Sunday, 24 November 2024

BirdNetPi: 12 months on, what now?

My system has collected a lot of data.

Can it tell me anything?

And what do I do now?

the current system

My mcguire89 based BirdNet-Pi system has been quietly sitting in the loft, gathering data via a small microphone hanging out the roof at the back of the house. While its range is naturally limited, it certainly identifies some of the birds in my back garden, some of the noisy birds on my roof, and a few of the louder birds flying overhead.


There are a number of adjustable settings on this system, most of which I've left set to the default values.

The minimum % confidence level was set to 75% for most, but not all, of the year. I don't know exactly what %confidence represents but should imagine that (in this case) a given detection matches 75% of the reference points for a given species.

Resample resolution in the graphs below was set to 15 minutes; which I think means several detections of the same species within 15 minutes are registered as one count.

Species Occurrence Frequency Threshold is currently at its default value of 0.03. This means that for a detected species to be accepted by the system as a valid identification, it must have been reported on at least 3% of the eBird submitted checklists, for my location and during a given week.

I may lower this threshold for next year, as I keep thinking it may exclude valid detections in a location for close to the 'bird magnet' that is the Pagham reserve.

One other point is that I lost a few recording (they were purged by the system) due to lack of storage capacity, but no data was lost from the database. I've since fitted a bigger USB drive, so hopefully this wont happen again next year.

This system has two rather annoying issues;

1) The Species Stats seem to be compiled up until the last time/date the system was re-booted. Not a huge problem as I simply have to remember to reboot if I want to look at up-to-date graphical information.

2) The spectrogram image contains a spurious signal at around 8-9kHz.


I've tried altering microphone and USB audio module cables several times. Sometimes it appears that I've fixed it, only to return later!

Also note that Patrick McGuire stopped work on this system software during 2023, so there have been no recent updates to the software.

Data for (mostly) 2024

The system lists 101 species of birds detected in the last 12 month (December to December).

I suspect some of these are species errors (maybe where only a small number of detections occurred for a given species) , but the 2 recordings of cuckoo in June appear genuine. And as we are just a mile or two from the Pagham RSPB Reserve, the vast majority are certainly possible (e.g. Canada Goose, Northern Lapwing, Whooper Swan, kingfisher & so on).

However the single, mid-day recording of a tawny owl does seem a little suspect. It scrapes in with a %confidence of 75.97%, although the recordings certainly include an 'owl like' sound (...and owls have been heard on our estate at night).

The Top 20 by detection count;

 

Nothing too surprising here except for the redwing and the number of detections for the chiffchaff.

From 100 to 1000 detections;


From 20 to 100 detections;


The mention of 'Dog' is because some of these detections sound like they may be due to a neighbour's dog barking.

Less than 20;

5 or less;



I'm still trying to make sense of this data, but thought some of it worth posting.

Just to state the obvious; if a bird doesn't make a sound, it can't be recorded by this system. I say this because (as an example) we have wood pigeons in the garden virtually every day of the year, but they are almost 'mute' during the winter, although extremely vocal for the rest of the time.

 

During the summer, a pair of blackbirds nested somewhere in our garden, and the chart shows a correspondingly high level of vocalisation.

But these were not sweet calls, they were 99% alarm calls as the pair followed a neighbour's cat, as it prowled along the fence line.

 

At some point in the future, systems like this will classify calls based upon type; e.g. Song #1, #2, #3, Alarm call #1, #2 & so on. (...or maybe even; Alarm call cat, alarm call sparrowhawk!).

The redwing is an example of a bird generally known for its night flight calls (NFC) and from this circular 'time' chart almost all occur during [winter] night time.

 

The blue tit & great tit calls both follow a fairly predictable pattern, increasing in number during the mating/nesting season.

For the rest of the year they make good use of our bird feeders.

The peak of calls is probably down to just a single male, while the winter calls may be due to any number of individuals, maybe from a mixed species winter flock.

 

I don't think the coal tits bred in either our, or our near neighbours gardens, so this chart probably mostly reflects visits to our birdfeeders.

 

Likewise, the long-tailed tits and green finch were mostly visiting the feeders.,

 

The hawfinch interests me. According to the RSPB;

"Hawfinches prefer well-wooded landscapes with mature broadleaved trees. During the winter months, they become more numerous and widespread as resident birds are joined by Hawfinches that fly in from northern Europe to escape the harsh winter weather. "

 

Although we've never seen one in our garden, they have been seen in nearby gardens.

Likewise the bullfinch, which does not hang on our feeders, but does poke around under hedges & shrubs, so can be difficult to spot.

 

The black redstart is a bird we saw a lot on our recent trip to Gibraltar & southern Spain. I now believe I may have spotted these in our garden during the summer.

 

Unlike the Pied wagtail and the grey wagtail, the white wagtail is described as a 'passage migrant' passing through Britain from early March to May, and then going south from mid-August to early November.

I don't seemed to have any detection data early in the year, so maybe they only flock and pause to feed-up here on the south coast just in the autumn, before crossing the water back into europe.

 

The chiffchaff data shows a big increase in detection during the autumn.

 

Most migrate south, however a small but increasing percentage are resident.

A reference on the net said: "Most of them arrive in March and have departed by September." 

In my data, the 'garden peak' for detections is around the last week in September. Quite a few must have hung around into October.

plans for 2025 system

I plan to build a new 'software' image, based upon the Nachtzuster fork of BirdNet-Pi during December, ready for January.

I'm experimenting with a different method of locating the microphone, which seems to be producing better results; basically the mic module is pushed into the end of a bamboo cane which is then pushed out through the gap between roof & rear wall.

 

This ensures the mic is pointing the right way and stops it being blown around is strong winds.

Once set up at the start of 2025, I plan not to fiddle with any settings during the year, thereby [hopefully] maintaining a consistent month-by-month comparison.

 

 

See also:-

 https://github.com/mcguirepr89/BirdNET-Pi

 https://github.com/Nachtzuster/BirdNET-Pi


No comments:

Post a Comment