Inversion And Spatiotemporal Distribution Characteristics Of Surface Seawater PH In The North Pacific

With the rapid increase of atmospheric CO₂,the continuous acidification of seawater has disrupted the balance of the marine ecosystem and caused irreversible changes in the marine ecosystem,thus affecting the development of human society.Ocean acidification has become the third major environmental problem after global warming and environmental pollution.The pH value of seawater can directly reflect the degree of seawater's acidification,so real-time monitoring of seawater pH is urgent.At present,seawater pH monitoring data is scarce,and it is impossible to establish the spatial distribution of long-term and long-term series.It is an effective method to use readily available data to model surface seawater pH inversion.The North Pacific plays a vital role in global climate change.So it is crucial to inversely model the pH of the surface water of the North Pacific.In addition to the increase of atmospheric CO₂ concentration are the most critical factor leading to ocean acidification,seawater temperature,salinity,dissolved oxygen,partial pressure of carbon dioxide,total alkalinity,and chlorophyll-a concentration also affect the change of seawater pH value.This paper is based on the measured surface water parameters(voyage data and station data),such as sea surface temperature(SST),sea surface salinity(SSS),partial pressure of carbon dioxide(PCO2),dissolved oxygen(DO)and total alkalinity(TA),with a large amount of data.Linear regression modeling and BP neural network modeling methods were used to model the pH values of surface water in the North Pacific offshore waters(Yellow Sea,East China Sea,South China Sea,Sea of Japan,East coast of North Pacific),the North Pacific Ocean and the whole North Pacific Ocean area.The pH value of the surface water of the North Pacific Ocean was estimated by using the established optimal model,so as to analyze the temporal and spatial distribution characteristics of the pH value of the surface water of the North Pacific Ocean from 1999 to 2018.The conclusions are as follows:1)The R² of the model of the east coast of the North Pacific Ocean,the North Pacific Ocean,and the whole North Pacific Ocean are all more than 0.9,because the data used in the inversion modeling of the three sea areas are site data,which has strong stability and reliability.The best linear regression model for the east coast of the North Pacific Ocean is pH=8.7970-0.0068*SST-0.0069*SSS-0.0010*PCO₂.This model is the best inversion model among all the models,with not only the determination coefficient R2 as high as 0.9810,but also a small error in the four-season verification.The second is the BP neural network model of pH and SST-SSS-PCO₂ in the whole North Pacific Ocean,with a determination coefficient of 0.9768 and a small error in the four-season verification.The best model of the North Pacific Ocean is the PH-SSS-PCO₂ BP neural network model,with a determination coefficient of 0.9743,but its seasonal applicability is poor.2)The fitting degree of the model inversed by the measured voyage data is worse than that inversed by the measured station data.The optimal inversion model for the Yellow Sea is pH=9.2338-0.0112*SST-0.0286*SSS-0.0206*DO,with a determination coefficient of R² of 0.8755.It has high applicability in spring,autumn and winter.The best model of the East Sea is BP neural network model PH-SST-SSS-DO model,whose determination coefficient is 0.8711,but it only applies to summer and autumn.The modeling result of the South China Sea is the worst among all the models,and the coefficient of determination of the best model is 0.7060 when pH=8.3857-0.0116*SSS+0.0520*DO-0.0264*TA.The best inversion model for the Sea of Japan is pH=7.9459+0.0058*SST-0.0110*SSS+0.0767*DO,and the coefficient of determination R²is 0.8451,which is suitable for spring,summer,and winter.3)Two modeling methods are used in this paper:linear regression modeling and BP neural network modeling.In BP neural network modeling,each modeling result is different,while linear regression modeling will only have one modeling result,compared with linear regression modeling is more convenient;From the modeling process in this paper,it can be seen that the goodness of fit of BP neural network is higher than that of linear regression in most cases.However,in the verification of seasonal applicability,linear regression modeling is more suitable for multiple seasons.This is because of BP's over-fitting phenomenon.Although its training ability is high,its prediction ability is decreased.4)The optimal BP neural network model(PH-SST-SSS-PCO₂)of the whole North Pacific Ocean was used to estimate the pH value of the surface waters of the North Pacific Ocean from 1999 to 2018.From the perspective of spatial distribution,the pH value of the surface water of the North Pacific Ocean gradually decreases from northwest to southeast,and the coastal water is seriously acidified.From the perspective of inter-annual variation,the pH value of the surface water of the North Pacific Ocean has been decreasing continuously,and the pH value has reduces by about 0.03 in 20years.From the perspective of seasonal variation,the pH value of surface seawater in summer is lower than that in spring,autumn and winter.The acidification of seawater in autumn is the most serious.