing jet aircraft source, attenuation factors due to ground, air, and distance were applied as one-third octave band level corrections. The resultant attenuated band levels were A- weighted and then summed into an overall A-weighted sound level.

A significan The softw based on than the o L 0.01 and deemed to parameter from wha values tha values and F statisticall weighted significan W angle bec again not W significan provide a n important ce test that in are John’s M results gather bserved samp ow p-values lower (99% be statistic s were not an t was compu n the normal the model’s igure 47 sho y significant ground atten t input param hen the com omes a highl significant in hen air atten t input param significant i output char dicates the i acintosh Pr ed from EGA le result (Tri indicate statis degree of co ally significa important p ted. Highly statistically resultant out ws the DOE and not stat uation values eters, wherea FIGURE puted ground y significant this scenario uation and sp eter and el nfluence ove acteristic for nfluence of i ogram (JMP) . The p-va ola, 2001). tical signific nfidence and nt by JMP’s art of the cal significant i significant p put. results from istically sign calculated fr s distance is 47. P-valu attenuation parameter an . herical sprea evation angle r the resulta a DOE or nput parame is used to c lue is the pro ance for indiv above; a co analysis (th culation proc nput paramet arameters and JMP for th ificant input om EGA, flo non-significa e results of E is applied to d flow resist ding losses a is a signifi nt overall A sensitivity ter on the va alculate p-v bability of o idual inputs mmon selec e blue vertic ess, the resul ers are input indicate a s e main effe variables. W w resistivity nt in this con GA sensitiv an example ivity is a sign re added into cant parame -weighted le study is the riation in the alues for eac btaining a m on the outpu tion in statis al line in Fig t would be d variables w trong correla cts of the EG hen examin and elevatio text. ity analysis. jet aircraft s ificant param the model, d ter. Flow re vel in this c p-value fo examined re h input para ore extreme t results. P-v tical analysi ure 47). If rastically dif ith much low tion between A model an ing the overa n angle are h pectrum, elev eter. Distan istance is a h sistivity doe ase. Atmosp 6-13 r the sults. meter result alues s) are these ferent er p- their d its ll A- ighly ation ce is ighly s not heric

6-14 effects and spherical spreading, neither of which are functions of flow resistivity, dominate the source attenuation computation obscuring ground impedance effects. 6.5. Integrated Modeling Issues An examination and exercise of the acoustic algorithms in AEDT (Koopman et al., 2015) and INM (Boeker et al., 2008) was conducted to:  Determine the effect segment length has on noise metric prediction uncertainty, and  Identify a suitable manner in which an alternate ground effect propagation algorithm can be incorporated into AEDT. 6.5.1. Background Integrated modeling relies on pre-propagated noise-power-distance data from infinite length, constant condition, flight segments. The noise fraction (NF) determines the percentage of the infinite segment noise exposure that results from the finite length segment and is based on segment-receiver geometry. Within AEDT these contributions are summed for each modeled flight segment to obtain the total noise exposure. Capturing changes in geometric acoustic propagation features, such as varying ground impedance, requires a reduction in the segment length commensurate with the geometric feature to be modeled. For uniform ground impedance, infinite segments are suitable. However to model a shoreline and the corresponding acoustically hard and soft interface, one must consider the modeled fidelity of the ground impedance. For example, is 2000, 1000, 100 or even 20 foot impedance grid spacing required to capture the modeling results with the desired fidelity? There is a direct geometric link between the desired output fidelity and the maximum segment length that must be used in the integrated model. In the limit, as flight segment lengths are reduced, errors are introduced due to such pragmatic issues as numerical precision, discretization, algorithm implementation and fundamental assumptions in the noise fraction formulation. An assessment of the effect of segment length on integrated noise modeling prediction consistency is presented below. 6.5.2. Effect of Segment Length Modeling in AEDT/INM A set of analyses were performed using a special diagnostic version of INM 7.0d, which provides additional information regarding the elements of the noise calculations. These analyses modeled the notional operation itemized in Figure 48, which is a 10,000 ft long segment at 250 ft and 1000 ft AGL, with a B737-300 (utilizing spectral class CFM563) flight at 250 knots and 19,000 lbs. thrust over a uniform flat soft ground with 59 F and 70 % RH at a sea-level airfield. A custom module was developed that modified the TRK_SEGS.DBF file from the single 10,000 ft long segment to segments of varying length 5,000, 1,000, 500, 100, and 20 ft in the INM study.

