Data-driven decision making is key to providing effective and efficient wildfire protection and sustainable use of natural resources. Due to the complexity of natural systems, the decision(s) to alter those resources needs clear justification based on substantial amounts of information that are both accurate and precise at various spatial scales. To build that information and incorporate it into the decision-making process, new analytical processes and frameworks are required that incorporate novel computational, spatial, statistical, and machine learning concepts with field data and expert knowledge in a manner that is easily digestible by natural resource managers and practitioners.

UM has successfully built an open source working parallel processing library (Raster Tools) for spatial modeling that can be used to perform various GIS analyses quickly, easily, and at scale. Using this library we have built multiple examples demonstrating how cost revenue streams can be quantified at fine spatial scales across large extents (Spatial Modeling Tutorials). To expand upon the current success of the ongoing investigation, we are now integrating machine learning, computer vision, and spatial processing procedures into the existing Raster Tools software architecture. Additionally, we are expanding our use case scenarios to include functions to quantify wildfire suppression difficulty indices (SDI), potential control locations (PCL), and surface fuels using various data sources including remotely sensed data.

Fig 4. The Raster-Tools software suite provides a uniform interface to acquire geo-spatial data and process it in a highly scalable manner.

Fig 5. The Rice Ridge fire, 07/24/2017. Left panel shows the fire severity determined by MTBS (Eidenshink et al., 2007). Right panel shows the fire severity predicted by our random forest classifier. This work was enabled with the Raster Tools we develop in our lab.