Leveraging machine learning and high-performance computing to deliver the spatial data needed by Maine's forest industry
PI: Legaard, Kasey (Forest Resources, UMaine)
Sector: Computer Science, Forestry
Partners: UMS Advanced Computing Group
Abstract: Forest managers in Maine cite a lack of spatial information about forest resources (both timber and non-timber) as a key barrier to the planning and prioritization of management actions. Available commercial products are typically priced at levels that are viewed as too expensive by Maine landowners. More critically, available products suffer from systematic error originating from mapping algorithms or imperfections in reference data available to train mapping algorithms. To address the reliability shortcomings of current data products available to forest industry and forest researchers, we developed a machine learning method that is capable of minimizing both total and systematic error in estimates of forest attributes from satellite imagery. We would specifically like a student to lead the continued effort of producing map output from trained GA-SVM models.