Spatiotemporal modelling of abundance from multiple data sources in an integrated spatial distribution model

Abstract

AbstractAimIn biodiversity monitoring, observational data are often collected in multiple, disparate schemes with greatly varying degrees of standardization and possibly at different spatial and temporal scales. Technical advances also change the type of data over time. The resulting heterogeneous datasets are often deemed to be incompatible. Consequently, many available datasets may be ignored in practical analyses. Here, we propose a more efficient use of disparate biodiversity data to assess species distributions and population trends.LocationSwitzerland (Europe).TaxonBirds.MethodsWe developed an integrated, hierarchical species distribution model with a joint likelihood for all datasets using a shared state process (e.g. latent species abundance or occurrence), but distinct observation process for each dataset. We show how the abundance submodel of a binomial N‐mixture model can fuse four different data types (count, detection/non‐detection, presence‐only and absence‐only data) and enable improved inferences about spatiotemporal patterns in abundance. As case studies, we use data from multiple avian biodiversity monitoring schemes. In the first, the goal is estimating abundance‐based species distribution maps. In the second, we infer trends in population abundance across time.ResultsAccuracy and precision of abundance estimates increased when combining data from different sources compared to using a single data source alone. This is particularly valuable when data from each single data source are too sparse for reliable parameter estimation.Main conclusionsWe show that exploiting the complementary nature of ‘cheap’, but abundant, citizen‐science data and less abundant, but more information‐rich, data from structured monitoring programmes might be ideal to estimate distribution and population trends more accurately, especially for rare species. Joint likelihoods allow to include a wide variety of different datasets to (1) combine all the available information and to (2) mitigate weaknesses of one by the strength of another.