# CNOSSOS-EU road traffic source emission

> One page of the phonometry documentation, as Markdown. Index: https://docs.jmrp.io/phonometry/llms.txt

Canonical: https://docs.jmrp.io/phonometry/environment/sources/cnossos-road-emission/
Language: en
Alternate: https://docs.jmrp.io/phonometry/es/environment/sources/cnossos-road-emission/index.md
Version: phonometry 4.0.0
Install: pip install "phonometry[full]"
Author: José Manuel Requena Plens (https://orcid.org/0000-0003-1250-6212)
Published: 2026-07-31
Updated: 2026-11-16T21:36:14Z
Standards: OJ L 5, 10.1.2018, p. 35; Directive 2002/49/EC
Verified: 12 conformance checks (Directive (EU) 2015/996; Directive (EU) 2021/1226; Directive 2002/49/EC) https://docs.jmrp.io/phonometry/reference/conformance/
License: MIT (https://github.com/jmrplens/phonometry/blob/main/LICENSE)
Summary: CNOSSOS-EU road traffic emission in Python (Directive 2002/49/EC Annex II): rolling and propulsion power per vehicle category, corrections and power per metre.

---

Every strategic noise map drawn in the European Union since 2021 starts from the
same two numbers: how loud a vehicle is in each octave band, and how many of them
go past per hour. **Annex II of Directive 2002/49/EC**, as replaced by the
CNOSSOS-EU methods, turns those into a **directional sound power per metre of
source line**, which the propagation stage then carries to the receiver.

This guide covers the road source, section 2.2 of Annex II with the coefficient
database of Appendix F. The railway source (2.3, Appendix G) and the CNOSSOS-EU
propagation model (2.5) are separate.

## How do I compute CNOSSOS-EU road traffic noise emission in Python?

Describe the flow of each vehicle category with `environment.RoadTraffic`, then
pass the list to
`environment.road_source_power(traffic, surface=..., temperature_c=...)`. It
sums the rolling and propulsion sound powers of Annex II for each category and
octave band and turns them into a directional sound power per metre of source
line: `result.total_line_power` holds the octave bands and
`result.a_weighted_line_power` their A-weighted total, which a propagation
model then carries to the receiver. For the urban arterial below, whose
spectrum is drawn further down, `result.a_weighted_line_power` rounds to
`84.2` dB(A) per metre.

```python
from phonometry import environment

# An urban arterial: vehicles per hour of each category, all at 50 km/h.
traffic = [
    environment.RoadTraffic(category=environment.RoadVehicleCategory.LIGHT, flow_rate=1200.0, speed_kmh=50.0),
    environment.RoadTraffic(category=environment.RoadVehicleCategory.MEDIUM_HEAVY, flow_rate=90.0, speed_kmh=50.0),
    environment.RoadTraffic(category=environment.RoadVehicleCategory.HEAVY, flow_rate=45.0, speed_kmh=50.0),
    environment.RoadTraffic(category=environment.RoadVehicleCategory.MOTORCYCLES, flow_rate=60.0, speed_kmh=50.0),
]
result = environment.road_source_power(
    traffic, surface=environment.RoadSurface.THIN_LAYER_A, temperature_c=12.0, road_slope_percent=3.0,
    junction_distance=60.0, junction_type=environment.JunctionType.CROSSING,
)
print(result.frequencies)                             # [  63.  125.  250.  500. 1000. 2000. 4000. 8000.] Hz
print(result.total_line_power.round(1))               # [89.1 82.7 81.3 80.2 80.4 76.2 71.7 66. ] dB re 1 pW per metre
print(round(float(result.a_weighted_line_power), 1))  # 84.2 dB(A) per metre
```

## 1. Which text is implemented

Annex II is a moving target and getting the layering wrong is the easiest way to
ship a wrong table. Three instruments are in play:

| Instrument | What it does to the road source |
| :--- | :--- |
| Commission Directive (EU) **2015/996** | Replaces the whole of Annex II. Supplies formulae 2.2.1 to 2.2.20 and Tables F-1 to F-4. |
| **Corrigendum** of OJ L 5, 10.1.2018 | Corrects 2.2.1: the sound powers are calculated "for each octave band from **63 Hz to 8 kHz**", not the 125 Hz to 4 kHz the original text printed. Appendix F always covered 63 Hz to 8 kHz, so the uncorrected text contradicted its own tables. |
| Commission Delegated Directive (EU) **2021/1226** | Replaces **Table F-1** and **Table F-4** in their entirety, merges the former 4a and 4b rows of Table F-4 into one, and prints the octave-band A-weighting to be used in 2.5.5. |

The library implements the consolidated result. Tables F-2 (studded tyres) and
F-3 (junctions) have never been amended. The 2021 amendment is not cosmetic: the
current road source is about **2.5 to 3.5 dB(A) louder** than the one published
in 2015, so any comparison with pre-2021 literature carries that offset.

