The Analytical Study On Ecosystem Of The Beibu Gulf Based On Physical-Ecological Model And Machine Learning

The Beibu Gulf is an important semi-enclosed gulf located in south-western China with abundant natural resources,and it is an emerging key strategic marine planning area in the coastal areas of China.Therefore,the ecological condition of the Beibu Gulf is related to the ecological security pattern and the environmental development in the southwestern China and the South China Sea(SCS)coast.With the rapid development of the Beibu Gulf Economic Zone in recent years,the gulf is facing environmental and resource issues.Therefore,it is essential to study the process of the Beibu Gulf ecosystem.In this study,a physical-ecological coupling model was established to simulate the temporal and spatial distribution characteristics of physical and ecological variables,and the accuracy of the model has been verified against remote-sensing and field investigation data.Taking the simulated output for the year of 2015 as the example,the distribution of physical and ecological variables in representative months,depths,and sections were discussed.The results indicate that the interface process has an important impact on the ecosystem of the Beibu Gulf.The concentrations of organic matters,inorganic nutrients and plankton are generally higher in the eastern and northern gulf than in the western and southern gulf,higher in nearshores than in offshores,and higher in the upper layer than in the lower layer;most of the northern Beibu Gulf is a phosphorus-restricted area;the estuary of the Red River in Vietnam,the southwestern coast of the gulf,and the areas affected by the intrusion of the SCS in the southeastern gulf in spring and summer are nitrogen-restricted areas.The contribution of physical and biochemical processes to the nutrient budget in the gulf was analyzed,the results show that the phytoplankton photosynthetic uptake is the nutrient sink,and phytoplankton dead cellular release is the largest source of DIN,while phytoplankton respiration is the largest source of DIP.In addition,the biochemical processes have greater impact on the nutrient budget than physical processes.The results from the analysis of the nutrient flux across the gulf mouths show that in the Qiongzhou Strait,the nutrient transported from outer sea to the Beibu Gulf throughout 2015;on the latitudinal section at the southern gulf mouth,the nutrient is transported from the Beibu Gulf to outer sea;on the longitudinal section,nutrients are transported from the outer sea to the Beibu Gulf.The nutrient inflows from outer sea to the Beibu Gulf are larger in autumn and winter than in spring and summer.In summary,there are 113709 t of DIN and 5277 t of DIP transported from outer sea to the Beibu Gulf in 2015.Based on the simulated data from 2007 to 2015,it can be found that the ecological variables at the stations in the northern gulf affected by monsoon or river input have more significant seasonal fluctuations,and the stations in the central and southern gulf have less seasonal fluctuations.The nutrients across the Qiongzhou Strait section are transported from east to west,from the outer sea into the Beibu Gulf in general.Only in July of 2007 to 2011 there are transportations from west to east.The direction of nutrient transport across the latitudinal section on the southern gulf mouth is from north to south,while across the longitudinal section on the southern gulf mouth is from east to west.The La Ni?a incident occurred from 2011 summer to 2012 spring has greater impacts on the ecological variables in the northern gulf.It results in the decrease of sea temperature and the enhancement the vertical mixing of the water,thus promoting various physical and biochemical processes and water transport.Based on the different variables of from model simulated results,the spatially constrained multivariate clustering(SCMC)method of unsupervised machine learning was used to make the geographical-environmental regionalization for the Beibu Gulf.In most experiments,an individual region can be identified in west of the Qiongzhou Strait,indicating that this area is an important ecological zone of the Beibu Gulf.This region represents the impact of the Qiongzhou Strait current invading the Beibu Gulf.The estuary along the Guangxi coast can be identified as an individual region in the summer at the surface layer,which represents the influence range of the river input.There is usually a zonal boundary near 20°N,which represents the specific ecological boundary of the southern and northern part of the gulf.The progressive characteristics of the regionalization in the southern gulf reflect the water intrusion and ecological impact of SCS outer water in different time and space.In addition,the regionaliztation for ENSO(El Ni?o-Southern Oscillation)years also reflects the ecological boundaries in the Beibu Gulf under different climatic conditions.Based on the model simulated result which has a big data feature,backpropagation neural network(BPNN)and random forest(RF)of supervised machine learning were applied to make multi-to-single factor prediction for the horizontal distributions.The results show the correlation of the final prediction of both two methods are above 0.9,and the accuracy of RF is better than BPNN.Recurrent neural network(RNN)was used for multi-to-single factor prediction in time series while long short term memory(LSTM)was used to predict single-to-subsequent single factor in time series.The predicted trend is consistent with the original data in general.In summary,this study has determined the key regions and key periods affecting the Beibu Gulf ecosystem,and solved the channel effect issues through quantitative analysis,thus makes up for the insufficiency of the field investigation.In addition,it has also proved the ability to use machine learning to extract the features of model simulated data.This study can provide a scientific basis for further scientific research and management of natural resources in the Beibu Gulf,as well as the application of ocean big data.