New Southern Ocean transfer function for subsurface temperature prediction using radiolarian assemblages

Abstract

Radiolarians are microzooplankton that produce siliceous shells that preserve well in sediments and allow for paleo-reconstructions. Previous studies have used them for sea surface temperature (SST) reconstructions. However, radiolarians peak in abundances between 100 and 400 m in the Southern Ocean (SO), suggesting that their assemblages are more representative of subsurface conditions. Here, we aim to develop a SO-wide transfer function (TF) for subsurface temperature reconstructions (subST) using the Southern Ocean RAdiolarian Dataset (SORAD), a compilation of data for about 240 radiolarian taxa in 228 surface sediment samples from the SO. This exhaustive dataset has been simplified using common TF criteria to minimize noise and optimize SORAD for subST prediction. Ordination tests and Q-mode Factor Analyses (QFA) were applied to the resulting training dataset, which includes 212 samples and 75 taxa (SORAD212_75). The results suggest that, out of six environmental variables, radiolarian assemblages in SORAD are mainly driven by summer temperatures at 200 m and that the first five factors of the QFA explained over 75% of SORAD212_75 variance. We applied common TF methods, IKM, MAT and Weighted-MAT (WMAT), to two data transformations (relative abundances of species and log-transformed). The six models of modern temperature predictions all show excellent performance over the −2 to 18 °C interval. To compare with previously published results generated with the same method, IKM% was applied to radiolarian census data of three cores from the Atlantic, Indian and Pacific sectors of the SO, thus calculating new subST reconstructions and testing the new TF performance.