## 2. The source and the source line

Each vehicle is one **point source 0.05 m above the road surface**, radiating
uniformly; the first reflection on the pavement is already inside the sound
power, which is why the method calls it a semi-free-field quantity. A traffic
flow is an incoherent **source line**, ideally one line per lane at the lane
centre.

For each vehicle category $m$ and octave band $i$, formula (2.2.1) turns a
single-vehicle sound power into a power per metre of line:

$$
L'_{W,\mathrm{eq,line},i,m} = L_{W,i,m} + 10\,\log_{10}\!\left(\frac{Q_m}{1000\,v_m}\right),
$$

with $Q_m$ in vehicles per hour and $v_m$ in km/h. The vehicle power itself is
the energy sum of a **rolling** and a **propulsion** term (2.2.2), except for the
powered two-wheelers of category 4, which have no rolling noise at all and take
the propulsion term alone (2.2.3):

$$
L_{W,i,m} = 10\,\log_{10}\!\left(10^{L_{W\mathrm{R},i,m}/10} + 10^{L_{W\mathrm{P},i,m}/10}\right).
$$

Table [2.2.a] defines five vehicle categories: **1** light motor vehicles, **2**
medium heavy vehicles, **3** heavy vehicles, **4** powered two-wheelers and
**5** the "open" category reserved for vehicle types not yet defined. Category 4
splits into the subclasses **4a** mopeds and **4b** motorcycles, which operate
in very different driving modes, so the coefficient database carries five rows:
1, 2, 3, 4a and 4b. The first four categories are mandatory and category 5 is
optional; it has no coefficients in Appendix F and is not modelled here.

![Octave-band source-line power of an urban arterial: grey bars for the total and marker lines for the light, medium heavy, heavy and motorcycle contributions, with the heavy vehicles above the light ones from 125 Hz to 500 Hz and the light ones taking the lead again at 1 kHz](https://docs.jmrp.io/phonometry/media/cnossos_road_emission.svg)

*Light vehicles carry the flow, but 45 heavy vehicles per hour still take over
the mid frequencies: the per-metre spectrum shows exactly where each category
governs.*

**Show the code for this figure**

```python
import matplotlib.pyplot as plt
from phonometry import environment

traffic = [
    environment.RoadTraffic(category=environment.RoadVehicleCategory.LIGHT, flow_rate=1200.0, speed_kmh=50.0),
    environment.RoadTraffic(category=environment.RoadVehicleCategory.MEDIUM_HEAVY, flow_rate=90.0, speed_kmh=50.0),
    environment.RoadTraffic(category=environment.RoadVehicleCategory.HEAVY, flow_rate=45.0, speed_kmh=50.0),
    environment.RoadTraffic(category=environment.RoadVehicleCategory.MOTORCYCLES, flow_rate=60.0, speed_kmh=50.0),
]
result = environment.road_source_power(
    traffic, surface=environment.RoadSurface.THIN_LAYER_A, temperature_c=12.0, road_slope_percent=3.0,
    junction_distance=60.0, junction_type=environment.JunctionType.CROSSING,
)
print(result.total_line_power.round(1))
# [89.1 82.7 81.3 80.2 80.4 76.2 71.7 66. ]  dB re 1 pW per metre of line
print(round(float(result.a_weighted_line_power), 1))     # 84.2 dB(A) per metre
result.plot()
plt.show()
```

**What a road source is worth.** This urban arterial (1 200 light, 90 medium
heavy, 45 heavy vehicles and 60 motorcycles per hour at 50 km/h) comes out at
**84.2 dB(A) per metre of line**; the free-flowing motorway link of section 6
(1 000 light and 120 heavy vehicles per hour at 90 and 80 km/h) reaches
**88.6 dB(A) per metre**. Those are the two magnitudes to check your own result
against. The arithmetic behind them is worth holding on to: doubling the flow
adds exactly 3 dB, while halving the speed adds 3 dB to the flow term
$10\log_{10}(Q/1000v)$ and subtracts from the rolling term, so the two partly
cancel and an urban link is only a few decibels below a motorway one. The
spectrum says which mechanism is in charge: the motorway peaks at 1 kHz because
rolling noise leads, the arterial peaks at 63 Hz and stays flatter, because at
50 km/h with heavy vehicles and a junction 60 m away the propulsion term has
taken over. The check a reader can run on either: `a_weighted_line_power` must
equal the A-weighted energy sum of the printed band values, using the octave-band
`CNOSSOS_A_WEIGHTING` of section 6.

