Averaging Noise Levels
chris.hoops
If you look at the standard acoustical parameters:
L1, L10, L90, Leq, Lmax etc
Clearly there is no basis for averaging the Lmax, whereas the Leq is a
continuous average level.
If you have a series of values in the other parameters e.g. a series of L10
values, could these be averaged effectively using a simple log average of
the results or is this the wrong way? How would you do it for the different
percentile parameters?
Ken Plotkin
There are two philosophies for declaring the average of a series of
Lx's. One is to just do a numerical average. If you delve deeper
with that approach, you can look at the distribution of Lx's over the
series. The other is to figure the distribution of levels within each
period, then combine the distributions to one pooled distributiion.
You'd then extract the Lx's for the pooled distribution.
This was a heavily debated topic a few years ago for analysis of
noise/soundscapes in US national parks. After considerable
discussion, everyone realized that the pooled method tended to crunch
the final L10 and L90 toward extremes, and told you nothing about how
the soundscape varied from day to day. The conclusion was that the
best way to represent the statistics of sequences Lx's (and also Leq)
was to do simple numeric averages.
That's also consistent with the process of fitting distributions to
levels, rather than distributions to the antilogs of levels.
Ken Plotkin
Brian Marston
attempting to describe. Over say, a day of 15 min samples, you would
probably be looking for the "lowest" L90 values. Then, do not even
think of averaging, think about what you are looking for!
In our region, the day is defined as 7:00am to 6:00pm (44 sample
periods), the evening is from 6:00pm to 10:00pm (16 sample periods) and
the night is from 10:00pm to 7:00am (36 sample periods). By general
consensus each day starts at dawn :)
For each division - day / evening / night - the LA90,15min that is
exceeded for 90% of the time is taken as the "design" environmental
LA90,15min noise level. To describe the environment, the same can be
done for the LA10,15min if required. This approach trims off anomalous
very low LA90,15min readings.
For large projects, the monitoring could go on for say 3 weeks, giving
for each day of the week (assuming consistent weekly noise patterns),
132 daytime readings, 48 evening readings and 108 night time readings.
This was developed on a project with about 12 monitoring locations each
monitored for 3 to 6 weeks each. We were in the middle processing the
data from 3 other consultants with our local regulatory body looking
over our shoulders (and scribbling notes). The data was trimmed of rain
affected readings and wind affected readings. This process produced day
/ evening / night LA90,15min values for each week day and each Sunday /
Saturday. The weekday LA90s were then arithmetically averaged and the
week end LA90s were arithmetically averaged to get a final results.
The end result was a set of LA90s that all consultant involved in agreed
were true LA90s for that project area and no one could argue about. The
local regulatory body then made it standard practice.
Noral Stewart
"Brian Marston" <mars...@pacific.net.au> wrote in message
news:1nHZg.18232$b6.1...@nasal.pacific.net.au...
> Ken Plotkin wrote:
>> On Thu, 19 Oct 2006 09:00:46 +1000, "chris.hoops"
>> <ch...@palmeracoustics.com> wrote:
>>
>>
>>>Hi all,
>>>
>>>If you look at the standard acoustical parameters:
>>>
>>>L1, L10, L90, Leq, Lmax etc
>>>
>>>Clearly there is no basis for averaging the Lmax, whereas the Leq is a
>>>continuous average level.
>>>
>>>If you have a series of values in the other parameters e.g. a series of
>>>L10 values, could these be averaged effectively using a simple log
>>>average of the results or is this the wrong way? How would you do it for
>>>the different percentile parameters?
>>
>>
>> There are two philosophies for declaring the average of a series of
>> Lx's. One is to just do a numerical average. If you delve deeper
>> with that approach, you can look at the distribution of Lx's over the
>> series. The other is to figure the distribution of levels within each
>> period, then combine the distributions to one pooled distributiion.
>> You'd then extract the Lx's for the pooled distribution.
>>
>> This was a heavily debated topic a few years ago for analysis of
>> noise/soundscapes in US national parks. After considerable
>> discussion, everyone realized that the pooled method tended to crunch
>> the final L10 and L90 toward extremes, and told you nothing about how
>> the soundscape varied from day to day.
It would be typical that the longer the period, the more extreme the L10 and
L90 would be, and it is correct that this does cover up variations over
shorter periods.
The conclusion was that the
>> best way to represent the statistics of sequences Lx's (and also Leq)
>> was to do simple numeric averages.
