Date: 9/1/25 10:49 am From: Matt S. <accipiter22...> Subject: [MASSBIRD] Late Post: August 30, 2025 nearly 200,000(!!!) Tree Swallow Cloud at Parker River, Newburyport
Hi all,
Through the Friends of Parker River, I had heard that there were large
flocks of tree swallow congregating over the impoundment north of the
Hellcat Dike in the evenings. I have not been to Parker in the evening in
ages, and the chance for more tree swallows, plus phalaropes and knots
being spotted were enough to send me out on a Saturday evening. The knots
were at the pan, three of them, and after watching them for a while I
headed to Hellcat to see if the tricolored was still at the Old Blind. It
was not. But I noticed lots of tree swallows overhead, a few hundred, which
was nice. I headed back to the lot and saw some familiar cars there and
decided to head to the dike to see if any birding friends were around.
There were several people lined up watching egrets, and some yellow-crowned
night heron.
Someone pointed out that there were large masses of tree swallows in the
distance, to the north of the dike. I saw SOME, but not a ton, then I
brought up my bins and started exclaiming. The magnification brought
absolutely ENORMOUS clouds of tree swallows into view. I panned across
maybe 45-60 degrees of horizon and it was top to bottom tree swallow. I
have seen large flights during the day, large flocks…and in the moment I
estimated maybe 40,000 were out there. I got my camera up and took a rapid
series of shots, moving my camera enough to eliminate overlap, and doing so
quick enough (and against the grain of flock movement) to try and prevent
birds from looping back and getting double counted. I ended up with a 28
shot panorama, and looking at some of the shots I had a feeling there may
have been way more than 40K in the flock; some of the shots looked like
they had almost 10,000 birds in them.
When I got home I prepped the photos and uploaded them to ChatGPT, and
using Python ran SmallBlobAnalysis on each photo, then had it search for
overlap in the shots, and correct for that. I also assumed there were some
birds looping back, and deducted a certain percentage from the overall
number. I also did not capture the “top” of the flock, so added a
percentage back into the count. The analysis showed that there were
SEVERAL frames in the panorama that had 10,000 or more tree swallows in
them. The final count: 188,971!!!!!!!!!!!! You can see the methodology
described below, and see a sampling of photos in the eBird list which is
linked.
Birding has given me many happy, fun, and breathtaking moments in my life.
As far as breathtaking, I do not think anything tops this thus far in
life. I have heard at the mouth of the Connecticut River you can see 1
million tree swallows, I may have to check that out next year. In the
meantime, this was the greatest natural spectacle I have witnessed in
person. My goodness. If that’s not worth protecting, what is?
That's all for now,
Matt S.
Newton, NH
<Accipiter22...>
---------------
Parker River NWR, Essex, Massachusetts, US
Aug 30, 2025 6:18 PM - 7:46 PM
Protocol: Traveling
4.87 mile(s)
25 species
Mallard (Anas platyrhynchos) 2
Mourning Dove (Zenaida macroura) 1
Common Nighthawk (Chordeiles minor) 1
Semipalmated Plover (Charadrius semipalmatus) X
Lesser Yellowlegs (Tringa flavipes) 13
Greater Yellowlegs (Tringa melanoleuca) 3
Red Knot (Calidris canutus) 3
Least Sandpiper (Calidris minutilla) 8
Semipalmated Sandpiper (Calidris pusilla) X
Ring-billed Gull (Larus delawarensis) 1
American Herring Gull (Larus smithsonianus) 2
Double-crested Cormorant (Nannopterum auritum) 27
Yellow-crowned Night Heron (Nyctanassa violacea) 2 Pics, north of dike
Black-crowned Night Heron (Nycticorax nycticorax) 1
Snowy Egret (Egretta thula) 22
Great Egret (Ardea alba) 13
Great Blue Heron (Ardea herodias) 1
Osprey (Pandion haliaetus) 3
Red-tailed Hawk (Buteo jamaicensis) 1
Great Crested Flycatcher (Myiarchus crinitus) 1
Tree Swallow (Tachycineta bicolor) 188971 Flock Count Analysis Summary:
Per-photo counts (not all photos are attached, space is limited here. I can
provide them all if needbe):
Notes per ChatGPT BlobDetector analysis:
- Overlap detected primarily in AJ9A1586–AJ9A1589 and AJ9A1590–AJ9A1591
regions.
- Later batches (1595–1604 and 1605–1612) showed minimal to no overlap,
indicating distinct frames.
- Counts are based on strict blob-detector settings with red dot overlays
for visual verification.
I took the pictures, attempting to have little to no overlap in them, in
rapid succession, ending up with 28 photos altogether, over 24 seconds.
Note that I did not get the entire flock, there are no pictures showing
clean sky/clouds above a group of birds, for instance. I did get most of
the left and right flanks, and a good portion of the bottom. I tried to
pan in a way that went against the movement of the cloud to avoid getting
pictures of the same birds traveling through the various frames. Obviously
this will not be 100% successful, and the movement was more of a boiling
mass as opposed to a steadily moving flock going in one direction.
I then had ChatGPT run a SimpleBlobDetector analysis (you can see the
parameters below, full Python code is saved). I have uploaded several
samples of the blob analysis, and have included some of the raw photos as
well. I also included the overlap subtractions, and notes from the AI. I
was surprised by the overall accuracy of the blob analysis. I would
estimate that only roughly 5% of the dots were false-positives, and that
may be high from some scanning I did. I would also estimate that it missed
less than 0.5% of potential birds, and have not adjusted the raw account
for this minute amount. I also assume that as many as 10% of the birds may
be cycling back through, though this is more of an educated guess. Again,
I tried to pan in a way that was against what little grain of movement
there was in the flock, while attempting to have a space buffer between
each frame. The buffer was not always present, as I rapidly moved through
shots to try and capture it as completely as possible. However, that
buffer would add additional birds to the count. I have not adjusted the
count up for this however, as I have no way of estimating the impact.
I would also estimate that I left perhaps 8% of the flock above the maximum
height of these photos. Taking the overlap-adjusted total of 205851, and
subtracting 5% for false positives, and another 10% for swallows cycling
back through from that overlap-adjusted total, you get 174973 swallows in
the pictures. Adjusting up for the estimated 8% that I did not capture,
yields 188971.
Note that totals at tops of photos are pre-overlap-adjustments, although
total adjustment was relatively small.
