This dataset is a weekly time-series of images from 2005-2017 that depict the probability of large fire across western US forests and woodlands. Specifically, the images depict the probability that an area on the landscape will burn in a large (i.e., > 405 ha) fire following an ignition event, on the given date. Each product in the dataset is a three-band GeoTIFF image (250-m resolution) in the WGS84 geographic coordinate reference system (EPSG:4326). Names of each product correspond to the image prediction date. Band 1 values are the output of a Random Forest classification algorithm, trained on 10 independent, random samples of small and large wildfires that occurred from 2005-2014, and represent the mean predicted probability of an individual pixel burning in a large fire. Band 1 values range from 0-100 (probability scaled by 100). Band 2 values represent the standard deviation of predicted probability of an individual pixel burning in a large fire, and values also range from 0-100 (standard deviation scaled by 100). Band 3 values indicate the quality of MODIS predictor variables. Multiple MODIS products were used as predictor variables to describe the vegetation and land surface immediately preceding a fire event. Only good quality pixels were retained for model training, but all pixels were retained when creating spatial predictions. Therefore, Band 3 indicates if one of these MODIS predictors had unreliable quality, where0 = All MODIS pixels were processed and good quality and 1 = At least one MODIS pixel was not processed or had bad quality. No Data values in each image are set to 255.
|Date made available||2018|