Global data-driven prediction of fire activity

All files within this directory are used to generate the model/data/plots used in the submissions. Software is available at code ocean
https://codeocean.com/capsule/8570224/tree

For the generation of the dataframes used for training the various data-driven models we use (Input data not present in this folder are available from ERA5-Land) :

#### MAKE_PICKLE_*.py

This generates the following pickle files:

#### TRAIN_*_10PER_2019_2021.pk1

Once those pickle files are generated the XGBoost models are then trained using :

#### TR_*.py

This generates the model files, which are then used with the input data to generate ERA5 and forecast data :

#### MAKE_GLOBAL_MAPS_ERA5.py

The output modelled data are then available as :

#### POF_*_ERA5.nc

Computation of radar plot scores are made using :

#### SCORES_FULL.py

Output of statistical analysis are availble as:

#### correlation_fwi.pkl
#### obs_correlation_fwi.pkl
#### *RELIABILITY.pkl
#### *RADAR.pkl

Region specific observations and forecasts are provided in :

#### NEN*
#### NWN*

For plotting the plots are made using the following scripts :

#### FIG_*.ipynb

For access to data note present in this capsule please access the ftp site:

ftp server: ftp.ecmwf.int
username: ecmwf_fire
password: FhXekWMuy