Very surprised about using arithmetic average on Leq as this tends to
produce a lower result than the energy average and goes against the
philosopy that long-term Leq or DNL is a good indicator.
>>
>> That's also consistent with the process of fitting distributions to
>> levels, rather than distributions to the antilogs of levels.
>>
>> Ken Plotkin
>>
> Over a short period arithmetic averaging is OK but what are you attempting
> to describe. Over say, a day of 15 min samples, you would probably be
> looking for the "lowest" L90 values. Then, do not even think of
> averaging, think about what you are looking for!
>
> In our region, the day is defined as 7:00am to 6:00pm (44 sample periods),
> the evening is from 6:00pm to 10:00pm (16 sample periods) and the night is
> from 10:00pm to 7:00am (36 sample periods). By general consensus each day
> starts at dawn :)
>
> For each division - day / evening / night - the LA90,15min that is
> exceeded for 90% of the time is taken as the "design" environmental
> LA90,15min noise level. To describe the environment, the same can be done
> for the LA10,15min if required. This approach trims off anomalous very low
> LA90,15min readings.
>
Brian, I do not understand what you are saying in the paragraph above.
Brian Marston
> "Brian Marston" <mars...@pacific.net.au> wrote in message
> <snip>
>>In our region, the day is defined as 7:00am to 6:00pm (44 sample periods),
>>the evening is from 6:00pm to 10:00pm (16 sample periods) and the night is
>>from 10:00pm to 7:00am (36 sample periods). By general consensus each day
>>starts at dawn :)
>>
>>For each division - day / evening / night - the LA90,15min that is
>>exceeded for 90% of the time is taken as the "design" environmental
>>LA90,15min noise level. To describe the environment, the same can be done
>>for the LA10,15min if required. This approach trims off anomalous very low
>>LA90,15min readings.
>
>
> Brian, I do not understand what you are saying in the paragraph above.
Explanation:
Across the night time period, there would be 9 hours, giving 36 readings
of LA90,15min. Across say three weeks, using the data from three
consecutive Mondays would provide up to 108 values of night time LA90s.
The lowest 10% of these readings would be rejected, and the lowest in
the remaining set of results, would retained as the LA90 for Monday nights.
Noral Stewart
And the result is thus the 90 percentile value of the 15 minute 90
percentile values for those three weeks which may be unrepresentative of the
rest of the year.
Ken Plotkin
<mars...@pacific.net.au> wrote:
>Over a short period arithmetic averaging is OK but what are you
>attempting to describe. Over say, a day of 15 min samples, you would
>probably be looking for the "lowest" L90 values. Then, do not even
>think of averaging, think about what you are looking for!
[snip]
I'm thinking of continous hourly samples for a couple of years, trying
to get a grip on what the soundscape is and how it varies from season
to season as well as diurnally.
Keep in mind that Leq is widely used because it is much easier to
model than other metrics, and mathematically appealing, and not
necessarily the most representative measure of noise.
Ken Plotkin
Brian Marston
repeated at different times the consistency has been very good between
different consultants measurements. We do what we can to avoid those
periods of elevated LA90 that may occur due to public holidays and the like.
Longer than a day provides an indication of the the daily noise
patterns. Longer than a week provides an indication of the the weekly
noise patterns. Three weeks or more, brings in meteorological
fluctuations due to the passage of High and Low pressure cells.
When I first started, measurements were generally "eye-ball averaging"
off an analogue SLM over about 5 minutes at about midday, again at about
8:00pm and finally around midnight.
When it became possible to electronically obtain the parameters over at
least 15 minute periods, I was one of the first in our region to do
24-hour continuous attended monitoring scribbling down the results at
the end of each 15 minute period.
:(
Brian Marston
the soundscape. LAmax, LAmin, LA01, LA10 & LA90 provides more of a feel
for the soundscape.
Analysis of a couple of years of data would be a very interesting
exercise by season by month by week.
That was why we went to the 90 percentile analysis of the LA90 & LAmin
data, and a 10 percentile analysis of the LAmax, LA01, LA10 data. It cut
the data down into handleable daily chunks that could be compared week
to week. The LAeq got a double dip with the original data getting both
10 percentile analysis and 90 percentile analysis to provide a range. I
find LAeq data can be too easily contaminated by foreground noise in
environmental investigations.
:)