A single eve other leng difference Figure 51 model the noise rang segments highlighti operation using a gr similar to expected. receptor gri nt SEL (dBA ths indicated between the . One can ob same 10,000 es from a te (Figure 51e) ng the areas uncertainty id density of those presen d covering 2 ) results are above and I se results an serve that as ft operation nth of a dB f . Moreover of most unc is within 0.7 21 x 21 poi ted with the FIGURE nmi x 2 nmi presented in NM was reru d the baseline the segmen , there is a bi or two 5,000 , this incons ertainty. In 5 dB within nts and the r higher grid d 48. Modele comprised o Figure 49. T n. These res (10,000 ft) t length is de as introduced ft segments istency is co regions abea 0.75 nmi la esults are pr ensity analys d operation. f 51 x 51 gr he single seg ults are pres result yields creased, and in the resul to more than ncentrated n m of the flig teral distance esented in Fi is, but with d id points was ment was th ented in Figu the differenc more segme ts: an increa 1.3 dB for o ear the ends ht segment, s. The anal gure 52. Th ifferent tem modeled, an en modified re 50. Takin e results sho nts are requir se in the pred ne hundred of the segm for a 250 ft ysis was rep e uncertaintie poral resoluti 6-15 d the to the g the wn in ed to icted 100 ft ents, AGL eated s are on as

T results pre the ends geometric formulatio increased threshold. he vehicle fl sented in Fi of the flight area of influ n in the AED fidelity, how FIGURE ight altitude gure 53 indic track, and a ence. One T integrated ever the noi 49. Baseline single 10,00 was then inc ate a reducti reduction in must be awa model. Red se fraction fu results SEL 0 ft segmen reased to 1,0 on in the pea the lateral u re of the unc ucing segmen ndamental fo (dBA), 250 t, 51 x 51 gri 00 ft AGL a k uncertainty ncertainty, b ertainties in t sizes furth rmulation su ft overflight d. nd the analy to 1.1 dBA ut demonstr the fundame er might give ggests a min , sis repeated. in the region ate a much ntal segmen the appearan imum uncer 6-16 The s off larger tation ce of tainty

FIGURE 50. altitude: a) INM grid re 5,000 ft b) 1, sults (SEL, 000 ft c) 500 dBA) for dif ft d) 200 ft e ferent segme ) 100 ft f) 20 nt lengths, 5 ft. X and Y 1x51 grid, 2 distances in 6-17 50 ft ft.

FIGURE 51. ft alti Baseline SE tude. 1) 5,00 L (dBA) mi 0 ft b) 1,000 d nus different ft c) 500 ft d istances in n segment len ) 200 ft e) 10 mi. gth results, 0 ft f) 20 ft. 51 x 51 grid X and Y 6-18 , 250

FIGURE 52 250 ft a . Baseline S ltitude. 1) 5, EL (dBA) m 000 ft b) 1,0 d inus differen 00 ft c) 500 f istances in n t segmented t d) 200 ft e) mi. length resu 100 ft f) 20 lts, 21 x 21 g ft. X and Y 6-19 rid,

6.5.3. AE T al., 2008) infinite se lengths. T variation for the 20 have on i difference FIGURE 53 1000 ft a DT Noise F he noise fract may be expr gment. Figu his in effect in the noise f ft segments. ntegrated res s between sm . Baseline S ltitude. a) 5 raction and ion (NF) form essed as the re 55 illustra represents th raction scales One can rea ults, especia all angles. EL (dBA) m ,000 ft b) 1,0 d Ground Eff ulation (Fig ratio of the a tes the noise e percentage in Figure 55 dily see the in lly in the re inus differe 00 ft c) 500 f istances in n ect Impleme ure 54) (Eldr coustic energ fraction for a of the NPD e cover 5% to fluence geom gions off the nt segmented t d) 200 ft e) mi. ntation ed, 1980; Eld y of the fini single segm xposure that 90% for 1,0 etric numer end of the length resu 100 ft f) 20 red & Mille te segment to ent of 1,000 is summed in 00 ft segmen ical precision segments wh lts, 21x21 gr ft. X and Y r, 1980; Plotk the energy ft, 200ft and AEDT – no ts and 1% to and accurac ere one is t 6-20 id, in et. of the 20 ft te the 1.6% y will aking

W Adjustme Ground A INM Tech long segm The relati combined ithin INM nt” which is ttenuation an nical Manua ent. The ne ve impact of effect. FIGURE F and AEDT reflected by d Overall La l (Boeker et w ground eff the Overgrou 55. Noise fr IGURE 54. the propagat the compo teral Attenu al., 2008) an ects model d nd and Long action snap Noise Fract ion effects nents of Ov ation (Equati d illustrated g irectly replac Range Air t shot from a s ion Formula are embodie erground At on 3-28, 3-2 raphically in es the “Ove o Ground co ingle segme tion. d in the “L tenuation, Lo 9 and 3-33 r Figure 56 fo rall Lateral A mponents are nt with diffe ateral Attenu ng Range A espectively) r the first 1,0 ttenuation” p shown with rent lengths 6-21 ation ir to in the 00 ft iece. their .