![Plan and section of a two-lane urban arterial modelled with CNOSSOS-EU. In plan, a source line runs along the centre of each lane, one line is broken into 20 m segments with the middle segment labelled with its line power plus ten log of the segment length and its equivalent point source marked at the centre, a signal-controlled junction sits with the 60 m distance x dimensioned back along the road and a small inline graph of the taper max(1 minus x over 100, 0) beside it, and a dwelling facade stands 25 m from the near lane with the receiver marked on it. In section, the point source sits 0.05 m above the road surface, the receiver 4 m above ground at the facade, and the road gradient s is drawn with the uphill and downhill halves of a bidirectional flow labelled separately](https://docs.jmrp.io/phonometry/media/diagram_cnossos_road.svg)

### Where each input comes from

Every argument of `road_source_power` is a *defined* quantity in Annex II, and
getting one wrong is a silent error: the call succeeds and the map is off by
decibels. This is where each of them is read from.

| Argument | What it means | Where it is read from | The trap |
| :--- | :--- | :--- | :--- |
| `RoadTraffic.flow_rate` ($Q_m$) | Vehicles of category $m$ per hour | Yearly average per hour, per time period (day / evening / night), per vehicle class and **per source line**, from traffic counting or a traffic model (2.2.1) | A peak-hour count. $Q$ is an annual average for the period, which is what makes the result an $L_\mathrm{den}$ ingredient |
| `RoadTraffic.speed_kmh` ($v_m$) | Representative speed of category $m$ | In most cases the **lower** of the legal maximum for the road section and the legal maximum for the vehicle category; where local measurement data is unavailable, the legal maximum for the category (2.2.1) | Assuming a measured average is required. It is not: the standard's default *is* the legal limit, and heavy vehicles usually take a lower one than the section |
| `temperature_c` ($\tau$) | Air temperature | The **yearly average** air temperature of the period, from long-term meteorological data (2.2.10) | The survey-day value. At $K_1 = 0.08$ dB/°C, a 12 °C annual mean against a 20 °C afternoon is 0.6 dB on the light-vehicle rolling term |
| `RoadTraffic.studded_fraction`, `studded_months` ($Q_{\mathrm{stud,ratio}}$, $T_\mathrm{s}$) | Fleet share and season | The average ratio of light vehicles per hour fitted with studded tyres **during** the months they are fitted, and the number of those months; the method forms $p_\mathrm{s} = Q_{\mathrm{stud,ratio}}\,T_\mathrm{s}/12$ itself (2.2.7) | Passing the year-averaged share, which applies $T_\mathrm{s}/12$ twice |
| `junction_distance` ($x$) | Distance to the junction | Measured along the road from the point source to the nearest **intersection of this source line with another source line** (2.2.5) | Measuring to the stop line or the kerb. It is the line crossing, and only the last 100 m is affected |
| `junction_type` ($k$) | 1 crossing with lights, 2 roundabout | The junction control, not its shape | A priority junction is neither: the correction is defined for signals and roundabouts |
| `road_slope_percent` ($s$) | Signed longitudinal slope, % | The terrain model along the source line, saturating at ±12 % | Applying one sign to a two-way flow. The standard requires splitting the flow in half and correcting one half uphill and the other downhill |
| `surface` | Pavement type | The road register, matched to a Table F-4 row **and its declared speed range** | Using a row outside its validity range |
| geometry | Source line and receiver | One line per lane at the lane centre (one line per carriageway, or one for a two-way road, is also acceptable), the point source 0.05 m above the surface; the Directive's receiver points are 4 m above ground | Placing the source at the kerb, or at a vehicle-body height |

Two consequences worth stating outright. First, the temperature has to belong to
the *same* period as $Q$ and $v$: a night flow paired with an annual mean
temperature and a night speed is what makes the resulting line power an $L_\mathrm{den}$
ingredient rather than a snapshot of one evening. Second, the source line is a
modelling object, not a lane marking: moving it from the lane centre to the
kerb changes every propagation distance downstream of it.

## 3. Rolling noise (2.2.4 to 2.2.10)

Rolling noise is tyre-road noise. It grows logarithmically with speed from the
reference speed $v_\mathrm{ref} = 70\ \text{km/h}$:

$$
L_{W\mathrm{R},i,m} = A_{\mathrm{R},i,m} + B_{\mathrm{R},i,m}\,\log_{10}\!\left(\frac{v_m}{v_\mathrm{ref}}\right)
             + \Delta L_{W\mathrm{R},i,m},
$$

where the correction collects four independent effects (2.2.5): the road
surface, the studded tyres, the junction and the air temperature.

**Air temperature** (2.2.10) is a straight line, applied equally to all eight
bands: $\Delta L_{W,\mathrm{temp}} = K_m\,(20\ \text{°C} - \tau)$, with
$K_1 = 0.08\ \text{dB/°C}$ for light vehicles and
$K_2 = K_3 = 0.04\ \text{dB/°C}$ for the two heavy categories. Cold air makes tyres
harder and the road louder, so the correction is *positive* below 20 °C.

