Mapping Co2 Fluxes Of Cypress Swamp And Marshes In The Greater Everglades Using Eddy Covariance Measurements And Landsat Data

Wetlands play a critical role in the global carbon (C) budget by sequestering carbon dioxide (CO2) and storing significant amounts of C, but climate change is altering C dynamics in these ecosystems. Therefore, predicting and mapping net ecosystem exchange (NEE) of CO2 in wetlands has emerged as a focus of research to better understand where, how, and when biogeochemical cycles will be modified by human and natural disturbances like climate change. To address this, we developed an object-based machine learning ensemble approach to upscale eddy covariance (EC) CO2 flux measurements across two major wetland ecosystems (cypress swamp and freshwater marshes) in the Greater Everglades of south Florida. We linked 30-m Landsat, water depth, and air temperature data with EC flux measurements using the temporally composite concept for model development, and mapped fluxes within Big Cypress National Preserve (BCNP) and Everglades National Park (ENP) to characterize two distinct seasonal flux patterns. Using an ensemble analysis of three machine learning model outputs (Artificial Neural Network, Support Vector Machine, and Random Forest), our models of NEE at BCNP had a high predictive power with a R-2 greater than 0.8, while NEE was predicted at ENP with a R-2 greater than 0.5. We also generated an uncertainty map to quantify the prediction diversity caused by application of multiple models in upscaling. This uncertainty map identifies regions that are either easy or challenging to predict from model based analyses. We conclude that integration of Landsat, water depth, and air temperature data is valuable for quantifying CO2 exchange between the atmosphere and Everglades ecosystems, and the developed paradigm is promising for upscaling EC flux measurements in wetlands to study and mitigate disturbances to wetland ecosystem services such as hurricanes, fires, water management and/or climate change.