F6.5.4. App T AEDT: a A Attenuatio AEDT Ac Geometric and acous T uncertaint IGURE 56. lication an here are two lgorithm imp lgorithm: A n” (Equatio oustic Modu Application tic uncertaint  The m imped the ma  A rep analys fractio (simila provid the ex altitud  The gr geospa imped In the will be used.  The ge consid he segment le y recognizin INM NPD la d Integrated elements as lementation a new ground n 3-33 in th le and be un : The follow ies considerin aximum cha ance should b ximum chang resentative s is and operat n angles). r to the anal es a relations ample shown e over approx ound impeda tial fidelity ance changes limit for a un longest side ometric fide ered. ngth used in g that as se teral attenu f Modeling I sociated with nd geometric effect algor e INM Tech iformly app ing steps sho g noise fract nge in prop e determined e in acoustic et of source ions under co The Noise F ysis presente hip between earlier, 100 imately 0.5 n nce modelin of the impe are defined. iform imped irrespective lity — for w AEDT shoul gment length ation “adjus t segment len ssues implementa application. ithm in AED nical Manua licable to all uld be taken ion and grou agation effec for the sour results (delt receiver ge nsideration ( raction unce d in this res NF uncertain ft segment mi while 5,0 g length scal dance grid. Input fidelity ance grid (on of whether a hich the aco d be the larg is decrease tment.” 51 x gth. tion of a ne T should su l) portion of metrics (SE in determini nd impedanc t for the mi ce receiver g a-dB) due to ometry (SRG source heigh rtainty for t earch) for un ty (delta-dB) lengths yield 00 ft segmen e (IMLS) sh The IMLS is not the sa ly 4 corners higher fidel ustic results est possible t d, acoustic 51 grid, 250 w ground ef pplement th the acoustic L, LmaxA, E ng the variou e: nimum and eometries (S ground effect ) should be ts, receiver d he SRG sho iform ground and segmen 1.3 dB unc ts yield 0.1 d ould be evalu is the dis me as the inp need to be d ity grid with are to be co o achieve the uncertainty ft altitude, fects algorith e “Overall L algorithm i PNL, PNLT s segment le maximum g RG). This de physics. selected fo istances and uld be quan impedance t length scale ertainty for 2 B uncertainty ated based o tance over w ut grid resol efined), the I constant val mputed shou lowest achie due to the N 6-22 1,000 m in ateral n the etc). ngths round fines r the noise tified . This . For 50 ft . n the hich ution. MLS ues is ld be vable F is

6-23 increased. The NF uncertainty sets both the minimum uncertainty possible and the smallest required segment size while the ground effect physics and IMLS will potentially permit longer segments without increasing uncertainty. The point-to-point attenuation accounting for the actual intervening terrain with the specified ground impedance will need to be calculated for each segment end point and the physical (on-segment) point of closest approach. The maximum attenuation of the three should be selected and applied, as is the protocol in NOISEMAP. 6.6. Prioritized List of AEDT Improvements This section of the report details what the Research Team has identified as a list of potential improvements for AEDT. As detailed above, integrated modeling relies on pre-propagated noise-power-distance results from infinite length, constant condition, flight segments. The noise fraction determines, based on segment-receiver geometry, the percentage of the infinite segment noise exposure that results from the finite length segment. Within AEDT these contributions are summed for each modeled flight segment to obtain the total noise exposure. Capturing changes in geometric acoustic propagation features, such as varying ground impedance, requires a reduction in the segment length commensurate with the geometric feature to be modeled. For uniform ground impedance, infinite segments are sufficient. However to model a shoreline and the corresponding acoustically hard and soft interface, one must consider the modeled fidelity of the ground impedance. For example, the above study showed that a grid resolution of 10 ft reduced the disparity between the 2D profile cut and Fresnel ellipse calculation to 10 % for most one-third octave bands. There is a direct geometric link between the desired output fidelity and the maximum segment length that must be used in the integrated model. The study above utilized a method for calculating the average flow resistivity in the Fresnel ellipse that relied on grids with no interpolation at the grid boundaries. It is possible that a different methodology could be employed that did not utilize grids. A vector-based characterization of the flow resistivity of the ground could be used to eliminate the reliance on grids resolution altogether. Such an implementation would be more exact in determining the ratio of the profile cut that was hard versus soft at a water-land boundary. The investigation here relied on the grid-based methods used in most models that consider terrain. 6.6.1. Importing Terrain Data AEDT currently allows for importing terrain data in three formats: a grid float format, the Digital Elevation Model (DEM) file format from USGS, and the 3cd file format (also a USGS file format). All three file formats represent ground elevation with a two-dimensional grid. The location of each elevation in the grid is georeferenced using an origin with real world coordinates and fixed spacing in each dimension of the grid. This allows for the identification of an elevation profile between source and receiver by interpolating elevation values along a line on the ground starting below the source to the receiver. The grid float format is a generic format while the other two are obsolete file formats previously supported by the USGS. In order to identify the impedance profile between the source and receiver a similar grid as the elevation data should exist which contains the necessary parameters for characterizing the ground. This will allow using the same routines to interrogate the data to find the impedance profile from source to receiver.

