Forecast model to predict future fire spatial distribution (fire spread) and intensity.
| Namelist Option | Namelist Description |
|---|---|
| frp_input | Filename of initial fire map |
| model_input | Filename of forecast model input |
| model_output | Filename of model output |
| frp_source | Option for initial FRP data (0: RAVE) |
| time_start | Start/initial time stamp in YYYYMMDDHHMM |
| time_end | End time stamp in YYYYMMDDHHMM |
| time_freq | Integer time steps in hours (default = 1) |
| lat_lower_left | Latitue in degree of the lower left corner of domain of interest |
| lat_upper_right | Latitue in degree of the upper right corner of domain of interest |
| lon_lower_left | Longitude in degree of the lower left corner of domain of interest |
| lon_upper_right | Longitude in degree of the upper right corner of domain of interest |
| opt_frpgen | Option for gridded frp reprocessor (0: off, 1: on) |
| opt_inputgen | Option for model input generator (0: off, 1: on) |
| opt_forecast | Option for main forecast model (0: off, 1: on) |
| opt_mapgen | Option for fire mapper (0: off, 1: on) |
| opt_corr | Option for intensity correction model (0: off, 1: on) |
| scale_opt | Option for final FRP scaling (0: off, 1: on) |
| scale_val | Scale factor for final FRP scaling, only available when scale_opt = 1 |
| path_frp | Local location of inital FRP data |
| path_elv | Local location of elevation data |
| path_ast | Local location of fuel/surface type data |
| path_fh | Local location of canopy/forest height data |
| path_vhi | Local location of vegetation health index data |
| path_mete | Local location of metelogical model data |
python src/fire_model.py
Computational resource for a 24h cycle: 12 nodes, 8G mem, 1 hour
./build/compile.sh
./fire_model
Spatial distribution and intensity of individual fires and gridded fire map for domain of interest.
-
Gridded FRP preprocessor (optional): Create gridded FRP file as initial fire map. The NOAA Regional Hourly ABI and VIIRS Emissions (RAVE) products are used as example.
-
Input generator: Extract required variables from provided datasets, select individual fires from domain of interest, pre-processing, and create model input file.
-
Main forecast model: A Convolutional autoencoder model that creates prediction of the spatial distribution of future FRP. Post-processing performed.
-
Intensity correction model: A multiple linear regression model that corrects the FRP prediction generated by main forecast model.
-
Fire mapper: Map individual fires to gridded domain and create predicted fire map as final product.