HYBRID MODELING OF ECOSYSTEM-LEVEL METABOLISM

Abstract

 Metabolic prediction and management has significant funding to manage change at the ecosystem level because it involves management of the environment.  The proposed research will present a hybrid modeling design where field flux measurements can be coupled with remote sensing indices and machine learning to characterize and support predictions of biogeochemical processes in diverse ecosystems.  We received information on forest, wetland, estuarine and grassland biomes including GPP, NEP, ER and nutrient fluxes.  Non-linear relationships between environmental variables were determined using Random Forest and XGBoost algorithms and mechanistic carbon modeling.  The model outperformed R 2 = 0.85 at every location and the SHAP analysis revealed that the most significant predictors were NDVI, temperature, and soil moisture.  These findings are supported by 9 big tables and 12 high quality images that demonstrate the way by which carbon, and nutrients flow space- and time-wise.  Experts, interviewed, confirmed that the outcomes of the models can be applied in real life practices in the monitoring of the environment and policy formulation.  Combining mechanistic and data-driven approaches with respect to obtaining a complete picture of the metabolism of ecosystems: this work demonstrates just how useful such an integrated method can be. This may be of use when it comes to future climate-resilient management plans.