RAVE RAP FRP July 2019 to 2021 Merged Dataset

The RAP_RAVE_df_2019_2021.csv is a file that contains a point-wise machine learning dataset. It was created by merging data the RAVE fire radiative power dataset, the rapid refresh (RAP) weather model dataset, time of the RAVE detection to the nearest hour, and location information. The RAVE data is aggregated over a 13km resolution grid, with identical structure to the RAP grid, over the hour indicated in "hour." The data is not provided in grid format, however. It is a tabular csv where if there was a non-zero fire radiative power (FRP) from the RAVE dataset, it is listed as an entry andmerged with the corresponding RAP information. The lat and lon provided are the RAP lat and lon points. It is intended to be used to train a machine learning model to forecast hourly FRP

The RF_RAP_RAVE_all_vars.sav file is a trained ranfom forest model created based on the following credentials that can be imported into a python program. The following list of inputs must match and are defined with corresponding rap variable names in the collab notebook. Note that "24hr_roll_frp" corresponds to the 24 hour rolling mean of FRP up to the hour before desired modeling period: 

Training dimensions:   [195707, 26]   

Testing dimensions:  [83875, 26]   

Credentials of input data: Rolling 24 hour FRP mean > 0   

Withheld dates for blind validation  08-01-2019 through 11-01-2019 and 08-25-2020 through 09-25-2020   

Model structure: RandomForestRegressor(n_estimators = 1000, random_state = 30)

Inputs must be: 'lat', 'lon', 'hour', 'temp2m', 'wind_gust', 'precip', 'dpt', 'uwind', 'vwind',
                                 'temp_surf', 'snod', 'spfh', 'pwat', 'weasd', 'dswrf', 'uswrf', 'dlwrf', 'ulwrf', 
                                 'cape', 'ltng', 'tsoil', 'soilw', 'vgtyp', 'sotyp', '24hr_roll_frp'

Output variable will be an hourly FRP value for that input's hour, latitude, and longitude.