6-24 The study by Plotkin et al. (2013) outlined methods that would be useful for interrogating data on a grid. As noted above, the National Land Cover Database (NLCD) is an available set of data with ground cover classification. Care should be taken in identifying how the data is presented to AEDT regardless of its source. Because the file formats at USGS change over the years, it would likely be best for AEDT users if there were a preprocessor to create a generic file format like the grid float to enter the data into the program. This would simplify maintenance of the code itself by using a preprocessor to obtain the data much the same way that BASEOPS does for NMAP. A separate issue relates to the determination of the parameters that characterize the ground cover. The study by Sohlman et al. (2004) provides a good example on how to verify the impedance of ground cover originally identified based on satellite data. The results of this project may help with the classification scheme. Notable is the binary use of hard/soft by NMAP which is based upon Plovsing’s parametrization of Rasmussen’s theory (Delta, 1993). Two practical asides in regards to the importation and use of terrain data:  A method should be provided by the terrain module to identify impedance values graphically. For example, an image of a map showing water and land in a graphical user interface would allow the user to assign a flow resistivity to the two areas. In the event that it is a complex scattering of the two surfaces, the user would be able to graphically ‘draw’ the boundaries that delineated the two types of surfaces. This functionality exists in BASEOPS.  The interpolation of grid point data to find ground parameters should consider how to delineate the edges of geographic features. For example, the edge of runways and taxiways at airports delineate hard ground. There should be a way for the program to associate those boundaries with a high flow resistivity. As important is whether those boundaries are treated as finite. This would affect whether a location between a pavement edge and soft ground would be an interpolation between the values or associated with the soft ground since the pavement edge is well defined. The grid density of the NLCD is 30 m. The interpolation of any values between a pavement edge and soft ground would introduce an uncertainty in the location of the impedance discontinuity of that order. 6.6.2. Method to Determine Terrain Profile If a grid-based representation of the flow resistivity of the ground is used by AEDT, then the algorithms used by NOISEMAP to interrogate terrain data are very straightforward and could be used by AEDT to identify the impedance profile along the ground from beneath the source to the receiver. One caveat discovered in this research is that the definition of the flow resistivity at each grid point must be maintained to the boundary of the grid point with its neighbors. This is because there should be no interpolation of flow resistivities between grid points. Whether that is enforced by creating a finer resolution grid based on the same information as was done above or achieved by altering the algorithm to interrogate the terrain will be up to the AEDT developers. As part of the incorporation of a new lateral attenuation adjustment based upon mixed-surface impedance, AEDT should be allowed to switch between using the results from the mixed impedance surface and turning it off and using the current lateral attenuation adjustments (SAE-AIR-5662) to allow for comparison testing. In general, the updated ground effects could be a drop-in replacement for the current lateral attenuation adjustments. A separate engine installation directivity effect will need to

6-25 supplement the new ground effects method(s) because AEDT currently includes engine installation directivity as part of the lateral attenuation. 6.6.3. Flexible Segmentation of Flight Path As a result of the research on the point-to-line source application of the ground effect, it will be necessary to have AEDT further refine the length of the line segments of the flight path. In the study by Plotkin et al. (2013) the relationship between applying the lateral attenuation calculated from points on various length segments was explored. The study found little error in applying the lateral attenuation for three points on a line segment (both ends and the closest point of approach) for segments up to 2,000 feet in length. The further study by the Research Team showed an uncertainty between segment length and source to receiver geometry when considering the variation of the noise fraction. AEDT will have to be changed in a way that flight path segments are reduced from what is input to the model. 6.7. Data Validation As evidenced by the TNM results presented in Section 4, the change in sound level over variable impedance terrain can be quantified using straight ray theory. The EPD model was used to expand the INM database as part of the process to generate the noise-power-distance curves (Bishop, et al., 1986); thus, straight ray theory is already an integral part of AEDT’s noise engine. When the study at Washington National Airport (DCA) by Downing (Downing, et al., 2004) showed that NMAP accurately predicted sound propagation over water at laterally placed noise monitors for civilian aircraft out to an elevation angle of 5 degrees, the case for the efficacy of straight ray theory became very strong. The Research Team for this project was encouraged to find airport data to validate the theory it was suggesting for modeling propagation over mixed-type ground. Searching for the data revealed that the best chance for obtaining airport data was from an airport with an ANOMS system. With the help from a Panel Member and Noise Officer at an airport, data were obtained with the necessary spectral time history and tracking data needed to exercise the theory over a mixture of surfaces. The participating airports are Portland International Airport (PDX), San Francisco International Airport (SFO), and Oakland International Airport (OAK). 6.7.1. Data Validation at PDX The airport has installed Brüel & Kjær’s Airport Noise Monitoring and Management System (ANOMS). The ANOMS data obtained from Portland International Airport (PDX) includes meteorological data, noise monitor data, and aircraft tracking data. The noise monitors provided A- weighted time history data for the duration of the interval. They also provided spectral time history data. At PDX, the closest monitors to the airport that are appropriate for measuring lateral attenuation for this project are noted as 107 and 101 in Figure 32 (Section 5.2.3). Aircraft departures from runway 28R fly directly over monitor 107; thus, the noise emitted from the aircraft at the point of closest approach to monitor 107 can be measured at monitor 101 to determine the lateral attenuation. Figure 57 shows the elevation angle at monitor 101 compared to 107. The range of elevation angles at the lateral monitor (101) is from approximately 20 to 40 degrees. It would have been preferable to have angles of elevation even lower in the range, but the Research Team did not find an airport with monitors closer to the airport with a similar setup as the one found at PDX.

