At the national scale, we compared deterministic and stochastic Leslie matrix models (which are usually used in practice for PVAs) against each other, and against a range of simple time series growth models. We applied these methods to data on abundance, breeding success and survival for 15 seabird species, for breeding colonies throughout the British Isles. We evaluated performance by applying the methods to abundance and demographic data, with these split in to a training period and a subsequent “test” period, assessing whether the predictions that the methods generated for the “test” period were consistent with the observed counts of abundance for that period. We considered four possible definitions of the test period – 1998-2017, 2003-2017, 2008-2017 and 2013-2017 – and in each case considered the training period to be all years (with suitable available population data) prior to this.

In one region (Forth/Tay), for five species, we also used a five year test period (2013-2017) to compared this suite of approaches against the Semi-Integrated Population Models (SIPMs) produced by Freeman et al. (2014) and Jitlal et al. (2017) (termed Bayesian State Space Models in those reports).

We assessed performance of methods by looking at: a) whether the method was possible to apply; b) how frequently it yielded “highly implausible” results (i.e., results that are more than 100 times larger or smaller than the actual abundance); c) whether it produced systematically biased results (i.e. over-estimated or under-estimated actual abundance); d) how much error the methods had, on average, in predicting the observed count; e) whether the method provided an accurate quantification of uncertainty; f) the level of uncertainty associated with each method and g) the computational time required to implement the method.