3.3.2. Impact Calculation

The computation of the ANE impact consists of determining the contribution of a particular event to the equivalent noise level of the recorded audio after applying the A-weighting filter [53], i.e., *LAeq*. This metric has been defined to evaluate the effect of each ANE on the noise map *LAeq*. It is calculated as the difference between the *LAeq* computed with all the raw data and the *LAeq* after removing the ANE by means of a linear interpolation (see Equation (3)). The event should be replaced by a lower-period linear interpolation to maintain the weight of the surrounding RTN level. That way, the road-traffic noise measurement is as accurate as possible in the whole integration time.

$$
\Delta L\_{Aeq} = L\_{Aeq, ANE} - L\_{Aeq, \overline{ANE}} \tag{3}
$$

where Δ*LAeq* is the contribution of this ANE to the *LAeq*, *LAeq*,*ANE* is the A-weighted equivalent noise level of the raw audio given an evaluation period and *LAeq*,*ANE* is the equivalent A-weighted equivalent noise level of the same audio after removing the ANE.

In the DYNAMAP project, noise-map values are updated every 5 min, thus the contribution of ANEs on the *LAeq* will be evaluated in this integration time, i.e., Δ*LA*<sup>300</sup>*s*. The low-level interpolations are carried out in a 1-s integration window, i.e., *LA*<sup>1</sup>*s*, as it heuristically proved to be a good trade-off between representing all audible short events and not adding imperceptible changes to the equivalent sound measurement. As the goal is to obtain the equivalent noise level completely unaffected by the ANE, a span of 500 ms is left before and after the exact ANE label. More details on the casuistry of this calculation may be found in [52].

Finally, it is worth mentioning that the *LA*<sup>300</sup>*<sup>s</sup>* measurement is conducted in a 5-min sliding window where the ANE is centered. This is a better approximation than using a 5-min fixed window as the mean distance between RTN samples and ANE samples is reduced. The fact that future samples are needed (plus the fact that the exact labeling can only be provided after listening and labeling the recordings) implies that this measurement can only be applied in an off-line analysis and not in the real-time operation mode of the WASN.
