Improving E3SM Land Model Photosynthesis Parameterization via Satellite SIF, Machine Learning, and Surrogate Modeling

The parameterization of key photosynthesis parameters is one the uncertain sources in modeling ecosystem gross primary productivity (GPP). Solar-induced chlorophyll fluorescence (SIF) offers a good proxy for GPP since it marks actual process photosynthesis; while machine learning (ML) provides robust approach to model GPP-SIF relationship. Here, we trained boosted regressing tree (BRT) and Random Forest ML models with Greenhouse Gases Observing Satellite SIF data situ observations from 49 eddy covariance towers. These were fed into Energy Exascale Earth System Model (E3SM) Land (ELM) generate ELM-simulated global estimates, which then benchmarked against satellite surrogate approach. Our results indicated performance ML-based ELM when also can well predict spatial-temporal variations SIF. We found high accuracy modeling. parameter sensitivity analysis suggested that fraction leaf nitrogen RuBisCO (flnr) most sensitive SIF; other include Ball-Berry stomatal conductance slope (mbbopt) vcmax entropy (vcmaxse). posterior uncertainty simulated was greatly reduced after benchmarking, produced improved spatial patterns mean relative FLUXCOM GPP. integrated new avenue improving land using remote-sensing SIF, be further future more ground- satellite-based observations.