**Studded tyres** (2.2.6 to 2.2.9) apply to category 1 only. The per-tyre excess
$D_{\mathrm{stud},i}(v)$ saturates below 50 km/h and above 90 km/h, and the
fleet-level correction weights it by the share of the year that studded tyres
are on the road, $p_\mathrm{s} = Q_{\mathrm{stud,ratio}} \cdot T_\mathrm{s}/12$:

$$
\Delta L_{\mathrm{studded},i} = 10\,\log_{10}\!\left[(1 - p_\mathrm{s}) + p_\mathrm{s}\,10^{D_{\mathrm{stud},i}/10}\right].
$$

**Junctions** (2.2.17, 2.2.18) add the same linear taper to both terms:
$\Delta L_{W\mathrm{R,acc}} = C_{\mathrm{R},m,k}\cdot\max(1 - |x|/100,\ 0)$ for rolling
noise and $\Delta L_{W\mathrm{P,acc}} = C_{\mathrm{P},m,k}\cdot\max(1 - |x|/100,\ 0)$ for
propulsion noise, with $x$ the distance in metres from the point source to the
nearest intersection of its source line with another source line, and $k$ the
junction type: 1 for a crossing with traffic lights, 2 for a roundabout. Both are
attributed equally to all octave bands. The rolling coefficients $C_\mathrm{R}$ are
*negative* (traffic near a junction is slower), the propulsion coefficients $C_\mathrm{P}$
positive (it is accelerating), and the standard states explicitly that both are
zero from $|x| \ge 100$ m.

![Three panels giving the A-weighted change in source-line power that each correction of section 3 produces, all computed on a flow of 1000 vehicles per hour. Left: air temperature swept from minus 10 to plus 40 degrees Celsius, a straight line worth about 2.1 dB at minus 10 for light vehicles and about 0.5 dB for heavy ones, crossing zero at the 20 degree reference. Middle: the studded-tyre correction against speed for two fleet shares, flat below 50 km/h, peaking near 50 km/h and flattening again above 90 km/h. Right: the junction correction against distance to the junction from 0 to 120 m, negative for both a signal-controlled crossing and a roundabout, deepest for the roundabout at the junction itself, and identically zero from 100 m out](https://docs.jmrp.io/phonometry/media/cnossos_road_corrections.svg)

*The three corrections of this section, each drawn as what it actually does to
the line power. The temperature term is the only one that is exactly linear and
the only one that changes sign; the studded term is a saturating band between 50
and 90 km/h; and the junction term is not monotone, because the negative rolling
coefficient and the positive propulsion coefficient taper together. For a
crossing the deepest reduction is around 50 m, not at the junction.*

**Show the code for this figure**

```python
import matplotlib.pyplot as plt
import numpy as np

# `road_source_power`, `RoadTraffic`, `RoadVehicleCategory` and `JunctionType`
# are imported by the figure snippet above.
def a_line(**kwargs):
    return float(environment.road_source_power(**kwargs).a_weighted_line_power)

light = [environment.RoadTraffic(category=environment.RoadVehicleCategory.LIGHT, flow_rate=1000.0, speed_kmh=50.0)]
heavy = [environment.RoadTraffic(category=environment.RoadVehicleCategory.HEAVY, flow_rate=1000.0, speed_kmh=50.0)]
temps = np.linspace(-10.0, 40.0, 60)
fig, (t_ax, s_ax, j_ax) = plt.subplots(1, 3, figsize=(13, 4.2))
for flow, name in [(light, "Light (1)"), (heavy, "Heavy (3)")]:
    ref = a_line(traffic=flow)
    t_ax.plot(temps, [a_line(traffic=flow, temperature_c=t) - ref for t in temps],
              label=name)
t_ax.set(xlabel="Air temperature tau [°C]", ylabel="Change in line power [dB(A)]")
t_ax.legend()

speeds = np.linspace(20.0, 130.0, 60)
for share in (0.2, 0.5):
    s_ax.plot(speeds, [
        a_line(traffic=[environment.RoadTraffic(category=environment.RoadVehicleCategory.LIGHT, flow_rate=1000.0, speed_kmh=v,
                                                studded_fraction=share)],
               studded_months=4.0)
        - a_line(traffic=[environment.RoadTraffic(category=environment.RoadVehicleCategory.LIGHT, flow_rate=1000.0, speed_kmh=v)])
        for v in speeds], label=f"Qstud,ratio = {share:g}, Ts = 4 months")
s_ax.set(xlabel="Speed v [km/h]", ylabel="Change in line power [dB(A)]")
s_ax.legend(fontsize="small")

xs = np.linspace(0.0, 120.0, 60)
ref = a_line(traffic=light)
for junction, name in [(environment.JunctionType.CROSSING, "Crossing with lights"),
                       (environment.JunctionType.ROUNDABOUT, "Roundabout")]:
    j_ax.plot(xs, [a_line(traffic=light, junction_distance=float(x),
                          junction_type=junction) - ref for x in xs], label=name)
j_ax.set(xlabel="Distance to the junction |x| [m]",
         ylabel="Change in line power [dB(A)]")
j_ax.legend(fontsize="small")
plt.show()
```

## 4. Propulsion noise (2.2.11 to 2.2.16)

Propulsion noise is the power train: engine, exhaust, transmission, intake. It
is **linear** in speed, not logarithmic, because at low speed the engine is
working hardest relative to the distance covered:

$$
L_{W\mathrm{P},i,m} = A_{\mathrm{P},i,m} + B_{\mathrm{P},i,m}\,\frac{v_m - v_\mathrm{ref}}{v_\mathrm{ref}}
             + \Delta L_{W\mathrm{P},i,m}.
$$

Where the two speed laws cross is what makes the road source behave the way
it does: below the crossover the source is an engine, above it a tyre.

