Noah_DL_Hybrid_parameterization

Optimizing Thermal and Hydrological Processing Simulation on the Qinghai-Tibet Plateau by Integrating Deep Learning and Land Surface Model.

Outline

1. Project Structure
2. Dataset Model

Project Structure

project-root/
│  README.md
│  
├─Parameter_Generator
│  │  LSMTransformer.py
│  │  PGDataset.py
│  │  PGNetwork.py
│  │  PG_TRAINING_UWC_WANDB.ipynb
│  │  
│  ├─GENERATOR
│  │      README.md
│  │      
│  └─PG_DATASET
│      │  static_propertise.csv
│      │  
│      ├─CMFD2FLOAT
│      │      README.md
│      │      
│      ├─GRID_NPY
│      │      README.md
│      │      
│      └─GRID_NPY_QQ
│              README.md
│              
└─Transformer_TEST
    │  LSMDataset.py
    │  LSMLoss.py
    │  LSMTransformer.py
    │  SURROGATE_TRAINING_STC_WANDB.ipynb
    │  SURROGATE_TRAINING_UWC_WANDB.ipynb
    │  
    ├─SURROGATE
    │      README.md
    │      
    └─TEMP
            README.md

• Parameter_Generator: the define of model and dataset and training code for parameter generator are in this directory.
  • LSMTransformer.py: self-defined model of surrogate
  • PGDataset.py: self-defined dataset to load cmfd, static propertice and SMCI(GROUND TRUTH) of study area
  • PGNetwork.py: self-defined model of parameter generator
  • PG_TRAINING_UWC_WANDB.ipynb: a training sample using wandb to train the parameter generator tested by UWC
  • GENERATOR: save the well-trained model in this directory.
  • PG_DATASET: the dataset of input and ground truth in this directory.
• Transformer_Test: the define of model and dataset and training code for surrogate are in this directory.
  • LSMDataset.py: self-defined dataset to load cmfd and simulation of Noah as ground truth of study area
  • LSMLoss.py: self-defined loss to add physics constraint
  • LSMTransformer.py: self-defined model of surrogate
  • SURROGATE_TRAINING_STC_WANDB.ipynb: a training sample using wandb to train the surrogate tested by STC
  • SURROGATE_TRAINING_UWC_WANDB.ipynb: a training sample using wandb to train the surrogate tested by UWC
  • SURROGATE: save the well-trained model in this directory.
  • TEMP: the datatset of input and ground truth in this directory.

Dataset Model

For parameter generator

• Input data: static propertise and vegetation coverage in 2015
• ground truth: SMCI dataset in 2015 with resolution of 10 km
• input for surrogate: CMFD dataset in 2015 with resolution of 0.1 degree

For surrogate

• Input data: CMFD dataset during 2010 and 2014 with resolution of 0.1 degree
• ground truth: simulation of Noah feeding CMFD during 2010 and 2014 with resolution of 0.1 degree