Validation of three aircraft noise calculation models Jonas Meister 1 , Stefan Schalcher, Jean-Marc Wunderli, Beat Schäffer Empa, Swiss Federal Laboratories for Materials Science and Technology Ueberlandstrasse 129, 8600 Dübendorf, Switzerland

ABSTRACT Aircraft noise affects large areas around airports. Noise calculation programs therefore need to account for air operations with high accuracy. In this contribution, we simulated several thousand single flights with three aircraft noise calculation models, namely sonAIR, FLULA2 and AEDT, and compared the results among each other and with corresponding noise measurements to evaluate their accuracy. While FLULA2 and sonAIR calculations are adjusted to local conditions, AEDT was run with default settings (NPD data and standard procedural profiles) to assess its performance when using unadjusted input data. We found that results of all three programs on average agree well with measurements and with each other. sonAIR is more accurate when using flight configuration data, but also in greater distances to the airports. Exemplarily adaptations of AEDT input data to local conditions showed that such adaptations clearly improve its calculation results. The contribution thus shows that, despite their inherently different modelling approaches, (a) all three programs are on average equally capable of reproducing the sound exposure in the vicinity of airports, (b) adaptions of input data to local conditions substantially improve results and (c) sonAIR is able to reproduce single flights with higher accuracy when provided with flight configuration data. 1. INTRODUCTION

Aircraft noise affects large areas and is hardly shielded by obstacles on the ground. Around 4 million people in Europe were estimated to be exposed to aircraft noise levels ( L den ) of 55 dB or higher [1]. Further, the most likely scenario according to EUROCONTROL [2] is a 44% increase in air traffic by 2050 compared to 2019, which will also increase the number of affected people. Since aircraft noise has negative public health effects, the WHO recommends to reduce aircraft noise below the an L den of 53 dB [3].

To be able to monitor aircraft noise for large areas, accurate calculation models are required. They need to represent real aircraft noise very accurately, since the results have large-scale impacts, e.g., on land-use planning.

To evaluate the accuracy of different aircraft noise modelling approaches, we conducted validation simulations of three inherently different models, namely sonAIR [4, 5], FLULA2 [6] (the latter officially used for aircraft noise calculations in Switzerland) and AEDT [7] (a Doc 9911 [8] compliant and Doc 29 [9] equivalent model) and compared the results among each other and with measurements. These results are published in greater detail in [10]. In a second part, we conducted AEDT simulations with adapted input data adjusted to local conditions to demonstrate its influence on the simulation accuracy.

1 jonas.meister@empa.ch

a Shea mar ce 21-24 AUGUST SCOTTISH BENT caso

2. Model Validation

2.1. Validation Concept

We used noise measurements of several thousand aircraft fly-overs at different locations around Zurich (ZRH) and Geneva (GVA) airports, Switzerland, and simulated these noise events with the three models. For approaches, the measurement points are located up to 53 km from touchdown. The acoustic quantity used for the model validations and comparisons is the A-weighted sound exposure level L AE (sometimes also referred to as SEL). The simulated aircraft types are the most frequently operating large aircraft in ZRH and GVA, which account for the largest share of the total noise.

For one part of the simulated flights, flight data recorder (FDR) data was available, which, beside time and position information, include the aircraft configuration and N1 (rotational speed of the low- pressure compressor) data, which sonAIR is able to account for in simulations. For the other part of the flights, only time and position information from radar data were available. Table 1 summarizes the most important dataset properties, which we used for the simulations.

Table 1: Properties of the different datasets used for the validation simulations.

Range to airport(s)

FDR data # Flights Airport(s) # Different aircraft types

close yes 1732 ZRH/GVA 8

close no 6659 ZRH/GVA 24

far (up to 53 km) yes 394 ZRH 5

For this validation, we ran AEDT without adjustments to local conditions (standard procedural profiles, default thrust setting, NPD tables and ISA) and had to apply corresponding decibel adjustments according to chapter 6.4 of the ECAC.CEAC Doc 29 (Volume 1) [9] for substituted aircraft types.

More details on the input data (e.g., weather, terrain, ground cover) for all three models can be found in [10].

2.2. Aircraft Noise Models

A detailed comparison of the three models can be found in [10]. The three models are briefly described in the following.

sonAIR [4, 5] (version 7) is a time-step aircraft noise simulation model that yields spectral sound pressure levels in third-octave bands, designed to represent single flights in great detail. Engine and airframe noise are separately modelled as three-dimensional directivity patterns. The sound propagation is calculated separately and accounts for various attenuation and meteorological effects [4].

