Integrating space and time in animal abundance prediction

Estimating animal abundance in space and time is (and will remain) a challange, even for dense observational networks like the ornithological monitoring network in the Netherlands. Accurate estimates of species abundance at large scales are important to detect trends and learn about the effects of environmental changes. We investigate the adequacy of an ensemble Kalman filter to enhance abudance estimates by using time series and survey data in combination rather than separate. Via synthetic examples it is shown that the description of a spatio-temporal process by a series of independent spatial models leads to a significant loss of information (and consequently inaccurate results). A time series analysis on spatially lumped units leads to inaccurate results in a similar way. A considerable improvement is made when a spatial or time series model is placed in a framework where both survey and time series data are assimilated with an ensemble Kalman filter (prediction errors are roughly halved). Also when noise is added to the system or the observations, the data assimilation system outperforms isolated spatial or temporal analyses. We apply this model framework to observations of Buzzard, Lapwing and Starling. The results are compared with those from separate spatial models (regression-kriging) and spatially-lumped time series models (ARMA). With data assimilation, cross-validation errors are considerably lower for Lapwing and Buzzard but not for Starling. This is explained by the relatively dynamic behaviour of Starlings which is not captured with the current observation frequency. Spatial as well as temporal estimates apear to be quite different between the different models, but consistent across the different species. The spatial models give relatively high estimates of bird abundance (60% more than the results from data assimilation for the entire Netherlands, averaged over the entire period), while time series models lead to lower estimates of bird abundancies by (35% less than the results from data assimilation). Our next research aim is to apply the data assimilation method to estimate the abundance of 60 species in the Netherlands over the past 25 years, starting from a set of spatial models (http://ecogrid.sara.nl/bambas/) and weekly counts at 6 airfields.