Breeding of White-browed Woodswallows in the Canberra area

The current sightings of breeding White-browed Woodswallows (WBWS - Artamus superciliosus) 

stimulated Steve Wallace to summarise the official records for Woodswallows in the ACT and circulate his summary to the COG Chatline.   I was interested to compare the 20% of records for WBWS having a breeding element with the 2% of COG Garden Bird Survey (GBS) records (for all species) showing breeding.   (As a general point for about 65% of the year there will be few if any breeding records whereas the WBWS tend to be recorded only in the breeding season.  However, the % of breeding records for WBWS still looked impressive.)

I was taken with Jack Holland's suggestion that when the WBWS hang around a bit they tend to exhibit breeding behaviour  Steve provided me with an extract from a dataset he maintains which includes both General Records and GBS data.  (For the purposes of this analysis the differences in methodology are felt to be not significant.)

As this gets a bit long and involved I can summarise my thoughts as two points.

• When there are a higher number of sightings of WBWS prior to November it is likely that breeding will be reported that year; and
• Larger flocks are more frequently reported in October than other months suggesting movement from over Wintering areas searching for food.

A first issue was that trying to analyse anything by calendar year was hopeless.  Being migrants WBWS start appearing (typically) around September and quite often hang around into the New Year. Of the 501 records 95 occur in the first half of a calendar year and are clearly part of a financial year based season (not that I am suggesting their behaviour is influenced by taxation considerations).  There is one “odd-ball” record for June which could be either very early or very late but it’s importance is I think low.  (The single record for August is almost certainly early.)  

So, with a bit of fiddling around I added a financial year code to the subset of data.  I also entered null records for the 4 financial years in which no WBWS were recorded in any surveys.

My next steps were to calculate 

• the number of days between first and last sighting for the calendar year; and
• the total number of Breeding records for each year

I then got Excel to give me a measure of correlation between the series “Duration of visit “ and “# breeding records”.  This was not great, at .0.46, especially since both series are 0 when no WBWS were sighted in a year (and a year with breeding records must have at least 1 observation record).   Here is a scatterplot of the two series.

I regard it as important to focus on the word "scatter" in interpreting this chart.  Not quite a randomplot but heading in that direction.

Another possible correlate was the number of reports in a year.  So I simply counted (OK got ACCESS to count) the number of reports in a year and Excel delivered a correlation coefficient between that variable and the number of breeding records previously acquired. 

Before Steve sent me the data he had identified 2002-03 as an outlier year as when plotting the number of obs records against the number of breeding records that year was well off the trend line.  Deleting that year from the series improved the correlation coefficient significantly. 

I was interested to identify why the year was unusual – it seemed that there were considerably fewer than expected breeding records for the number of observation records.  On looking at the data it was apparent that an influx of WBWS coincided with a Woodland Survey (WOO) weekend.  This is a time of high observer activity and 15 records were from that source in that year.  No other year recorded WBWS in WOO.  Thus for my purposes I have deleted the WOO records from my data set.

The result was quite astonishing, given the poor outcome from ‘length of observation period”  The coefficient was 0.91 which will certainly do me for significant!  (Of course there are a few variable independence issues since there must be >0 observations for there to be a breeding record. Here is a scatterplot.

The linear trend fits quite well (R² 0.83) but the 4^th order polynomial (I use that style of poly as it seems to be a fair compromise between fitting to a non-linear series and not chewing up too many degrees of freedom) has a very good fit at R² = 0.94. 

How I interpret this is that simply having a small number of birds “book-ending” a long season is not enough to generate a lot of breeding records.  It is a matter of there being enough birds around to be widely seen that indicates they are likely to start breeding.  This may then mean that the birds will be around for a fair while to allow the young to fledge and develop.

My next thought on this matter was to look at the timing of various breeding indicators.  There are about 12 of these in the COG data set (8 of which have been reported for WBWS) and I summarise them into three groups:

1. Pre-nesting (Display; Copulation, and Nest Building);
2. Active Nest (Nest with Eggs; Nest with Young, On Nest, Carrying Food);
3. Post nesting (Dependent Young) 

Plotting the number of occurrences by month gives the following.

This suggested that if the birds aren’t around by October/November they will be too late to start breeding activities.  Looking at the data there seemed to be a pattern of a (relatively) large number of observations before November correlating with breeding being reported in that season.  This is summarised by the following table.

	Breeding observed
4+ flocks before nov	Yes	No	Total
Yes	10	3	13
No	2	18	20
Total	12	21	33

This seems to confirm an hypothesis that if there are a large number of observations made before November in a year it is likely that the birds will “settle down” and breed.  In one of the cases where breeding was observed with few reports before November, there were 5 reports in November and 19 in December (so things were just a little late).  In the three cases where there were >4 reports prior to November and no breeding the number of reports was 5 or 6 in October and 2 or less reports in the rest of the year (ie overall a fairly weak WBWS turnout).

As part of an off-line discussion of the phenomenon Geoffrey Dabb raised some issues relating to flock size, likelihood of breeding; and whether the big flocks are just “passing through”.  The following table shows the number of flocks by size class of flock and month.  (Note: there are less flocks shown here than the 501 referred to above as some of the reports have been of presence/absence rather than giving a number of birds in the flock.)

Month	Flock size class
	1-10	11-20	21-40	41+	Total
Aug	1				1
Sep	7	3	6	7	23
Oct	60	21	16	29	126
Nov	66	19	15	9	109
Dec	70	19	11	8	108
Jan	43	11	6		60
Feb	8	3			11
Mar	5	1	1		7
Apr	9				9
May					0
Jun	1				1
Total	270	77	55	53	455

This shows the over-representation of October as a time for 'big flocks'.  Possibly this can be interpreted as the birds leaving their Wintering grounds in a big flock searching for food resources and when they find one they settle down and breed.  When conditions are not favourable - ie there is little food around - in this area they may be seen a small number of times but move on.

I may try to explore this area a little more, but at the moment can't quite see how to massage the data to do this (and I have taken up enough of your download limit anyway).