![A-weighted single-vehicle sound power against speed from 20 to 130 km/h for light and heavy vehicles, with the rolling and propulsion components dashed and dotted and the crossover speed marked on each pair](https://docs.jmrp.io/phonometry/media/cnossos_road_speed_law.svg)

*A light vehicle is tyre-dominated from about 25 km/h; a heavy vehicle only
from about 60 km/h, which is why a lorry is still an engine at urban speeds.*

**Show the code for this figure**

```python
import matplotlib.pyplot as plt
import numpy as np
from phonometry import environment

weights = np.asarray(environment.CNOSSOS_A_WEIGHTING)
a_weighted = lambda bands: 10.0 * np.log10(np.sum(10.0 ** ((bands + weights) / 10.0)))

speeds = np.linspace(20.0, 130.0, 221)
fig, ax = plt.subplots()
for category, name in [("1", "Light vehicles (1)"), ("3", "Heavy vehicles (3)")]:
    rolling = [a_weighted(environment.road_rolling_noise(category, v)) for v in speeds]
    propulsion = [a_weighted(environment.road_propulsion_noise(category, v)) for v in speeds]
    total = [a_weighted(environment.road_vehicle_sound_power(category, v)) for v in speeds]
    ax.plot(speeds, total, lw=2.4, label=f"{name} - total")
    ax.plot(speeds, rolling, ls="--", lw=1.2, label=f"{name} - rolling")
    ax.plot(speeds, propulsion, ls=":", lw=1.2, label=f"{name} - propulsion")
ax.set_xlabel("Speed v [km/h]")
ax.set_ylabel("A-weighted sound power [dB(A) re 1 pW]")
ax.legend(fontsize="small")
plt.show()
```

**Road gradient** (2.2.13 to 2.2.16) is the one correction whose published form
is genuinely asymmetric, and the asymmetry is not a transcription slip. For
light vehicles the downhill branch has **no speed factor**, while the two heavy
categories do and use different speed offsets:

| Category | Downhill ($s$ below the breakpoint) | Flat band | Uphill |
| :--- | :--- | :--- | :--- |
| 1 | $(\min(12\ \%, -s) - 6\ \%)/1\ \%$ | −6 % to 2 % | $(\min(12\ \%, s) - 2\ \%)/1.5\ \% \cdot v/100$ |
| 2 | $(\min(12\ \%, -s) - 4\ \%)/0.7\ \% \cdot (v - 20)/100$ | −4 % to 0 % | $\min(12\ \%, s)/1\ \% \cdot v/100$ |
| 3 | $(\min(12\ \%, -s) - 4\ \%)/0.5\ \% \cdot (v - 10)/100$ | −4 % to 0 % | $\min(12\ \%, s)/0.8\ \% \cdot v/100$ |

Category 4 takes no gradient correction. The slope saturates at 12 % in both
directions. Downhill noise comes from engine braking, uphill noise from load, so
both branches are positive. For a bidirectional flow, split the flow in two and
correct one half uphill and the other downhill.

![The A-weighted road-gradient correction to propulsion noise against the signed slope from minus 14 to plus 14 percent, at 50 and 80 km/h, for vehicle categories 1, 2 and 3. Every curve is flat inside its own dead band, rises steeply on the downhill side and less steeply uphill, and stops changing beyond plus or minus 12 percent where the slope saturates; the light-vehicle downhill branch is the only one that does not move with speed](https://docs.jmrp.io/phonometry/media/cnossos_road_gradient.svg)

*The asymmetry is a shape, not a typo. Each category has its own flat band
(−6 % to +2 % for light vehicles, −4 % to 0 % for the two heavy classes),
each saturates at ±12 %, and both branches rise: downhill the engine brakes,
uphill it works. At 80 km/h a 14 % descent is worth 6.0 dB(A) to a light
vehicle and 11.2 dB(A) to a heavy one, while the same climb is worth 5.3 and
12.0 dB(A), and the light-vehicle downhill branch is the only one of the six
that does not scale with speed.*

**Show the code for this figure**

```python
import matplotlib.pyplot as plt
import numpy as np

# `road_propulsion_noise` and `CNOSSOS_A_WEIGHTING` are imported by the
# speed-law figure snippet above.
weights = np.asarray(environment.CNOSSOS_A_WEIGHTING)
a_weighted = lambda bands: 10.0 * np.log10(np.sum(10.0 ** ((bands + weights) / 10.0)))

slopes = np.linspace(-14.0, 14.0, 141)
fig, ax = plt.subplots()
for category in ("1", "2", "3"):
    for speed, style in [(50.0, "--"), (80.0, "-")]:
        flat = a_weighted(environment.road_propulsion_noise(category, speed))
        ax.plot(slopes,
                [a_weighted(environment.road_propulsion_noise(category, speed, road_slope_percent=s)) - flat
                 for s in slopes], style, label=f"Category {category}, {speed:g} km/h")
ax.set_xlabel("Road slope s [%]")
ax.set_ylabel("Propulsion-noise correction [dB(A)]")
ax.legend(fontsize="small")
plt.show()
```

