General hypothesis

  • Building on Ryan’s 2020 PNAS paper about the timing of ‘cold snaps’ relative to breeding for tree swallows. He showed that in our Ithaca TRES population:

    • Springs are warmer over last 100 years
    • Despite that, the date of the last ‘cold snap’ (1/2/3 days with high < 18.5 C) has not changed
    • Birds breed earlier by ~9 days
    • Because of breeding earlier they are more often exposed to cold snaps while nestlings are at vulnerable age
    • Year to year, the date of last cold snap relative to breeding is strong predictor of breeding success

  • Idea for this paper is to essentially repeat that study but with many sites/years/species and to explicitly compare breeding performance for aerial insectivores (swallows, martins, flycatchers) to non aerial-insectivores (blubirds, chickadees, nuthatches, titmice, etc) that should not be as impacted by short cold snaps.

  • Predictions are the same as in the PNAS paper with the additions that:

    • Pattern should be present for aerial insectivores but not (or weaker) in non aerial insectivores
    • Geographic variation in timing of cold snaps relative to lay date/warming might modify where relationship is strongest

Cleaning Up

Cleaning breeding data for inclusion

  • Clean input data to these criteria
    • Has location.
    • Has clear date for at least egg laying/incubation start (ideally for hatch). When hatch date isn’t available, infer hatch based on lay date, clutch size, and species average incubation.
    • Has fate information fledge/failed (ideally with number fledged).
    • At least 500 records for the species in total.
    • For analyses by grid, has at least 10 records for the grid/year for that species.
    • I took out a few species that fit these criteria but weren’t worth including (e.g., starlings & house sparrows have tons of nest failures that I think must be intentional destructions?).
  • Filtering biologically implausible values
    • I plotted clutch size, lay date, and fledged number for each species to look at the distribution. Based on those plots, I filtered out values that were implausible for each species. I guess this was kind of a judgment call, but most of these are clearly implausible (e.g., breeding mid-January or a clutch size of 80). Probably this cleaning isn’t perfect. The exact ranges used are included in the main R code. We could make this into a supplemental table or something or just leave it in the code.

Cleaning and processing temperature data

  • Import temperature from NOAA using rnoaa package and choose stations that:
    • Have records starting in 1940 or earlier with max daily temperature (this leaves ~2,600 stations). Many of these stations go back <1900 but the number gets smaller that early.
  • Processing these temperature records:
    • Bin stations into the hexagonal grids (see below).
    • Average temperature for each day/year for all stations that appear in the same grid.
    • For each year/grid, identify the latest day that 1/2/3 day cold snap occurred in each year. Threshold = 18.5 C maximum daily temperature.
    • For counting cold snaps, limited to days of year 50 to 240 (otherwise you get fall cold snaps). Should we limit this more or by different grids?
    • Counted these

Standardizing breeding performance

  • For each species, calculate change in breeding phenology over time? For TRES we had records back to the 60/70s, but most of these only start in mid 90s so not sure how much we can detect here.

  • For each breeding record, calculate relative reproductive success. Will need to account for lat/long variation in average clutch/fledge size somehow. Also need to standardize across all species so that RS is a z-score of RS relative to species mean and relative to the expected RS for that particular location. This is actually going to be tricky because of course to the extent that temperature changes the clutch size pattern with latitude it might change some over the 25 years of data.

  • For each breeding attempt, calculate the offset in days from hatching to the last cold snap of the year using above criteria. Could get into more weather data specifically experienced for each nest but maybe that is best to skip at first.

  • Ask does date of last cold snap predict fledging RS (or fledge yes/no) and does the pattern differ for aerial insectivores vs. non aerial insectivores. Maybe need to also run this as phylogenetic correction for relationships?

Defining space used

Breeding grid

I made a hexagonal grid with 3 degree width hexagons. Because these are equal degree size, the actual land area varies a bit as you go north. I don’t think that is a problem since the total area doesn’t really factor into any analyses. I also plotted 1 or 2 degree grids, but those seem too small to work with in terms of having decent numbers of breeding records, etc. Of course the grid size is somewhat arbitrary here.


NOAA Stations

I downloaded daily weather data from a little over 2600 NOAA weather stations that started recording <1940 and had records from present day (some do have gaps in time). Here they are. You can see that these really only cover the US with a few in Canada, but the breeding records are also almost exclusively US so I think it’s fine to not worry about getting more records from farther north unless we find access to breeding records to go along with them.


Basic plots of breeding data

We have records from NestWatch at Cornell for a bunch of species going back to early/mid 90s (some older, but sample sizes get much smaller). We also have records for Purple Martins from a recent MartinWatch dataset archived with a JAB paper that we could add. Ryan has also was thinking we might be able to add a similarly large database of European nest records if we want to bring on a couple more collaborators.

Between the two US datasets, there are >400,000 nesting records, though that will definitely be reduced somewhat after filtering for having the info we need, etc and what exactly we decide to include.

Included breeding records

Here are all the records pooled over years and grouped together by similar species (number of records per group in corner of each plot).


Breeding records

This is the distribution of lay dates by day of year for each species. Aerial insectivores vs. not and cavity nesters vs. not are split by color. Numbers are the total number of records for the species in the database (though at this moment that number includes records that will be filtered out by cleaning).

Lay date