Crepes_veges data: Solving calibration and reanalysis challenges of ocean BGC dynamics with neural schemes: a 1D NNPZD case-study

Repository for the paper 'Solving calibration and reanalysis challenges of ocean BGC dynamics with neural schemes: a 1D NNPZD case-study.' Description: This study is codenamed 'VEGES - Vertical Experiment Gauging Estimation Strategy'. It is part of the PhD project "CREPES - Carbon REconstructed Per an Emulator through Supervision" (Carbone REconstruit Par Emulateur Supervisé). This dataset contains 7 different files: Calibration_DA.py to calibrate the BGC model using a 4Dvar-based scheme. Calibration_NN.py to calibrate the BGC model using a UNet-based scheme. environment.yaml to install the suitable python environment with correct package version. func_file.py that contains the needed functions. Generator_data.py to generate all the necessary data sets. model_file.py that contains the UNet model. Notebook_analysis_plot.ipynb to plot the paper results. The Data sets used in the article are compressed in "Generated_Datasets_part1.zip" and "Generated_Datasets_part2.zip". For their correct use, you can extract them and place them in a single "Generated_Datasets" folder.The results used in the article are compressed in "Res.zip".The Data are generated using a set of forcing profiles from polgyr, stored in "FORCING_40km.zip". 1. Install the correct packages with their associated version with the environment.yaml file.2. Generate the different data sets: run Dataset_Generator.py3. Use freely the different methods (run Calibration_DA.py, Calibration_NN.py)本代码仓库对应论文《基于神经方法解决海洋生物地球化学动力学的校准与再分析难题：一维NNPZD案例研究》。 本研究代号为「VEGES——垂直实验评估估算策略」，隶属于博士研究项目「CREPES——基于监督模拟器的碳重建（原英文全称为Carbon REconstructed Per an Emulator through Supervision，法语全称为Carbone REconstruit Par Emulateur Supervisé）」。 本数据集包含7个不同文件： 1. Calibration_DA.py：用于基于四维变分（4Dvar）方法校准生物地球化学（BGC）模型的脚本； 2. Calibration_NN.py：用于基于UNet架构校准生物地球化学模型的脚本； 3. environment.yaml：用于安装适配版本Python依赖环境的配置文件； 4. func_file.py：包含研究所需全部工具函数的脚本文件； 5. Generator_data.py：用于生成全部必要数据集的脚本文件； 6. model_file.py：包含UNet模型定义的脚本文件； 7. Notebook_analysis_plot.ipynb：用于复现论文实验结果图表的Jupyter Notebook文档。 本文所用数据集已分别压缩为「Generated_Datasets_part1.zip」与「Generated_Datasets_part2.zip」，如需正确使用，请将两个压缩包的解压文件合并至名为「Generated_Datasets」的文件夹中。本文所用实验结果已压缩为「Res.zip」。本研究使用的强迫剖面数据源自polgyr工具，存储于「FORCING_40km.zip」。 使用流程如下： 1. 通过environment.yaml文件安装对应版本的Python依赖包； 2. 运行Dataset_Generator.py生成各类所需数据集； 3. 可自由使用各类校准方法：运行Calibration_DA.py或Calibration_NN.py即可。