**Below 20 km/h the vehicle stops getting quieter.** Clause 2.2.1 states, after
both (2.2.2) and (2.2.3), that for speeds under 20 km/h the sound power takes its
20 km/h value: the tabulated regressions were fitted to free-flowing traffic and
are not extrapolated into a queue. The flow term keeps the *true* speed, so the
two halves of (2.2.1) pull in opposite directions and the line power has a
minimum. For 1 000 light vehicles per hour on the reference surface,
`road_source_power` returns 81.5 dB(A) per metre at 50 km/h, **76.2 at 20 km/h**,
79.2 at 10 km/h and 82.2 at 5 km/h; the same road gets *louder* per metre as the
stream slows below the floor, purely because more vehicles are on each metre of
it. A reader who does not know the floor cannot tell that from a modelled
congestion effect.

Say what the model therefore does not represent: acceleration and braking cycles,
idling in a queue, and the propulsion noise of stop-start driving, for which the
junction correction over the last 100 m is a deliberate crude proxy. Congested
links are normally modelled at their average travel speed with that correction
applied.

## 5. The road surface (2.2.19, 2.2.20 and Table F-4)

Table F-4 gives, for the reference surface and each of fourteen named
pavements, an octave-band coefficient $\alpha$ and a speed coefficient $\beta$,
per vehicle category, with the speed range over which the row is declared
valid. The surface enters the two terms
differently:

$$
\Delta L_{W\mathrm{R,road},i,m} = \alpha_{i,m} + \beta_m\,\log_{10}\!\left(\frac{v_m}{v_\mathrm{ref}}\right),
\qquad
\Delta L_{W\mathrm{P,road},i,m} = \min\{\alpha_{i,m};\,0\}.
$$

An absorbing surface reduces propulsion noise; a noisy one does not increase it.
That asymmetry is deliberate: the pavement can absorb the engine sound radiated
down onto it, but a rough texture only generates tyre noise, and tyre noise is
already the rolling term.

```python
from phonometry import environment

row = environment.road_surface_coefficients(environment.RoadSurface.TWO_LAYER_ZOAB_FINE)
row.speed_range_kmh          # (80.0, 130.0) km/h, the printed validity range
row.alpha["1"]           # alpha per octave band for light vehicles
row.beta["1"]            # -0.1, the speed coefficient
```

![Octave-band surface coefficient alpha for light vehicles against frequency for five Table F-4 rows spanning the range, plus the all-zero reference surface drawn as a flat line at 0 dB. The two-layer porous asphalt rows sit several decibels below the reference across the mid and high bands, the thin layers slightly below, and the hard-element paving well above it, with each curve's declared speed range printed in the legend](https://docs.jmrp.io/phonometry/media/cnossos_road_surfaces.svg)

*The reference surface is the zero line, so every curve reads directly as
"louder or quieter than the CNOSSOS reference": the two-layer porous asphalt is
6.3 dB below it at 1 kHz, hard elements not laid in herringbone are 31.4 dB above
it at 63 Hz. The declared speed range in the legend is part of the row: a porous
asphalt validated from 80 to 130 km/h is not a coefficient set for a 30 km/h
street, the reference row alone carries no range, and the method offers no
interpolation between rows.*

**Show the code for this figure**

```python
import matplotlib.pyplot as plt
import numpy as np

# `RoadSurface` and `road_surface_coefficients` are imported just above.
bands = np.array([63.0, 125.0, 250.0, 500.0, 1000.0, 2000.0, 4000.0, 8000.0])
fig, ax = plt.subplots()
for surface in (environment.RoadSurface.REFERENCE, environment.RoadSurface.TWO_LAYER_ZOAB_FINE,
                environment.RoadSurface.ONE_LAYER_ZOAB, environment.RoadSurface.THIN_LAYER_A,
                environment.RoadSurface.SMA_NL8, environment.RoadSurface.HARD_ELEMENTS_NOT_HERRINGBONE):
    row = environment.road_surface_coefficients(surface)
    # The reference row carries no speed range: it is valid by definition.
    span = ("all speeds" if row.speed_range_kmh is None
            else "{:g}-{:g} km/h".format(*row.speed_range_kmh))
    ax.semilogx(bands, row.alpha["1"], label=f"{surface.value} ({span})")
ax.axhline(0.0, color="k", lw=0.8)
ax.set_xlabel("Octave-band centre frequency [Hz]")
ax.set_ylabel("Surface coefficient alpha, category 1 [dB]")
ax.legend(fontsize="small")
plt.show()
```

The reference surface of the first row is all zeros: it is the virtual average
of a dense asphalt concrete 0/11 and a stone mastic asphalt 0/11, between two
and seven years old, in representative maintenance condition, dry, with no
studded tyres, at 20 °C, on the flat. Under exactly those conditions every
correction in 2.2 vanishes identically, so at the reference speed
$v_m = v_\mathrm{ref} = 70\ \text{km/h}$, where the two speed terms vanish as
well, the sound powers **are** the Table F-1 coefficients $A_{\mathrm{R},i,m}$ and
$A_{\mathrm{P},i,m}$.

