Dependencies

• Dependencies are listed in 'environments/pyproject.toml' file.

Data

The data was downloaded, preprocessed for the task and stored in .tif files. It is available at Google Drive as archive. Prior to the running a docker container (see next section), the project directory should be organized as follows (tree depth is limited by 4):

.
├── configs
│   ├── eval
│   |    └── eval_config.yaml
│   ├── process
│   |    └── process_config.yaml
│   ├── train
│   |    └── train_config.yaml
│   └── main_config.yaml
├── data
│   ├── 01-raw
│   |   ├── boundary
│   |   ├── Asia
│   |   |   ├── Climate
│   |   |   ├── LC_Type2.tif
│   |   |   └── ELv.tif
│   |   ├── Rice_map
│   |   ├── SEAsia
│   |   |   ├── Climate_future
│   |   |   ├── LC_Type2.tif
│   |   |   └── ELv.tif
│   |   └── FAOSTAT
│   ├── 01-prepared
│   ├── 02-models
│   ├── 03-results
│   └── 04-feed
├── README.md
├── environments
│   ├── Dockerfile
│   ├── poetry.lock
│   └── pyproject.toml
├── notebooks
│   ├── crops_statistics.ipynb
│   ├── notebook_yield.ipynb
│   ├── notebook_esg.ipynb
│   └── rice_mask.ipynb
└── src
    ├── load_data.py
    ├── models
    ├── plotting.py
    ├── rice_mask.py 
    └── utils.py

Pay attention the available years range in that source:

• climate data 'data/Asia/Climate' is available from 1966 to 2021,
• future climate data 'data/SEAsia/Climate_future' is available for years 2027, 2028 with 3 different CMIP5 simulations,
• land cover data is available from 2001 to 2022,

Docker

From repo folder run:

• docker build -t arable:1.0 environments/.
• docker run -it -v <CODE FOLDER>:/app -v <DATA FOLDER>:/app/data -w="/app" arable:1.0
• Inside the container run sh download.sh -- it will download all the necessary data from google drive, unpack it, delete the archived data

Executing program

Scripts in repo folder allow to perform main steps of algorithm. To run them properly, set the required parameters in all files (including subfolders) of 'config folder'.

• preprocess.py performs data processing and saves data in 01-prepared folder
• train.py trains the model
• run.py creates several plots

Alternatively, the same steps are implemented interactevely in notebook_esg.ipynb

Moreover, notebook_yield.ipynb performs modeling of fertilizers consumption and rice yield in countries of Southeastern Asia. More respecifically, it

• loads data
• defines set to countries to perform further analysis
• collects climate statistics for those countries
• builds SARIMAX model to predict fertilizers consumption for those countries
• performs yield modeling with XGBRegressor
• estimates the uncertainty of that model with boostrap procedure

'src' folder contains load_data.py, plotting.py and utils.py modules for loading data, plotting and auxilary procedures.