If the underlying dataset for the simulations are FDR data, all necessary parameters such as N1, air density, Mach number and aircraft configuration are available. Otherwise, as it is the case for 6659 validation flights (see Table 1), radar data are used, which only yield time and position information. Mandatory inputs, such as N1 and air density, have to be estimated or taken from other sources in the latter case. However, since it is hardly possible to estimate the aircraft configuration, a reduced model is used for these cases, which does not consider aircraft configuration.

FLULA2 [11] (version 004) is also a time-step aircraft noise simulation model but, contrary to sonAIR, directly yields an A-weighted sound exposure level at the receiver location for the total aircraft noise. Sound emission and propagation are combined in a fully empirical model description, and the sound source is two-dimensional (rotationally symmetric with respect to the roll axis). FLULA accounts for three-dimensional trajectories and has separate terms for lateral attenuation and shielding by terrain. It is designed for yearly noise calculations and does not use any additional

information from FDR data and therefore, only radar data information is used for calculations. FLULA2 is inherently adjusted to conditions around ZRH and GVA. Note that all sound source data used within this study are derived from the sonAIR models.

AEDT [7] (version 3d) is a software system to calculate fuel consumption, gaseous emissions and noise related to aircraft operations and is Doc 9911 [8] compliant and Doc 29 [9] equivalent. The flight trajectories are divided into segments [12]. The A-weighted sound pressure level at a certain receiver location is described as function of the slant distance to the receiver and power setting through so-called noise-power-distance (NPD) tables, with an implicit propagation description as in FLULA2. The NPD tables are derived from noise certification measurements and are therefore fully empirical. Additional terms account for lateral attenuation and engine installation effects.

AEDT is designed for noise calculations of larger scenarios, where either real or default flight profiles (with default thrust setting) can be used. It is possible to use adjusted input data to more accurately represent the real flight conditions.

2.3. Aircraft noise measurements

Figure 1 shows the measurement locations (black dots) around ZRH (left and center map) and GVA (right), which also represent the locations for the simulations. The left and the right map show the measurement points close to the airports with all simulated departure (red) and approach (blue) flights. The center map shows flights on two approach routes at ZRH, where the furthest distance between measurement location and touchdown is 53 km.

ee

Figure 1: Measurement locations around ZRH (left, center) and GVA (right) depicted as black dots. Left: Close range to ZRH, with all simulated departures (red) and approaches (blue). Center: Far range to ZRH, with two approach routes (purple and blue) up to 53 km from touchdown. Right: Close range to GVA, with all simulated departures (red) and approaches (blue). Maps adapted from [11].

2.4. Results

Figure 2 shows box-and-whisker plots of the differences between simulations and measurements for all three models. For simplicity, outliers are not shown. The plots are subdivided into simulations with and without FDR data and also into close as well as far range to the airports.

Figure 2: Differences between simulations and measurement for all three models. Left: Results for simulations with FDR data available, in close range to ZRH and GVA. Center: Results for simulations without FDR data available, in close range to ZRH and GVA. Right: Results for simulations with FDR data available, in far range to ZRH. Plots adapted from [11]. Overall, all three models show very good agreement with the measurements. The largest median difference of 1.1 dB was found for a subset of flights (departures calculated with AEDT).

The sonAIR calculations show the lowest scatter, especially if FDR data are available. If no FDR data are available and therefore a reduced modelling approach is used (see Section 2.2.), sonAIR performs similarly well as FLULA2 and AEDT. It reveals its full potential in the far range of ZRH, where the aircraft speeds are generally higher and the airframe noise is dominant over the engine noise. This is where the advantages of a separate model for engine and airframe become apparent.

FLULA2 shows the lowest overall median deviation from the measurements of all three models. In the close range to the airports, there is more scatter than with sonAIR, but less than AEDT. The largest scatter appears in the far range to ZRH, since the FLULA2 source models are derived from data in final approach and thus speeds, which do not represent the far range to touchdown.

AEDT calculations yield the largest scatter in close range, which is due to the usage of standard procedural profiles and thrust settings. It generally underestimates departures and tends to overestimate approaches. Nevertheless, the results are surprisingly good, since we used mainly default input data and had to make decibel adjustments in some cases (see Section 2.1.). This leads to the question how AEDT performs, if the input data are adjusted to conditions that are more realistic. This is investigated in the next section.

3. Adaption of AEDT Simulations

For the initial validation of AEDT (see above), default settings and unadjusted input data are used for the simulations. To investigate the potential of AEDT and the additional gain in accuracy with adjusted input data, we performed exemplary simulations to improve the agreement between AEDT calculations and the corresponding measurements.