## 6. Handing the source to a propagation model

The emission stage produces a power per metre. Splitting that line into
equivalent point sources is, in the words of section 2.5.3, "outside the scope
of the current methodology": a point source standing for a segment of length
$dL$ simply carries $L'_{W,\mathrm{eq,line},i} + 10\log_{10}(dL)$, which is
arithmetic and is offered as such.

```python
import numpy as np
from phonometry import environment

result = environment.road_source_power([
    environment.RoadTraffic(category=environment.RoadVehicleCategory.LIGHT, flow_rate=1000.0, speed_kmh=90.0),
    environment.RoadTraffic(category=environment.RoadVehicleCategory.HEAVY, flow_rate=120.0, speed_kmh=80.0),
])
segment = environment.line_source_segment_power(result.total_line_power, 20.0)  # a 20 m segment

# One segment, the one directly opposite a receiver 100 m from the road.
levels = environment.predicted_receiver_level(
    segment,
    environment.PropagationGeometry(distance=100.0, source_height_m=result.source_height_m, receiver_height_m=4.0),
    frequencies=result.frequencies,
)

# The road is the energy sum of its segments: 1 km either way, in 20 m steps.
energy = np.zeros_like(levels)
for x in np.arange(-990.0, 1000.0, 20.0):        # segment centres along the road
    energy += 10 ** (environment.predicted_receiver_level(
        segment,
        environment.PropagationGeometry(distance=float(np.hypot(100.0, x)), source_height_m=result.source_height_m, receiver_height_m=4.0),
        frequencies=result.frequencies,
    ) / 10)
road = 10 * np.log10(energy)
```

Two rules make a segmentation defensible, and neither is method: 2.5.3 declares
the split out of scope, so this is engineering practice. Each segment must be
short enough that the point source at its centre stands for the whole segment.
ISO 9613-2 allows one equivalent point source only where the propagation
distance exceeds twice the largest source dimension, so a segment seen from
100 m may be tens of metres long while one seen from 10 m may not. And the road
must be modelled far enough in both directions that the tails stop mattering:
beyond roughly ten times the perpendicular offset, adding segments changes the
total by less than a tenth of a decibel, which is why 1 km either way is enough
at 100 m here. The two spectra are read out in dB(A) two paragraphs down:
**52.9 dB(A)** for the single segment against **64.7 dB(A)** for the road, an
11.8 dB difference that is entirely the arithmetic above.

Note what this is and is not. CNOSSOS-EU has **its own** propagation method in
section 2.5 of Annex II, and it is not ISO 9613-2: the two differ in the ground
model, in the diffraction treatment and in the way favourable and homogeneous
conditions are combined. Chaining a CNOSSOS emission onto the ISO 9613-2
propagation of this library is a legitimate engineering estimate, but it is not
the normative CNOSSOS chain and must not be reported as one. If that
normative chain over a whole map is what you need,
[NoiseModelling](https://github.com/Universite-Gustave-Eiffel/NoiseModelling)
is an open tool built for it: a free Java library, not a Python package,
that computes environmental noise maps with the CNOSSOS-EU road and rail
emission and propagation methods on H2GIS and PostGIS spatial databases.

For the A-weighted total, use the octave-band weighting the Directive itself
prints in 2.5.5 as amended, rather than recomputing it:

```python
from phonometry import environment

environment.CNOSSOS_A_WEIGHTING   # (-26.2, -16.1, -8.6, -3.2, 0.0, 1.2, 1.0, -1.1)
result.a_weighted_line_power

# The two receiver spectra of the segmentation above, A-weighted:
weight = np.asarray(environment.CNOSSOS_A_WEIGHTING)
print(round(float(10 * np.log10(np.sum(10 ** ((levels + weight) / 10)))), 1))
# 52.9 dB(A) - one 20 m segment, not the road
print(round(float(10 * np.log10(np.sum(10 ** ((road + weight) / 10)))), 1))
# 64.7 dB(A) - the same road, summed over 2 km: 11.8 dB more
```

## 7. Substituting a national database

Appendix F is called a *database*, not a table of constants, because Member
States may substitute measured national values, and because the same equations
have already been evaluated with two different coefficient sets (the 2015 one
and the 2021 one). Both coefficient objects can be replaced:

```python
import dataclasses

from phonometry import environment

# One measured national row for light vehicles, the rest of Appendix F as published.
national = dataclasses.replace(
    environment.ROAD_COEFFICIENTS,
    rolling_a={
        **environment.ROAD_COEFFICIENTS.rolling_a,
        "1": (83.4, 89.0, 88.1, 93.6, 100.4, 96.9, 87.0, 76.5),
    },
)
```

A substituted database has to cover every category of the traffic it is used
with: reaching for a category it does not carry raises `ValueError` rather than
failing later.

