Prediction of krill swarm characteristics off East Antarctica

Brief description

This dataset contains estimates of krill swarm characteristics from statistical models based on underway acoustic observations along with underway and remote-sensed environmental data. Estimates of internal swarm density and depth across the study region (60-80 degrees E) are included for the time of the survey (Feb 2006). Estimates of February internal swarm density across the broader East Antarctic region (30-120 degrees E) are also included for the period 2001-2010.

Lineage

Progress Code: completed

Notes

Purpose
The acoustic data were collected during the multi-disciplinary Baseline Research on Oceanography, Krill and the Environment (BROKE)-West survey off East Antarctica in the austral summer of 2006 (Nicol et al. 2010). We used the easternmost half of the BROKE-West survey transects (T7-11) covering 60-80 degrees E. Individual krill swarms were isolated by applying the schools detection algorithm of (Barange 1994), implemented in Echoview. After (Tarling et al. 2009) school detection was carried out on a 7x7 identity matrix convolution of the 120 kHz pre-processed data using the detection parameters of (Tarling et al. 2009) and a mean volume backscattering strength Sv threshold of -70 dB re 1 m -1. These aggregations were classified as krill or not by applying the validated 'dB-difference' technique (Watkins and Brierley 2002) to the 7x7 convolution of 120-38 kHz pre-processed data falling within the detected aggregation boundaries. As implemented here, the dB-difference technique is a binary classification, and aggregations falling within a dB difference range were deemed to be krill. Outside of the dB difference range an aggregation was classified as coming from other species and excluded from further analysis. The dB-difference ranges were calculated using the krill acoustic target strength model (Calise and Skaret 2011) and based on length frequency distribution clusters (Kawaguchi et al. 2010). Once identified, volume integrations were carried out on 120 kHz falling within krill swarms at a -80 dB threshold and swarm internal density rho was calculated using rho = 10^(Sv-TS_kg)/10, where TS_kg is the target strength of 1 kg of krill. Statistical models were developed to predict two specific characteristics of individual krill swarms (1) internal swarm density (g m-3), and (2) mean vertical depth (m) of the swarm in the water column. A mixed-effects modelling approach was adopted to account for the inherent structure in the krill survey observations. We used Generalised Additive Mixed Models (GAMMs), with a Gaussian error distribution, implemented via the R package mgcv version 1.8-6 (Wood 2006a, b). Fourteen potential explanatory biophysical variables were initially considered on the basis of ecological relevance and also taking into account the availability and quality of data. A complete list of these predictors including details such as their original source and spatial resolution, and any data processing, are provided in Supplementary material Appendix 2, Table A2.1 of Bestley et al. (submitted). For the list of final variables selected for each model, and their relative importances and effects, see Bestley et al. (submitted). Spatial predictions of krill swarm characteristics (internal density and depth) were developed across the survey domain, from the model set built using remotely sensed predictor variables. To generate these, biophysical predictors were extracted for February 15th 2006. The results from the four different model structures, and the 10-fold data-subsampling procedure, were then summarised using the mean and interquartile range (IQR) of these (i.e. n = 40 model runs) per spatial grid cell. A climatological krill internal swarm density layer was also produced for the broader East Antarctic region (30-120 degrees E). To do this we made spatial predictions for four representative dates in February (2nd, 10th, 18th and 26th) across the 10 year period 2001-2010, using remotely sensed data. Predictions were made using the four models fitted to the full observational dataset to provide a total of 160 model fits (4 dates x 10 years x 4 models). The predictions were summarised as described above using the mean and interquartile range per grid cell. References Barange, M. 1994. Acoustic identification, classification and structure of biological patchiness on the edge of the Agulhas Bank and its relation to frontal features. S Afr J Mar Sci 14: 333-347. Bestley S, Raymond B, Gales NJ, Harcourt RG, Hindell MA, Jonsen ID, Nicol S, Peron C, Sumner MD, Weimerskirch H, Wotherspoon SJ, Cox MJ (submitted) Prediction of krill swarm characteristics that drive a marine predator 'hotspot' off East Antarctica. Ecography Calise, L. and Skaret, G. 2011. Sensitivity investigation of the SDWBA Antarctic krill target strength model to fatness, material contrast and orientation. CCAMLR Sci 18: 97-122. Kawaguchi, S. et al. 2010. Krill demography and large-scale distribution in the Western Indian Ocean sector of the Southern Ocean (CCAMLR Division 58.4.2) in Austral summer of 2006. Deep Sea Research Part II: Topical Studies in Oceanography 57: 934-947. Tarling, G. A. et al. 2009. Variability and predictability of Antarctic krill swarm structure. Deep-Sea Res I 56: 1994-2012 Watkins, J. L. and Brierley, A. S. 2002. Verification of the acoustic techniques used to identify Antarctic krill. ICES J Mar Sci 59: 1326-1336 Wood, S. N. 2006a. Generalized Additive Models: An Introduction with R. Chapman and Hall/CRC Press. 659 Wood, S. N. 2006b. Low rank scale invariant tensor product smooths for generalized additive mixed models. Biometrics 62: 1025-1036