A Ugg (Sim-Meas .W/FOR data close range w/o FOR data dose range ‘w/ FOR data farrange a| | ql “| een iro | [one al] Rove al] Rue al] Rove

3.1. Methods

We performed simulations with adjusted input data exemplarily for two narrow-body aircraft types, for which the aircraft noise was generally underestimated by AEDT, both for departures and approaches (see box-and-whisker plots in Figure 4). The flight events are the same as in the initial validation and amount to a total number of 773. The acoustic quantity for the comparison is the L AE .

For the simulations, we used the real trajectories as input on the one hand and adjusted the NPD tables on the other hand. For the former, fixed point profiles were used, where the aircraft altitude can be specified as a function of the distance to the runway endpoint.

Further, we used adapted NPD tables for both aircraft types, replacing the default thrust setting (kN or % of max. thrust) with N1, which is also the main input variable for engine noise in sonAIR. There are two reasons to use N1 in the NPD tables: On the one hand, the N1 values are known from the simulations with sonAIR, obtained from the FDR data, and can be entered into the fixed point profiles. On the other hand, we can generate adjusted NPD tables, or in this case NN1D tables, based on sonAIR simulations that represent the locally adapted noise data of the two narrow-body aircraft types.

To generate the adjusted NN1D tables, we conducted "pseudo-certification flights" with sonAIR, as illustrated in Figure 3. To that aim, we used standard atmospheric conditions around ZRH. As it is prescribed in chapter 2.5 of the ECAC.CEAC Doc 29 (Volume 2) [13], the standard aircraft speed for certification flights to generate NPD tables is 160 knots, and the selected heights above ground level (h AGL ) are set the same as in the original NPD tables. The minimum sound incidence angle with respect to the ground is chosen to be 10° to avoid any unwanted ground effects.

Since the certification aircraft speed is only used to define the noise exposure time, it is not representative for the simulated approaches. For the pseudo-certification flights with sonAIR, this has to be considered, as the airspeed has a major influence of the calculated airframe noise. Therefore, the aircraft speed input value for these simulations is changed to representative values at the measurement locations of the validation flights. However, these speeds are set for the currently emitted sound only, with the aircraft speed held constant at 160 knots for the noise impact calculation at the receiver location.

Figure 3: Scheme of pseudo-certification flights with sonAIR to generate NN1D (NPD) tables. The minimum sound incidence angle with respect to the ground is chosen to be 10° to avoid any unwanted ground effects. To summarize, the adjustment of the AEDT input data to local conditions includes the following three steps:

1. Pseudo-certification flights with sonAIR to generate new NPD tables, or in this case, NN1D tables. 2. Using fixed point profiles instead of default procedural profiles to consider the real flight profiles. 3. Input the N1 values into the fixed point profiles for each flight.

hag. = 200, 400, 630, 1000, 2000, 4000, 6300, 10000, 16000, 25000 ft V, = 160 kt ground soft ground

With these adjustments, we expected improved results especially for departures, since the engine noise is dominant in these cases. For approaches, the airspeed is still not considered by NN1D tables, which should lead to a somewhat larger scatter. Furthermore, decibel adjustments, as they were applied previously, should no longer be necessary. The results of these simulations are presented in the following.

3.2. Results

As in the initial validation, we compare the measurements with the simulations at the corresponding measurement locations at ZRH and GVA. Below, we show scatter plots of the measured and calculated noise levels (left) and box-and-whisker plots of the noise level differences between measurements and simulations (right).

3.3. Comparison of the event levels (L AE ) Figure 4 represents the L AE of the initial calculations with AEDT, where standard procedural profiles and NPD tables were used. Figure 5 shows the results with the adjusted profiles and the NN1D tables generated with sonAIR.

8 {(v)ap] woneinung 7% 60 Procedure 80 90 100 Lge Measurement (dB(A) 70

Figure 4: Comparison of L AE between simulation with AEDT and measurement before adjustment of input data, exemplarily for two narrow-body aircraft types (with decibel adjustments).

Figure 5: Comparison of L AE between simulation with AEDT and measurement after adjustment of input data, exemplarily for two narrow-body aircraft types (no decibel adjustments needed).

Procedure 100 90 80 Lge Measurement (dB(A) 70 8 8 8 8 @ £2 8B 60 {(v)ap] woneinung 7%

Both the median value of differences and the scatter improved strongly due to the adjustment of the input data. There are hardly any outliers anymore. As expected, departures are represented somewhat better, because the engine noise is dominant in these cases and the non-consideration of airframe noise is less important. Systematic deviations, as initially seen at lower levels, have now vanished. 4. CONCLUSIONS

In this study, we validated three different aircraft noise models, namely sonAIR, FLULA2 and AEDT using large measurement datasets. For the latter, we initially simulated with unadjusted input data and, in a second step, adapted these data to better match the local conditions.

