Flux Tower Measurements Combined With Deep Learning To Estimate Carbon Sequestration

The integration of multiple remote sensing data sources with artificial intelligence (AI) has emerged as a powerful approach for biomass estimation. Machine learning models can now synthesize information from optical imagery, radar, and LiDAR data to produce more accurate estimates of forest parameters. Converting tree parameters to carbon sequestration requires the use of allometric equations, necessitating significant parametrization and integration of local tree parameter estimates. However, change in tree dimensions acquired by manual surveys or LiDAR measurements remain challenging due to measurement uncertainties and sensors noise. Flux towers measurements are offering a direct way to estimate carbon sequestered by vegetation or soil and these changes can be compared with repeated LiDAR acquisitions over the same area. Ground validation networks, such as the NEON or AmeriFlux FLUXNET, provide crucial temporal data on exchanges of carbon and biomass accumulation across many ecoregions. We compare the impact of land use, land cover and ecoregions in estimating the growth of carbon pool and verify the impact of aerial LiDAR measurements and use of allometric equations. Similarly, airborne LiDAR campaigns have demonstrated capability to track forest growth and degradation over time with high precision. These measurements serve as essential calibration and validation data for satellite-based approaches to above ground-biomass monitoring over time.

In this work, we present our work to compare the estimates from flux towers, LiDAR and deep learning models to quantify the uncertainty in carbon sequestration and the yearly growth in carbon pool.