基于时空演变多重特性建模的近海叶绿素浓度时序预测

近海环境是沿海地区社会经济发展的关键支撑系统，近海环境的持续恶化对海洋经济的可持续发展带来了巨大挑战。叶绿素浓度的反映了水体理化性质的演变规律，对近海生态环境保护具有重要意义。尽管现有时序叶绿素浓度预测方法能从时空数据中挖掘有效信息，揭示时空数据的发展趋势和变化规律，但忽略了时空数据的结构化特征以及外界因素/突发因素对叶绿素浓度的影响。因此，本文提出基于时空演变多重特性建模的近海叶绿素浓度时序预测模型，并由四部分构成:自相关时序预测模块预测叶绿素浓度时序变化规律；多视角空间融合预测模块在构建预测点与其他位置叶绿素浓度空间关联性基础上，考虑海域气象状况，提高了空间叶绿素浓度预测的可靠性；基于环境上下文的突变模块通过对极端因素建模，挖掘突变因素与的叶绿素浓度变化的关联；时空动态聚合模块利用结构化模式，结合时间、空间叶绿素预测结果，实现不同圈层全要素近海叶绿素浓度建模。在渤海叶绿素浓度数据上的实验结果表明，该算法模型极大程度提升了近海叶绿素预测模型的准确性与可靠性。 

Temporal Prediction of Chlorophyll Concentration in Coastal Waters Based on Multi-characteristics Modeling of Spatio-temporal Evolution

? ?Chlorophyll plays an important role in the study of marine ecosystem. However， chlorophyll concentration is affected by many coupling factors， so it is a difficult problem to predict chlorophyll concentration accurately. These factors include not only the prediction of the site’s own attributes， but also the range of other site attributes that can have an impact on the chlorophyll concentration of the prediction site. At present， the time series prediction of chlorophyll concentration is realized by only considering the self-factor of prediction point， and the effect of space factor on chlorophyll concentration is ignored. Moreover， the combination of flat factors lacks structure， and it is difficult to achieve reasonable feature expression. In this paper， a data-driven spatio-temporal aggregation model is proposed to predict chlorophyll concentration. In this model， autocorrelation time series prediction model and multi-view spatial fusion prediction model are dynamically combined， and the influence of sudden change of environmental factors is considered. This paper evaluates the models using 2017 data from the Bohai Sea Region， and the results are superior to several baseline methods.