Spatial reconstruction of long-term (2003–2020) sea surface pCO2 in the South China Sea using a machine learning based regression method aided by empirical orthogonal function analysis,Earth System Science Data

当前位置： X-MOL 学术 › Earth Syst. Sci. Data › 论文详情

Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)

Spatial reconstruction of long-term (2003–2020) sea surface pCO2 in the South China Sea using a machine learning based regression method aided by empirical orthogonal function analysis
Earth System Science Data ( IF 11.4 ) Pub Date : 2022-10-04 , DOI: 10.5194/essd-2022-322
Zhixuan Wang , Guizhi Wang , Xianghui Guo , Yan Bai , Yi Xu , Minhan Dai

Abstract. The South China Sea (SCS), the largest marginal sea of the North Pacific Ocean, is one of the world’s most studied model ocean margins in terms of its carbon cycle, where intensive field observations including sea-surface carbon dioxide partial pressure (pCO₂) have been conducted over the last two decades. However, the datasets of cruise-based sea surface pCO₂ are still temporally and spatially incomplete. Using a machine learning-based method facilitated by empirical orthogonal function (EOF) analysis capable of constraining the spatiality, this study provides a reconstructed dataset of the monthly sea surface pCO₂ in the SCS with a reasonably high spatial resolution (0.05º×0.05º) and temporal coverage between 2003 and 2020. We validate our reconstruction with three independent testing datasets where, TEST.1 includes 10 % of our observed data, TEST.2 includes four independent underway datasets corresponding to four seasons, and TEST.3 includes a continuous observed dataset from 2003–2019 at the South East Asia Time-Series (SEATs) station located in the northern basin of the SCS. Our TEST.1validation demonstrated that the reconstructed pCO₂ field successfully simulated the spatial and temporal patterns of sea surface pCO₂. The root-mean-square error (RMSE) between our reconstructions and observed data in TEST.1 averaged to ~10 μatm, which is much smaller (by ~50 %) than that between the remote sensing (RS) and observed data. TEST.2 verified the accuracy of our reconstruction model in data months lacking observations, showing a near-zero bias (RMSE: ~8 μatm). TEST.3 tested the accuracy of the reconstructed long-term trend, showing that at the SEATs Station, the difference between the reconstructed pCO₂ and observations ranged from -10 to 4 μatm (-2.5 to 1 %). In addition to the typical machine learning performance metrics, we present a new method to assess the uncertainty that includes the bias from the reconstruction and its sensitivity to the features, and successfully quantifies the spatial distribution patterns of uncertainty. These validations and uncertainty analysis strongly suggest that our reconstruction is effectively captures the main features of both the spatial and temporal patterns of sea surface pCO₂ in the SCS. Using the reconstructed dataset, we show the long-term trends of sea surface pCO₂ in 5 sub-regions of the SCS with differing physico-biogeochemical characteristics. We show that mesoscale processes such as the Pearl River plume and China Coastal Currents significantly impact sea surface pCO₂ in the SCS during different seasons. While the SCS is overall a weak source of atmospheric CO₂, the northern SCS acts as a sink, showing a trend of increasing strength over the past two decades.

更新日期：2022-10-04

点击分享查看原文

点击收藏

公开下载

阅读更多本刊最新论文

全部期刊列表>>