Reconstruction of Surface Seawater pH in the North Pacific

• Published in: Sustainability Vol. 15; no. 7; p. 5796
• Main Authors: Wang, Jie, Yao, Peiling, Liu, Jiaming, Wang, Xun, Mao, Jingjing, Xu, Jiayuan, Wang, Jiarui
• Format: Journal Article
• Language: English
• Published: Basel MDPI AG 01.04.2023
• Subjects: Acidification; Algorithms; Alkalinity; Analysis; Biodiversity; Carbon cycle; Carbon dioxide; Climate change; Neural networks; Ocean acidity; Regression analysis; Salinity; Seawater; Variables; Pacific Ocean; Hawaii; United States > US
• ISSN: 2071-1050; 2071-1050
• DOI: 10.3390/su15075796

In the recent significant rise in atmospheric CO2, seawater’s continuous acidification is altering the marine environment’s chemical structure at an unprecedented rate. Due to its potential socioeconomic impact, this subject attracted significant research interest. This study used traditional linear regression, nonlinear regression random forest, and the BP neural network algorithm to establish a prediction model for surface seawater pH based on data of North Pacific sea surface temperature (SST), salinity (SSS), chlorophyll-a concentration (Chl-a), and pressure of carbon dioxide on the sea surface (pCO2) from 1993 to 2018. According to existing research, three approaches were found to be highly accurate in reconstructing the surface seawater pH of the North Pacific. The highest-performing models were the linear regression model using SSS, Chl-a, and pCO2, the random forest model using SST and pCO2, and the BP neural network model using SST, SSS, Chl-a, and pCO2. The BP neural network model outperformed the linear regression and random forest model when comparing the root mean square error and fitting coefficient of the three best models. In addition, the best BP neural network model had substantially higher seasonal applicability than the best linear regression and the best random forest model, with good fitting effects in all four seasons—spring, summer, autumn, and winter. The process of CO2 exchange at the sea–air interface was the key factor affecting the pH of the surface seawater, which was found to be negatively correlated with pCO2 and SST, and positively correlated with SSS and Chl-a. Using the best BP neural network model to reconstruct the surface seawater pH over the North Pacific, it was found that the pH exhibited significant temporal and spatiotemporal variation characteristics. The surface seawater pH value was greater in the winter than the summer, and the pH decline rate over the past 26 years averaged 0.0013 yr−1, with a general decreasing tendency from the northwest to the southeast. The highest value was observed in the tropical western Pacific, while the lowest value was observed in the eastern equatorial region with upwelling, which is consistent with the findings of previous studies.