Reconstructing chlorophyll-a (Chl-a) vertical profile is a promising approach for investigating the internal structure of marine ecosystem. Given that the process of profile classification in current process-oriented profile inversion methods are either too subjective or too complex, a novel Chl-a profile reconstruction method was proposed incorporating both a novel binary tree profile classification model and a profile inversion model in the Mediterranean Sea. The binary tree profile classification model was established based on a priori knowledge provided by clustering Chl-a profiles measured by BGC-Argo floats performed by the profile classification model (PCM), an advanced unsupervised machine learning clustering method. The profile inversion model contains the relationships between the shape-dependent parameters of the nonuniform Chl-a profile and the corresponding Chl-a surface concentration derived from satellite observations. According to quantitative evaluation, the proposed profile classification model reached an overall accuracy of 89%, and the mean absolute percent deviation (MAPD) of the proposed profile inversion model ranged from 12%–37% under different shape-dependent parameters. By generating monthly three dimensions Chl-a concentration from 2011 to 2018, the proposed process-oriented method exhibits great application potential in investigating the spatial and temporal characteristics of Chl-a profiles and even the water column total biomass throughout the Mediterranean Sea.

重建叶绿素-a (Chl-a) 垂直剖面是研究海洋生态系统内部结构的一种很有前景的方法。鉴于当前面向过程的剖面反演方法中剖面分类过程过于主观或过于复杂，提出了一种新的Chl-a剖面重建方法，该方法结合了新的二叉树剖面分类模型和地中海剖面反演模型海。基于二叉树轮廓分类模型的建立先验由轮廓分类模型 (PCM) 执行的由 BGC-Argo 测量的 Chl-a 轮廓聚类提供的知识，这是一种高级无监督机器学习聚类方法。剖面反演模型包含非均匀 Chl-a 剖面的形状相关参数与卫星观测得出的相应 Chl-a 表面浓度之间的关系。根据定量评价，提出的剖面分类模型的总体准确率达到了89%，在不同形状相关参数下，提出的剖面反演模型的平均绝对百分比偏差（MAPD）在12%~37%之间。通过生成从 2011 年到 2018 年的月度三个维度 Chl-a 浓度，