Search

Advances in artificial intelligence for computer vision hold great promise for increasing the scales at which ecological systems can be studied. The distribution and behavior of individuals is central to ecology, and computer vision using deep neural networks can learn to detect individual objects in imagery. However, developing supervised models for ecological monitoring is challenging because it requires large amounts of human-labeled training data, requires advanced technical expertise and computational infrastructure, and is prone to overfitting. This limits application across space and time. One solution is developing generalized models that can be applied across species and ecosystems. Using over 250,000 annotations from 13 projects from around the world, we develop a general bird detection model that achieves over 65% recall and 50% precision on novel aerial data without any local training despite differences in species, habitat, and imaging methodology. Fine-tuning this model with only 1000 local annotations increases these values to an average of 84% recall and 69% precision by building on the general features learned from other data sources. Retraining from the general model improves local predictions even when moderately large annotation sets are available and makes model training faster and more stable. Our results demonstrate that general models for detecting broad classes of organisms using airborne imagery are achievable. These models can reduce the effort, expertise, and computational resources necessary for automating the detection of individual organisms across large scales, helping to transform the scale of data collection in ecology and the questions that can be addressed.

Antarctic krill (Euphausia superba) and salps are major macroplankton contributors to Southern Ocean food webs and krill are also fished commercially. Managing this fishery sustainably, against a backdrop of rapid regional climate change, requires information on distribution and time trends. Many data on the abundance of both taxa have been obtained from net sampling surveys since 1926, but much of this is stored in national archives, sometimes only in notebooks. In order to make these important data accessible we have collated available abundance data (numerical density, no. m−2) of postlarval E. superba and salps (combined aggregate and solitary stages and species) into a central database, KRILLBASE, together with environmental information, standardisation and metadata. The aim is to provide a temporal-spatial data resource to support a variety of research such as biogeochemistry, autecology, higher predator foraging and food web modelling in addition to fisheries management and conservation. Previous versions of KRILLBASE have led to a series of papers since 2004 which illustrate some of the potential uses of this database. With increasing numbers of requests for these data we here provide an updated version of KRILLBASE that contains data from 15,194 net hauls, including 12,758 with krill abundance data and 9,726 with salp abundance data. These data were collected by 10 nations and span 56 seasons in two epochs (1926–1939 and 1976–2016). Here, we illustrate the seasonal, inter-annual, regional and depth coverage of sampling, and provide both circumpolar- and regional-scale distribution maps. Krill abundance data have been standardised to accommodate variation in sampling methods, and we have presented these as well as the raw data. Information is provided on how to screen, interpret and use KRILLBASE to reduce artefacts in interpretation, with contact points for the main data providers.