The following  Events 107 is  The tim were u slant d temper  The N files w ellipse was fo the elli  The E ground ellipse  The at calcula sound airplan added distanc FIGU procedure wa where the w less than 70 d e of an aircr sed to find th istances to th ature) to find ational Land ith flow resi s for the cent und using the pses occupie GA model w effect. The s associated w mospheric ab ted using me traveled from e to monitor back to the sp e. RE 57. Ele s used to exe ind speed is egrees were aft’s point o e arrival time e monitors w the arrival ti Cover Datab stivity and el er frequencie geometry o d by water w as used to c EGA model ith one-third sorption usi teorological the airplan 107. The c ectrum from vation Angl rcise the EG greater than discarded. f closest appr s of sound fr ere used wi mes. ase and Natio evation data. s of the one- f the aircraft ere also calcu orrect the so used the ave octave band ng SAE AIR data at the tim e to monitor alculated lev 101. This ‘r es at PDX M A model usin 10 knots and oach to moni om the aircra th the sound nal Elevatio The averag third octave b to each of the lated. und spectrum rage flow re s from 50 Hz 5534 and s e of the eve 101 as com els for these emoves’ the onitors. g the PDX d the elevatio tor 107 and t ft at each of speed (calcu n Data were e flow resisti ands from 5 monitors. T measured a sistivity for e to 10,000 H pherical spre nt along with pared to the two propag effects of pro ataset: n angle at mo he air tempe the monitors lated from t used to creat vity in the Fr 0 Hz to 10,00 he percent a t the monito ach of the Fr z. ading effect the extra dis distance from ation effects pagating the 6-26 nitor rature . The he air e grid esnel 0 Hz rea of rs for esnel were tance the were extra

6-27  The A-weighted levels of the resulting spectra were calculated and corrected for the engine installation effect according to SAE AIR 5662. Figure 58 shows the result of these calculations for each event. If the A-weighted level from monitor 101’s corrected spectrum equals monitor 107’s, then the EGA model is correctly accounting for lateral attenuation. There were 369 departures from runway 28R of fixed-wing aircraft that met the wind and elevation angle criteria outlined above during the first week of July 2016 at PDX. The majority, 300 of the 369 events, of the aircraft were Airbus and Boeing narrow-body jets. The area around monitor 107 is grass and sand; thus, it was modeled as soft ground. The reason that the estimates for flow resistivity associated with the NLCD ground cover were not used for the ground flow resistivity under the monitor was because the granularity of the NLCD grid had the monitor’s location in a grid cell of water. This was discovered using satellite imagery of the area. The flow resistivity of the area around Monitor 101 was determined by the impedance file created from BASEOPS using estimates associated with the NLCD ground cover. The file’s estimates of the flow resistivity of the area around the monitor were 19,500 kPa s/m2. To see if the difference of the monitor’s corrected levels follows a normal distribution, a probability plot was created (Figure 59). As can be seen in the figure, the difference levels exhibit normal behavior near the center of the distribution with long tails away from normal toward the ends of the range. The mean difference level for this dataset is 0.1 dB with a standard deviation of 2.0 dB. Performing a similar calculation on only the narrow-body jets resulted in the same mean and a standard deviation of 1.9 dB. The spread in this data is very close to that measured in the Boston study by Fleming et al. (2002) as can be seen in Figure 31 (Section 5.2.1).

