DeepCarbon: Machine learning-based software tool for characterization of carbon assets and impacts of land-use changes for informed planning and decision making.

During Phase I, We will demonstrate the feasibility of a proposed platform for automated characterization of carbon assets by developing foundational machine learning tools. This will include identification and validation of useable datasets for areas with known biomass density, and development of machine learning models linking multi-modal geospatial data, both on a pixel and regional level, with mappings of above-ground biomass. We will apply these mappings to validated historical data in order to develop predictive models based on long-short term neural networks, capable of forecasting biomass accumulation over time for vegetated areas. These models will be applied to existing predictions of land-use change in a demonstration region, identifying high-priority targets for conservation efforts and planning intervention. Combining historical data and accurate and automated measurement of carbon sequestration capacity with predictions of land use changes will serve to provide invaluable insight into priority areas that represent a high-risk of sequestered carbon loss on a regional scale. Additionally, such tools will enable rapid identification of the carbon cost of local actions impacting land use conversion, allowing examination of these impacts in consideration of triple bottom line.