Estimates of Late Quaternary Sea-Surface Water Temperature Fluctuations in the Eastern Equatorial Pacific Ocean From an Artificial Neural Network (ANN) Analysis of Radiolarian Data
Abstract
The radiolarians and other microfossils have been widely used to estimate sea-surface temperatures (SSTs) by the Imbrie-Kipp Transfer Function (IKTF) method, which evolved in 1971. Besides IKTF, the modern analog technique (MAT), artificial neural network (ANN), weighted-averaging partial least squares (WAPLS) regression, and maximum likelihood (ML) methods were used to estimate SSTs based on radiolarian fauna, and the accuracies associated with these methods were compared. Study suggested that the ANNs represent more optimum procedure for estimating SSTs than these other methods. Relative abundance (%) data for 38 modern radiolarian species in the core tops from the Pacific Ocean (Pisias et al., 1997) were first calibrated to measured February-April and August-October SSTs (World Ocean Data CD, 2001) by the ANNs and the other methods. Average root mean square error of prediction (RMSEP) for the ANNs is considerably lower (1.5° for February-April and 1.4° for August-October) in 10 independent hold-back (HB) test sets of observations, that were not part of the training of ANNs, than for the other methods (RMSEP ~2°) using the same split of training and test sets. Correlation coefficients between observed and estimated SSTs by the ANNs in 10-HB sets for both the February-April and August-October seasons are 0.98. Results suggest that the ANN-derived SSTs are ~25% and ~19% more accurate than those derived by the other methods during February-April and August-October seasons, respectively. Therefore, we estimated paleo-SSTs by the ANNs using same radiolarian species down core data in 10 sediment cores from the eastern equatorial Pacific (Pisias and Mix, 1997), and paleo-SST time series of the last ~800,000 years (~800-ka) were prepared. Besides, the ANN derived SST maps were made for the modern time, Last Glacial Maximum (LGM, ~23-ka), and Eemian (~125-ka) time-slices in the cores and analyzed. The ANN based SSTs suggested that the Peru upwelling was stronger during the LGM than at the modern and the Eemian times.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2008
- Bibcode:
- 2008AGUFMPP23C1489G
- Keywords:
-
- 4900 PALEOCEANOGRAPHY (0473;
- 3344);
- 4944 Micropaleontology (0459;
- 3030);
- 4950 Paleoecology;
- 4954 Sea surface temperature;
- 4964 Upwelling (4279)