That mechanism is also what pins the implementation: the European Commission
published a test set of 4 875 road emission cases computed with the 2015
coefficients, and feeding the shipped equations that superseded database (not
the 2021 tables the library ships by default) reproduces every published band
level to 0.005 dB, inside the two decimals the test set prints.

## What this guide covers

- **Covered.** Section 2.2 of Annex II to Directive 2002/49/EC in the consolidated text: the
  traffic-flow line power (2.2.1), the vehicle power (2.2.2, 2.2.3), rolling
  noise with its studded-tyre and temperature corrections (2.2.4 to 2.2.10),
  propulsion noise with the road-gradient correction (2.2.11 to 2.2.16), the
  junction correction (2.2.17, 2.2.18) and the road-surface effect (2.2.19,
  2.2.20), together with the whole Appendix F database, through
  `road_source_power`, `road_vehicle_sound_power`, `road_rolling_noise`,
  `road_propulsion_noise` and `road_surface_coefficients`.

- **Not covered.** The railway source (2.3 and Appendix G), the industrial source (2.4 and
  Appendix H), the aircraft source (2.6 and 2.7) and the CNOSSOS-EU propagation
  method of section 2.5, which is a different model from the ISO 9613-2 one
  implemented in this library. The open vehicle category 5 has no coefficients
  in Appendix F and is therefore not modelled. How a source line is split into
  point sources is declared out of scope by the method itself.

## See also

- API reference: [`environment.sources.cnossos_road`](https://docs.jmrp.io/phonometry/reference/api/environment/cnossos-road/).
- [Outdoor Sound Propagation](https://docs.jmrp.io/phonometry/environment/propagation/outdoor-propagation/): the
  ISO 9613-2 chain that carries a source power to a receiver.
- [Environmental noise levels](https://docs.jmrp.io/phonometry/environment/assessment/environmental-levels/): the
  $L_\mathrm{den}$ and $L_\mathrm{night}$ indicators the resulting maps are drawn for.

## References

- **European Union. (2002).** [*Directive 2002/49/EC of the European Parliament and of the Council of 25 June 2002 relating to the assessment and management of environmental noise*](https://eur-lex.europa.eu/eli/dir/2002/49/oj) (Directive 2002/49/EC).
  The directive whose Annex II the two Commission directives above write: section 2.2 and Appendix F, which this guide implements, are its annex as they amend it, read in the consolidated text.
- **European Union. (2015).** [*Commission Directive (EU) 2015/996 establishing common noise assessment methods according to Directive 2002/49/EC of the European Parliament and of the Council*](https://eur-lex.europa.eu/eli/dir/2015/996/oj) (Directive (EU) 2015/996).
  Replaces Annex II of Directive 2002/49/EC with the CNOSSOS-EU methods. The implemented road source is section 2.2 (formulae 2.2.1 to 2.2.20) with the coefficient database of Appendix F.
- **European Union. (2018).** [*Corrigendum to Commission Directive (EU) 2015/996*](<https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32015L0996R(01)>) (OJ L 5, 10.1.2018, p. 35).
  Corrects the octave-band range of the road source to 63 Hz - 8 kHz, replaces the whole of Appendix G and fixes the railway roughness conversion to metres per second.
- **European Union. (2021).** [*Commission Delegated Directive (EU) 2021/1226 amending, for the purposes of adapting to scientific and technical progress, Annex II to Directive 2002/49/EC as regards common assessment methods for noise*](https://eur-lex.europa.eu/eli/dir_del/2021/1226/oj) (Directive (EU) 2021/1226).
  The amendment in force since 31 December 2021. Replaces Tables F-1 and F-4 in their entirety and prints the octave-band A-weighting used to sum the source spectrum.
- **International Organization for Standardization. (1996).** [*Acoustics — Attenuation of sound during propagation outdoors — Part 2: General method of calculation*](https://www.iso.org/standard/20649.html) (ISO 9613-2:1996).
  The propagation model used here to carry the emission to a receiver. CNOSSOS-EU has its own propagation method in section 2.5 of Annex II, which is not ISO 9613-2.
- **Kephalopoulos, S., Paviotti, M., & Anfosso-Lédée, F. (2012).** [*Common noise assessment methods in Europe (CNOSSOS-EU)*](https://publications.jrc.ec.europa.eu/repository/handle/JRC72550) (EUR 25379 EN). Publications Office of the European Union. https://doi.org/10.2788/31776
  The Joint Research Centre reference report the Directive text descends from, with the measurement campaigns behind the Appendix F coefficients.