The validation reveals that sonAIR performs better than the other models if FDR data are available. In this case, the most detailed sonAIR emission models can be used, which are able account for the aircraft configuration. The separation of engine and airframe noise into two independent emission models is advantageous especially for approaches in regions further away from the airports, where the aircraft speed is usually high and hence the airframe noise the dominant sound source. If no FDR data are available, sonAIR is run with a reduced model approach that does not account for configuration and performs similarly well as the other two models.

AEDT and FLULA2 are both designed to calculate complex airport scenarios with many aircraft types and a large number of flights. Accordingly, the median results, while exhibiting a somewhat larger scatter than those of sonAIR, represent the measurements accurately in average. AEDT was found to agree surprisingly well with measurements despite the use of standard procedural profiles and default NPD tables.

To demonstrate the potential improvement of the modelling results obtainable with AEDT if using adjusted input data, we performed exemplary simulations using input data adjusted to local conditions. To that aim, we generated new NPD tables, respectively NN1D tables, by means of pseudo-certification flights with sonAIR and included the real flight trajectories by means of fixed point profiles. Furthermore, we accounted for thrust, respectively N1, using the values from the sonAIR simulations. With these simulations, the differences between calculations and measurements as well as the scatter decreased considerably. Systematic deviations and most outliers could thereby be removed. This shows on the one hand the strength of AEDT's modelling approach and on the other hand the importance of adjusted input data. 5. ACKNOWLEDGEMENTS

We thank the Swiss International Airlines, as well as Zurich and Geneva Airport for permission to use the data for this study. We further would like to acknowledge the funding by the Swiss Federal Office for the Environment FOEN (assignment No. 5211.01812.100.01).

6. REFERENCES

1. European Environment Agency (EEA): Environmental noise in Europe — 2020 . EEA Report No 22/2019. Luxembourg, 2020. 2. EUROCONTROL: Aviation Outlook 2050 - Main Report. 2022. 3. WHO: Environmental Noise Guidelines for the European Region. Copenhagen, Denmark, 2018. 4. J. M. Wunderli, C. Zellmann, M. Köpfli, M. Habermacher, O. Schwab, F. Schlatter, B. Schäffer. sonAIR - a GIS-Integrated Spectral Aircraft Noise Simulation Tool for Single Flight Prediction and Noise Mapping . Acta Acustica united with Acustica , 104 , 440-451 (2018). 5. C. Zellmann, B. Schäffer, J. M. Wunderli, U. Isermann, C. O. Paschereit. Aircraft Noise Emission Model Accounting for Aircraft Flight Parameters . Journal of Aircraft , 55 , 682-695 (2017). 6. W. Krebs, R. Bütikofer, S. Plüss, G. Thomann. Sound source data for aircraft noise simulation . Acta Acustica united with Acustica , 90 , 91-100 (2004). 7. U.S. Department of Transportation Volpe National Transportation Systems Center: Aviation Environmental Design Tool (AEDT) Technical Manual, Version 3d. 2021. 8. ICAO: Recommended Method for Computing Noise Contours Around Airports. Doc 9911. 1st ed. , International Civil Aviation Organization (ICAO), Montréal, Canada, 2008. 9. ECAC: ECAC.CEAC Doc 29: Report on Standard Method of Computing Noise Contours around Civil Airports, Volume 1: Applications Guide. 4th ed. , European Civil Aviation Conference (ECAC), Neuilly-sur-Seine, France, 2016. Available from: https://www.ecac- ceac.org/documents/ecac-documents-and-international-agreements (last viewed on 01/19/2022). 10. J. Meister, S. Schalcher, J.-M. Wunderli, D. Jäger, C. Zellmann, B. Schäffer. Comparison of the Aircraft Noise Calculation Programs sonAIR, FLULA2 and AEDT with Noise Measurements of Single Flights . Aerospace , 8 , 388 (2021). 11. S. Pietrzko, R. F. Hofmann. Prediction of A-weighted aircraft noise based on measured directivity patterns . Applied Acoustics , 23 , 29-44 (1988). 12. U. Isermann, B. Vogelsang. AzB and ECAC Doc.29-Two best-practice European aircraft noise prediction models . Noise Control Engineering Journal , 58 , 455-461 (2010). 13. ECAC: ECAC.CEAC Doc 29: Report on Standard Method of Computing Noise Contours around Civil Airports, Volume 2: Technical Guide. 4th ed. , European Civil Aviation Conference (ECAC), Neuilly-sur-Seine, France, 2016. Available from: https://www.ecac- ceac.org/documents/ecac-documents-and-international-agreements (last viewed on 01/19/2022).