FIGURE 58. Corrected Levels at PDX Monitors. Compariso Dashed Lin n. e with Unity Slope Draw 6-28 n for

F6.7.2. Da T Monitorin meteorolo provided history da airfield an S would be propagatio acting as measured the levels noise leve IGURE 59. an ta Validatio he San Fran g and Mana gical data, n A-weighted t ta. As show d a number o everal geome useful for thi n to NMT11 a reference, from the mo measured at ls. Normal Pro d 107 at PDX n at SFO cisco Interna gement Sys oise monitor ime history d n in Figure f Noise Mon tries and pa s project. De because of t albeit at gra nitors when t NMT11 whe bability Plot . Dash-dot tional Airpo tem (ANOM data, and air ata for the d 39 (Section itor Terminal irs of noise partures from he fact that n zing inciden he airplane w n the airplan of Differenc ted Line Rep rt (SFO) ha S). The A craft tracking uration of the 5.2.8), there s (NMT) aro monitors we either runw early half is o ce, a good c as between e is closest t e of Correct resents Gau s installed B NOMS data data for Au interval. T are two grou und the airpo re investigat ay 28 L or R ver water an omparison c them; howev o the monito ed Levels fr ssian distrib rüel & Kjæ obtained fr gust 2016. T hey also prov nd run-up un rt. ed to determ would have b d half over la ould be mad er, as can be rs are too cl om Monitor ution. r’s Airport om SFO inc he noise mo ided spectra its (GRU) o ine whether een a good t nd. With GR e with the seen in Figu ose to the am 6-29 s 101 Noise ludes nitors l time n the they est of U35 levels re 60, bient

l FIGURE 60 ines represen . A-weighted t arrival tim time histor es of sound point ies at GRU 3 emitted from of closest ap 5 and NMT B747-8 dep proach. 11 noise mo arture from nitors. Vert Runway 28 6-30 ical R at

6-31 The slant range at the point of closest approach to GRU35 for a departure from runway 28R is 387 ft. Similarly, the slant range to NMT11 is 11, 317 ft. The sound resulting in the maximum level measured by NMT11 shown in Figure 60 was emitted from the aircraft when it was much further down the runway as evidenced by the fact that the peak level was measured by the monitor several seconds after the arrival of sound from the aircraft at it point of closest approach. This means the aircraft will be going much faster and emitting a different noise level than it was at the beginning of the runway near GRU35. This makes accounting for lateral attenuation using the maximum sound levels from the two monitors impossible. Departures on runways 10 L and R were investigated to see if noise measurements at GRU35 and NMT11 could be utilized. There were only 23 departures, and the noise measured by NMT11 was also unusable for the same reason as described for runway 28. The other pair of monitors on either side of the East-West runways are GRU34 and NMT8. While the same Boeing 747-800 opertion discussed above was measured at GRU34, it was not distiguishable in the time history data at NMT08, nor were any other operations from these runways. This is due to the high ambient noise levels at the monitor and a slant range of 7,500 ft to runways 28R. The remaining possibility to determine if the EGA model correctly predicted lateral attenuation was to treat NMT01 and NMT05 in like manner as at PDX. NMT01 is inline with runway 28L and NMT05 is laterally displaced by 2,000 ft from NMT01. Using a similar procedure to cull the ANOMS data as was used at PDX, a data set was found to exercise the model. The procedure used the same test to reject an event if the wind speed was too high, but because the SFO monitors did not continuously collect spectral time histories, the test to determine if the event noise measured at the monitor was above the ambient noise levels relied upon the A-weighted level time history data which was collected continuously. Figure 61 shows the elevation angles at the two monitors for operations that occurred when the wind speed was less than or equal to 10 knots and the A-weighted sound level received from the aircraft at its point of closest approach to the monitors was at least 10 dB above the ambient levels. The data set comprises 532 operations. All but 5 of the operations are narrow and wide-body jet aircraft. The area around the monitors is classified as Developed Medium Intensity by the National Land Cover Database and is estimated to have a flow resistivity of 19,500 kPa s/m2. The area between the monitors is classified as Developed High Intensity with a flow resistivity of 25,500 kPa s/m2. The comparison for the SFO data follows a similar procudure outlined for the data at PDX above:  Spectra captured by the monitors emitted from the aircraft at the closest point of approach were corrected for ground effect using the EGA model with flow resitivities mentioned above.  The spectrum from the lateral monitor (NMT05) was further corrected for atmospheric absorption and spherical spreading for the extra distance the sound traveled beyond the distance from the aircraft to the undertrack monitor (NMT01).  The A-weighted levels of both corrected spectra were calculated.  The engine installation effects SAE AIR 5662 (SAE, 2006) were removed from the A- weighted sound levels found from the corrected spectra.  The corrected levels were compared.

