Integrating process-based model and artificial intelligence for calculating cropland carbon budget in the U.S. Midwest
Abstract
Accurate estimation of carbon budgets is vital to assessing the climate change mitigation potentials of terrestrial ecosystems. Cropland carbon budgets play an important role in regional carbon budgets over the U.S. Midwest where cropland dominates the landscape. However, there is still no reliable product on cropland carbon budget with high spatial and temporal resolutions over the U.S. Midwest. Empirical studies use flux tower observations to quantify different components of cropland carbon budget at local scale, such as net ecosystem exchange (NEE), but it is difficult to scale local observations up to regional scales. Process-based models can simulate individual components of cropland carbon budget, but are lacking effective constraints from observations. To fill this gap, we integrated the advanced ecosystem model, ecosys, with a new remotely-sensed daily ecosystem gross primary production (GPP) observations using the Long Short Term Memory networks (LSTM) to estimate the crop yield, ecosystem respiration (Reco), and NEE at field scale in the U.S. Midwestern cropland. Specially, we built a simulated carbon budget database over Illinois, Indiana, and Iowa under rainfed corn-soybean rotation systems using the ecosys model, and connected the simulated ecosystem autotrophic respiration (Ra), ecosystem heterotrophic respiration (Rh), and crop yield with local soil, climate, and simulated GPP using LSTM model at daily scale. Based on the trained LSTM model, we estimated Ra, Rh, and crop yield with local soil and climate information as well as the remotely sensed GPP data as inputs. We validated our method using the Reco and NEE observations at 11 cropland eddy-covariance sites, and county-scale crop yield data in the U.S. Midwest. Our study highlights the benefits of intertarging of process-based models, state-of-art AI techniques, and high resolution remote sensing products in quantifying and monitoring field-level carbon budget.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2021
- Bibcode:
- 2021AGUFMGC45R..08Z