Controlling Factors And Tenative Predictions Of Relative Distribution Of Major Marine Diazotrophic Groups

Marine biological nitrogen fixation provides up to half of the new nitrogen source in the ocean,promotes primary marine production,regulates export production,and affects carbon and nitrogen cycle in the ocean.The main diazotrophic groups in the ocean include Trichodesmium,unicellular cyanobacteria(UCYN),diatom-diazotroph associations.The difference of size and physiological between diazotrophic groups make them occupy different niches.Studying the controlled environmental factors of the relative distribution of major diazotrophic groups and predict their distribution characteristics in the global ocean,which is of great significance to reasonably estimate marine nitrogen fixation levels,study the effects of nitrogen fixation on marine ecosystem structural functions and carbon export,predict the response of marine nitrogen fixation to global climate change.Based on the global observation data of diazotrophic groups nifH gene abundance,this study aims to explore the controlled environmental factors of distribution of diazotrophic groups and to predict the relative abundance of diazotrophic groups in the global ocean.Data analysis and prediction were performed on the relative abundance nifH gene of Trichodesmium and two major unicellular cyanobacteria(UCYN-A and UCYN-B)by using single factor regression,multiple linear regression and machine learning method.Single factor regression screened out controlled environmental factors;Multiple linear regression was used to discriminate the environmental factor set that significantly affect the relative distribution of diazotrophic groups,and to compare the influence degree of environmental factors.The linear regression regression equation is used to predicted the relative distribution of the main diazotrophic groups.In machine learning,we use two supervised learning algorithms random forest regression and support vector machine regression,slecting k-fold Cross Validation and Hold-out Validation methods to divide training and verification set to exploring whether it is possible to effectively predict the relative distribution of different diazotrophic groups based on existing observations.In the single factor regression results,temperature has a greatest influence on the relative distribution of diazotrophic groups,which the R2 is above 0.2,the highest is 0.53.In the multiple linear regression results,The R-squared for the relative abundance of Trichodesmium,UCYN-A and UCYN-B is 0.5,0.5,0.3,respectively.we can see that temperature,light radiation intensity,surface wind speed,dissolved iron and Chlorophyll-a concentration are the major factors affecting the relative distribution of three diazotrophic groups.Nitrate,phosphate and silicate have little effect on the relative distribution of three dizotrophic groups.Combined with the results of single factor regression and multiple linear regression.Trichodesmium is dominant under high temperature,high light radiation,low surface wind speed,high dissolved iron concentration condition;UCYN-A is dominant under low temperature,week light radiation intensity,high surface wind speed,high dissolved iron concentration and high Chlorophyll-a concentration confditions.UCYN-B is less likely to be the major diazotrophic group,which is suitable for sea areas with high temperature,low chlorophyll-a concentration and low dissolved iron concentration conditions.In the machine learning results,support vector machine regression with Hold-out Validation method performs better in prediction of relative distribution of Trichodesmium and UCYN-A,R2 in validation set reaches 0.7 and 0.7,respectively;Random forest regression performs better in the prediction of UCYN-B relative distribution on the validation set of Hold-out Validation,R-squared is 0.5.Machine learning methods can more closely fit the trend of the relative distrnbution of three diazotrphic groups.The Trichodesmium or UCYN-A-dominatedted area is well boundaries.In the equatorial warm currents,the Trichodesmium is dominant;25-40°subtropical ocean circulation center with lower temperature and nutrition concentration,UCYN-A is dominant;UCYN-B is mainly present in relatively low abundance in the low-altitude oligotrophic area,In the Pacific 15°N and 15°S latitudes areas UCYN-B has higher abundance than other areas.From the ocean predictionresults,in the Atlantic warm current region,Trichodesmium is dominant,while UCYN-A is the dominant in the cold current region;UCYN-A is widely distributed in open Pacific Ocean,and it is the main diazotrophic group in the western Pacific;Southern Indian Ocean cyanobacteria is rare.In summary,the single factor can hardly explain the relative distribution of didifferent diazotrophic groups,which have to be considered as the results of the combination of many environmental factors.Machine learning method show a strong potential and make a good prediction in the relative distribution of major diazotrophic groups globally.Due to the limitation of data size limitation,However,further improvement in the prediction of the relative distribution of diazotrophic groups requires more support from observational data.