Abstract
An important step in understanding the climate system is simulating and studying the past climate variability, using oceanic models, atmospheric models, or both. Toward this goal, long-term wind stress data, as the forcing of oceanic or climate models, are often required. In this study, the possibility of reconstructing the past winds of the tropical Pacific Ocean using historical sea surface temperature (SST) and sea level pressure (SLP) datasets was explored. Four statistical models, based on principal component (PC) regression and singular vector decomposition (SVD), were developed for reconstructing monthly pseudo wind stress over the tropical Pacific for the period 1875–1947. The high-frequency noise was removed from the raw data prior to the reconstruction. These models are SST-based PC regression (model 1), SLP-based PC regression (model 2), SST-based SVD (model 3), and SLP-based SVD (model 4). The results show that reconstructed wind stresses from all models can account for more than one-half of the total variances. In general, the SLP is better than SST as a predictor and the SVD method is superior to the PC regression. Forced by these reconstructed wind stresses, an oceanic general circulation model can simulate realistic interannual variability of the tropical Pacific SST. However, the wind stress reconstructed by SST-based models leads to better simulation skill in comparison with that from SLP-based models. Last, a long-term wind stress dataset was constructed for the period from 1875 to 1947 by the SST-based SVD model, which provides a useful tool for studying the past climate variability over the tropical Pacific, especially for El Niño–Southern Oscillation.