6-32 As before, if the corrected level from the lateral monitor matches the one from the undertrack monitor, then the EGA model is accounting for lateral attenuation correctly. Figure 62 shows a comparison of the corrected levels with the mean and standard deviation of the difference levels. The standard deviation is similar to that found for the data at PDX above, but the mean shows a -0.8 dB bias. The negative bias in the difference levels means the corrected levels from the lateral monitor (NMT05) are smaller than the corrected levels from the undertrack monitor. This may indicate that the EGA routine is not accounting for enough lateral attenuation at NMT05 or it is accounting for too much at NMT01. To further explore this bias, a constant flow resitivity of 150 kPa s/m2 was used for the whole area in place of the estimates associated with the NLCD ground cover. The results are plotted in Figure 63 along with the mean and standard deviation. Using soft ground for the processing of the SFO data results in a mean of -0.1 dB. This shows a 0.7 dB shift in the mean difference level by using soft ground cover. An explanation of why the soft ground resulted in better agreement between the corrected levels may be found by looking at the area around the monitors. Figure 64 show that the area around NMT05 is uniform one story buildings. Given that the roof tops of those buildings appear to be asphault, one would expect that the flow resistivity of the surface would be much lower than the 19,500 kPa s/m2 estimated by the impedance file created from BASEOPS using the flow resistivity estimates associated with the NLCD ground cover classification. Additionally, there is a second aspect of the ground surface that has not been addressed in this project: roughness. The scattering of sound from a surface will reduce the energy of the reflected sound at the receiver. While this appears to have the same effect as reducing the flow resistivity of the surface, it is not the same effect physically. The roughness of a surface is considered a feature best addresed by terrain characterization rather than be focused solely on the impedance as in the present study. While using the estimates for flow resitivity associated with the ground cover in the NLCD results in a mean difference of the correctel levels, it is less than 1 dB and should be considered acceptable for a measurement that is not dedicated to the purpose of exploring a specific phenomenon.

FIGURE 61. Elevation angles of operations satisf criteria at SF ying wind sp O. eed and signal to noise l 6-33 evel

FIGURE 62. a Compariso round SFO. n of correcte Dashed lin d sound leve e with unity ls using the slope drawn estimates of for compar flow resistiv ison. 6-34 ities

FIGURE 63. Compa SFO rison of cor data set. Da rected sound shed line wi levels using th unity slop a flow resis e drawn for tivity of 150 comparison kPa s/m2 for . 6-35 the

6.7.3. Da T data, nois weighted history da Monitor T T propagatio monitors promising ta Validatio he ANOMS e monitor da time history ta (see below erminals (NM he focus of an n over mix were investi pair consist F n at OAK data obtained ta, and aircr data for the ). There ar T) around t alyzing the d ed impedanc gated to dete ed of an ext IGURE 64. from Oakla aft tracking d duration of t e ground run he airport. ata from Oa e surfaces. rmine wheth raordinary gr Area around nd Internatio ata for Aug he interval, a -up units (G kland Interna As with SF er they wou ound run-up NMT05 at nal Airport ( ust 2016. T long with lim RU) on the a tional Airpor O, several g ld be usefu noise monit SFO. OAK) includ he noise mon ited amoun irfield and a t (OAK) was eometries a l for this an or (GRU36) es meteorolo itors provid ts of spectral number of ground-to-g nd pairs of alysis. The and a comm 6-36 gical ed A- time Noise round noise most unity

6-37 noise monitor (NMT12). GRU36 was placed on the airfield to capture the noise from jet departures on runway 30 as shown in Figure 65. The community monitor, NMT12, is 8,363 ft from the threshold. A red line connecting the two monitors in Figure 65 shows the relative angle of emission from the nose of the aircraft assuming the maximum noise level is emitted when the aircraft is at the threshold of the runway. GRU36 is 714 ft from the threshold of runway 30 The ANOMS data gathered at OAK is different from that at PDX in that the noise monitors gather time histories of the A-weighted sound levels continuously, but they only collected spectral time history data when the A-weighted level exceeded a threshold. This has two implications: 1) there would be no ambient spectrum to subtract from the event spectrum, and 2) there was almost no spectral data collected at the community monitor (NMT12) because A- weighted levels of the noise from departures off runway 30 were below the threshold to trigger spectral data capture at the monitor. Due to the issue cited above, an alternate pair of monitors was considered for this analysis. As can be seen in Figure 65, there is another noise monitor across the water from the runway: NMT02. However, it was considered inappropriate to analyze the data captured by the NMT02 monitor as the noise emitted from the aircraft that arrives at that monitor is from a different angle relative to the nose of the aircraft. Because the noise emitted from jets is directional, the noise captured by GRU36 is going to be several decibels different from the noise emitted by the aircraft in the direction of NMT02. The data from a number of other monitors at OAK were provided with the data set representing noise and operations at OAK for August 2016, but the other monitors were too far from runway 30 to differentiate noise from the aircraft and noise from the community. Despite the lack of spectral data at NMT12, a validation dataset was developed using the monitor pair NMT12 and GRU36. The process to apply straight ray theory and analyze the results to see if EGA correctly predicts lateral attenuation had to be altered because of this lack of spectral data at NMT12. Instead of correcting the distant monitor’s data for lateral attenuation and propagation effects back to the reference monitor for comparison, the opposite was done. The reference monitor’s spectra were corrected for lateral attenuation and the effects of propagation to the further monitor were subtracted. Because the reference monitor, GRU36, is located at grazing incidence to the aircraft, it is also a test of the EGA model’s ability to account for lateral attenuation. The following procedure was used:  Exclude all events outside 00:00 and 06:00 because the A-weighted ambient noise levels at NMT12 are too high outside these hours to differentiate the noise from departing aircraft off runway 30.  Exclude all events where the maximum A-weighted sound level at either monitor did not exceed the ambient A-weighted level by 10 dB. The ambient level was, defined as the minimum A-weighted sound level during the previous 85 s in the time history data of the monitor.  Exclude all events during times when the wind speed exceeded 10 knots.  It is assumed that the maximum A-weighted sound level measured at GRU36 was emitted from the aircraft when it was at the threshold of runway 30.  The spectrum recorded by GRU36 at the time of the maximum A-weighted sound level had the ground effect removed by subtracting the output of EGA from the spectrum.

