The Unet3+ model, a deep learning approach originally introduced for use in medical imaging, is particularly skillful at image segmentation tasks. By combining full scale skip connections with an encoder-decoder architecture, the model can capture both fine-grain detail and coarse-grain semantics simultaneously allowing for more accurate image segmentation. By combining the capabilities of the Unet3+ model with a neighborhood loss function, Fractions Skill Score (FSS), we can quantify model success by predictions made both in and around the location of the original fire occurrence label.
The model is trained on fuel, weather, and topography observational inputs and on weather forecast inputs. The model is trained using labels representing fire occurrence. We source our observational fuel and topography data from the Landscape Fire and Resource Management Planning Tools of LANDFIRE (LF), a product of collaboration between the wildland fire management programs of the U.S. Department of Agriculture Forest Service and U.S. Department of the Interior, which includes topographical products for aspect, elevation, and slope. From NOAA’s Global Ensemble Forecast System (GEFS) Reforecast dataset, a daily, 5-member ensemble numerical weather prediction model, used to produce retrospective gridded meteorological forecasts for CONUS, we source both analyzed and forecast weather data. Our fire occurrence labels are sourced from the U.S. Department of Agriculture’s Fire Program Analysis fire-occurrence database (FPA-FOD), which contains spatial wildfire occurrence data for CONUS, combining data sourced from the reporting systems of federal, state, and local organizations.
Our research focuses on exploring the additional performance offered by incorporating forecast weather data into short-range fire occurrence predictions. We will also explore the impact on model performance offered by datasets containing less correlation, as our previous work identified large amounts of correlation within the observed weather and fuel variables as sourced from gridMET, a daily, CONUS-wide, high-spatial resolution dataset of surface meteorology variables including the weather-derived fuel variables. Lastly, where permitted by the lack of multicollinearity, we seek to quantify the individual contributions of variables or classes of variables to the fire occurrence prediction model.

