Filling Data Gaps in the Southern Ocean: Fusion of Remote Sensing Observations with Historic Krill Data to Explain Coastal and Offshore Variability in Krill Abundance

Continuous sampling and analysis of data from the Atlantic sector of the Southern Ocean is key to monitoring rapid, stochastic ecosystem changes in the region. Antarctic krill (hereafter krill), the subject of this research, is particularly prone to regional warming, with a southward contraction of its habitat forecast. While coastal regions of the Antarctic Peninsula, South Orkney Islands and South Georgia are relatively well surveyed using trawl and acoustic survey methods, offshore environments represent significant data gaps. Many important stages in krill life cycle take place offshore including spawning and grazing, and moreover transit between the Antarctic Peninsula and South Georgia entails over 200+ days of advective oceanic (mainly passive) transport with Antarctic Circumpolar Current and its associated fronts. To fill in this significant data gap and infer patterns in temporal and spatial offshore distribution patterns necessitates the integration of diverse data sources, including primary historical observations of krill abundance from surveys and fishing activity as well as secondary observations from remotely sensed environmental variables.

Ideally, we could detect krill directly using hyperspectral imaging to measure the concentration of astaxanthin pigments in surface waters (Basedow et al. 2019). However, given such methods are still in development, we utilize Species distribution models (SDM) to infer spatiotemporal krill distributions. SDMs are models that relate abundance/ occurrence of species with environmental data for a given set of sample locations (Elith and Leathwick 2009).  In this research we use multivariable regression methods to build SDMs to predict krill abundance in relation to both static (geographical area, bathymetry) and dynamic (SST) environmental features, and numerical density of krill as a target variable. We explore the accuracy of several nonlinear methods including Random Forest and Boosted Regression Trees, through comparisons of model accuracy (R² values, standardized RMSE values and so on) and cross-validation. We then compare these predictions to eddy statistics calculated from satellite altimetry data, and phytoplankton concentrations derived from ocean colour data, with both products accessed through the Copernicus Marine Service. In this way, we will use SDMs for spatiotemporal predictions and use these mapped predictions to explain important relationships e.g. krill density as a function of eddy size.

References:

Elith, Jane, and John R. Leathwick. 2009. Species Distribution Models: Ecological Explanation and Prediction Across Space and Time. Annual Review of Ecology, Evolution, and Systematics 40 (1): 677–97. doi: https://doi.org/10.1146/annurev.ecolsys.110308.120159

Basedow, S.L., McKee, D., Lefering, I. et al. Remote sensing of zooplankton swarms. Sci Rep 9, 686 (2019). doi: https://doi.org/10.1038/s41598-018-37129-x