If the model F be seen in are higher be seen in  Subtra distanc  Apply runwa  Calcul maxim the A-weigh ing of EGA f FIGURE igure 66 show the figure, th than those m the figure, t ct from the e from GRU ground effec y 30 and NM ate the A-w um level me ted level of th or the ground 65. Noise m s a compari ere is a distin easured. A he direct line corrected sp 36 to NMT12 t using EGA T12. eighted leve asured at NM e corrected s effect to GR onitor placem son of the lev ct bias in the closer look a of site to th ectrum air a . for the geo l of the co T12. pectrum from U36 and NM ent at OAK Runway 3 els resulting distribution t the area aro e runway is b bsorption an metry and te rrected spec GRU36 ma T12 is corre to capture 0. from the pro of the data sh und NMT12 locked by ho d spherical rrain betwee trum and c tches the lev ct. noise from d cedure outlin owing that t is shown in uses. To se spreading fo n the thresho ompare with el at NMT12 epartures o ed above. A he corrected Figure 67. A e if this elim 6-38 r the ld of the , then n s can levels s can inates

the groun corrected to it. The to the spe the correc probabilit distributio T the measu from the 0.0 dB w effect at g F T It may als the maxim would be measuring model co incidence d effect from for ground ef ground effe ctrum. Figur ted levels no y plot was c n with consid here were 27 red sound le airplane to th ith a standar razing incide IGURE 66. he spread in t o indicate th um level at to have obser the noise be rrectly asses . The estima attenuating fect to the m ct calculated e 68 shows longer show reated and i erable varian 4 events that vels at NMT e airfield mo d deviation o nce. Comparison Dashe he data is rep e uncertainty the monitors vations of th fore the hou sed ground ted flow resi the sound onitor and ha for propagati a comparison the bias evid s displayed ce. met the crit 12 and the G nitor along w f 4.8 dB. T of correcte d line with u resentative o in the locati . A method e aircraft loc sing at the w effect on th stivity associ of departing d only the ai on from the following th ent in Figur in Figure 69 eria using the RU36 levels ith air absor his would in d levels from nity slope d f the conside on of the airp to improve t ation when it ater’s edge; e airfield fo ated with the aircraft, th r absorption plane to NM is procedure e 66. To see . This figu above proc that were o ption and sp dicate that th GRU36 an rawn for com rable distanc lane when it he understan emits the ma otherwise, as r airplane no NLCD grou e spectrum f and spherical T12 by EGA . As can be if this distrib re shows a r edure. The m nly corrected herical sprea e housing d d measured parison. e over which emitted the ding of what ximum soun the evidenc ise measure nd cover of rom GRU36 spreading ap was NOT ap seen in the f ution is norm easonable n ean differen for ground ding to NMT isrupts the g levels at NM the sound tr sound captur is happening d levels as w e stands, the d at near-gr the area is a 6-39 was plied plied igure, al, a ormal ce of effect 12 is round T12. avels. ed as here ell as EGA azing round

14,000 kP monitor lo high flow a s/m2 for cation does resistivity. all Fresnel z appear to hav FIGUR one ellipses. e a high perc E 67. Area a The ground entage of tax round NMT cover betw iway coverin 12 with line een the runw g the area w to GRU36. ay threshol hich should h 6-40 d and ave a

FIGURE 68 airfield, air a . Comparis bsorption, a lin on of levels f nd spherica e with unity rom GRU36 l spreading w slope draw corrected o ith measur n for compa nly for grou ed levels at N rison. nd effect on MT12. Das 6-41 the hed

FIGU ground RE 69. Pro effect, atmo bability plot spheric abso Dash-dot of difference rption and ted line repr between GR spherical spr esents Gaus U36 levels eading and sian distribu only correct measured le tion. ed for airfiel vels at NMT 6-42 d 12.