Research On Approaches To Forecasting The Temporal And Spatial Distribution Of Chlorophyll-a Concentration In Coastal Waters

The concentration of chlorophyll-a(Chl-a)in coastal waters is an important indicator to measure the primary productivity of the ocean,and it is also a key indicator to measure the water eutrophication.The evolution rules of Chl-a concentration reflect the dynamic change in the physical and chemical properties of sea water.Therefore,the prediction of Chl-a concentration is of great significance to discover the evolution of physical and chemical properties in coastal waters,prevent eutrophication,protect the ocean ecological environment,and guarantee the development of coastal marine agriculture.At present,the related works paid more attention to the correlation analysis between the Chl-a concentration and water environment.Researches on prediction of Chl-a concentration were mainly oriented to the non-sea areas,such as rivers or lakes,and lack methods of Chl-a concentration prediction for coastal waters.This thesis takes the coastal environment monitoring data set as the research object.Based on integrated learning method,we analysis and study the comprehensive influence of the coastal environmental factors,such as spatial-temporal characteristics,physical and chemical characteristics and hydrological and meteorological characteristics of the coastal sea environmental monitoring data,as well as the artificial zoning of the sea area and land-based emissions,to the concentration of Chl-a,and realizes the prediction of the spatial and temporal distribution of Chl-a concentration of in coastal waters.The main works of this thesis are as follows.First,we give a set of pre-processing methods for marine environment monitoring data,which includes three steps of raw data pre-processing,feature construction and feature processing.Among them,raw data pre-processing includes such steps as data deduplication,down sampling,interpolation,and normalization;feature construction includes feature simplification and feature combination;feature processing includes feature reorganization and feature importance analysis.Second,we present a cluster-regressive stacking method to predicting the Chl-a concentration in coastal waters.We first apply K-means clustering to group the monitoring data,and the study region is thus divided into 7 types of sub-regions according to the clustering results.Each type of the sub-regions has similar evolution characteristics of Chl-a concentration;then in each sub-region,a stacked regression model based on KNN,MLP,SVR,XGBoost and RF is constructed to realize seasonal prediction of Chl-a concentration.Third we propose a method of Chl-a concentration prediction based on KNN-LSTM.The Euclidean distance and KNN algorithm are applied to select the monitoring stations that are most similar and related to the current monitoring station to form a prediction area.Then,we construct time series composed of the environmental factors of these monitoring stations as training data,apply a LSTM model to realize the spatial and temporal prediction of the Chl-a concentration in a specific local coastal water.In this paper,the environmental monitoring data set of Weihai coastal waters from 2015 to 2018 is used as the experimental dataset,and the experiments result verified the effectiveness of the proposed models and methods to predicting the spatial and temporal distribution of Chl-a concentration in coastal waters.