Territory plots from CBC-style bird survey 2020
Analysis of CBC-style data
The plots shown here summarise territory data for 2020.
Relevant data were recorded using basically the same methods as in the BTO's Common Bird Census (CBC). The main difference was that data were recorded on many days, on an 'ad hoc' basis.
Initial approximate territories were estimated using various standard clustering algorithms.
These clustering (or 'classification') algorithms do not take into account relational information (row to row info) and need to be corrected.
Summary of results
- Territories can be estimated successfully for species where about 20 or more records are available.
- In some species adjacent territories appear to overlap to some extent (e.g. blackcap, whitethroat).
- Territories seem to vary in size with some species. There may be only one large blackbird territory on the meadow but the clustering suggests two smaller ones.
- More reed warbler (4 or 5) and sedge warbler (3 or 4) territories were discovered than expected.
- Recording is almost certainly biassed towards those species that spend much time in raised or exposed locations, and against those that stay low down in thick cover (such as reeds).
Clustering fault correction
Two types of fault were identified:
type | fault | remedy |
---|---|---|
TYPE.0 | Two points are assigned to the same cluster but relate to individuals from apparently different territories (e.g. two singers, registered simultaneously) | Assign one of the points randomly to a different cluster |
TYPE.1 | Two points are assigned to different clusters but originate from just one indiividual, for example, when the singer moved | Choose one of the points randomly, set the cluster to be the same as for the other point |
The randomisation stage makes it necessary to run a number of trials and then select the best-performing trial. Twenty-five trials for each k value were thought to be sufficient.
Performance was scored by calculating fracSS, the fraction of total sum of squares accounted for by the clustering process, after 'correction'.
The best value for the number of clusters was assessed from the raw plots and the dendrograms, some of which are reproduced below. The increase in fracSS as k was increased also guided the process.
Dendrograms from clustering algorithms
Points are numbered in order south to north (i.e. the point furthest south is point 1).
Approximate numbers of territories
(Note that some of these territories overlap into land outside the 1.5 hectare plot.)
species | number | notes |
---|---|---|
blackbird | 3 (or 4) | Clustering suggests two meadow territories. More likely the meadow held a single territory, with some overlap with an external territory north-west of the plot |
blackcap | 4 (or 5) | Two of the territories are largely outside plot boundary. Some individual variation in song characteristics |
chiffchaff | 4 (plus 2 nearby) | Territory at top blackthorn (on meadow) was probably temporary |
reed bunting | 3 (or 4) | Diifficult to discern territories |
reed warbler | 5 (plus 1 temporary) | Simultaneous registrations support at least 4 good territories. Two more clusters may represent a single singer that moved to the east side (points 40, 55, 57, 59). Territories are mostly very compact (10-15m across, see diana dendrogram) |
sedge warbler | 3 (or 4) | Singers go quiet for long periods, hide in cover |
song thrush | 3 (or 4) | Hard to discern true number |
whitethroat | 4 | Territories overlap at Top Blackthorn |
wren | 6+ | Some activity in reeds almost certainly missed |