Skip to main content
2023 ESA Annual Meeting Main banner
Icon Legend
This session is not in your schedule.
This session is in your schedule. Click again to remove it.
Back

Contributed Talk

Session: : Back and Forecasting in Ecology

COS 72-5 - Landscape-scale iterative forecasting of terrestrial carbon pools and fluxes : a Harvard Forest testbed

Tuesday, August 8, 2023
11:00 AM – 11:15 AM PDT
Location: B117
Abstract: Our ability to monitor and predict the terrestrial carbon cycle is critical at scales from local to global for responding to monitoring, reporting, and verification (MRV) requirements (e.g., carbon markets, UN Framework), climate change adaptation and mitigation, and understanding source/sink variability. Near-term iterative forecasting has emerged as a powerful tool to meet these needs. Here we report on work extending a continental network of tower-scale carbon forecasts to now account for landscape-scale heterogeneity, using Harvard Forest in central Massachusetts as an initial testbed. This testbed focused on the 5x5km footprint of the National Ecological Observatory (NEON) AOP footprint and was initialized using a combination of NEON AOP leaf area, Landtrendr biomass, and NLCD land cover. The system was used to hindcast 2012-2022 and then was run as a true daily forecast for 2023. Both hindcast and forecast were driven by the 35-day NOAA GEFS weather forecast and iteratively updated based on the assimilation of hourly flux tower eddy-covariance data (NEE and LE), daily NASA SMAP microwave soil moisture, 8-day MODIS LAI, and Harmonized Landsat Sentinel-2 phenology (2-3 day frequency). Dynamic observation operators were used to combine measurements operating at different spatial scales and capture the daily variability in the eddy-covariance footprint. In scaling up our previous tower-scale forecast, which assessed the diurnal and seasonal variability in the predictability of multiple carbon pools and fluxes, a goal of this project was to extend those analyses to look at spatial variability in predictability across a heterogeneous landscape that includes a mixture of hardwood and conifer forests of different stand ages, woody wetlands, and pasture. In general, the optical remote sensing (LAI, phenology) had the greatest impact on constraining landscale-scale forecast uncertainty. Eddy-covariance data had the largest impact on the deciduous forest plant functional type, as this is the PFT that dominates the tower footprint, but other PFTs experienced non-negligible constraints due to the forecast spatial covariance. To externally validate our forecasts we leveraged a network of inventory plots with both dendrometer bands to measure tree growth and permanently installed collars to measure soil respiration. The forecast was better able to capture spatial variability in forest growth than soil respiration, largely because of the challenges in capturing the spatial heterogeneity in soil carbon initial conditions. An important future direction of this work is to leverage landscape-scale forecasts as an input into adaptive monitoring approaches to optimizing the deployment of limited field resources.