Abstract:The parameterization of key photosynthesis parameters is one of the key uncertain sources in modeling ecosystem gross primary productivity (GPP). Solar‐induced chlorophyll fluorescence (SIF) offers a good proxy for GPP since it marks the actual process of photosynthesis; while machine learning (ML) provides a robust approach to model the GPP‐SIF relationship. Here, we trained the boosted regressing tree (BRT) and the Random Forest ML models with Greenhouse Gases Observing Satellite SIF data and in situ GPP observations from 49 eddy covariance towers. These trained ML GPP‐SIF models were fed into the Energy Exascale Earth System Model (E3SM) Land Model (ELM) to generate ELM‐simulated global SIF estimates, which were then benchmarked against satellite SIF observations with a surrogate modeling approach. Our results indicated good modeling performance of the ML‐based GPP‐SIF relationship. The ELM model when fed with the ML GPP‐SIF models also can well predict the spatial‐temporal variations in SIF. We also found high model accuracy for the surrogate modeling. Model parameter sensitivity analysis suggested that the fraction of leaf nitrogen in RuBisCO (flnr) is the most sensitive parameter to the SIF; other sensitive parameters include the Ball‐Berry stomatal conductance slope (mbbopt) and the vcmax entropy (vcmaxse). The posterior uncertainty in simulated GPP was greatly reduced after benchmarking, and the model produced improved spatial patterns of mean GPP relative to FLUXCOM GPP. Our integrated approach provides a new avenue for improving land models and using remote‐sensing SIF, which can be further improved in the future with more ground‐ and satellite‐based observations. Model estimation of photosynthesis product, that is, gross primary productivity (GPP), is a challenging but vital task. One of the keys is to find better values for key parameters. This parameter searching process requires good proxies for GPP that can be widely available across space and time, good statistical methods to relate proxies to GPP and to make best estimations that reduce the gaps between modeled results and observations. Here, we designed a new method that use solar‐induced chlorophyll fluorescence (SIF, a good proxy for photosynthesis) as a key input, and employ machine learning (a robust way to relate SIF and GPP) and surrogate modeling (a good method for finding the best parameters), to improve the photosynthesis parameterization in the Energy Exascale Earth System Model (E3SM) Land Model (ELM), a state‐of‐the‐art terrestrial biosphere model. Our results demonstrate that this new integrated approach has great potential for improving the parameterization of key photosynthesis parameters in land models. We built a unique method to improve gross primary productivity (GPP) modeling in land models This method integrates solar‐induced chlorophyll fluorescence observations, machine learning, and surrogate modeling The method reduced posterior uncertainties in simulated GPP and improved the modeling of its spatial patterns We built a unique method to improve gross primary productivity (GPP) modeling in land models This method integrates solar‐induced chlorophyll fluorescence observations, machine learning, and surrogate modeling The method reduced posterior uncertainties in simulated GPP and improved the modeling of its spatial patterns

Representing nitrogen, phosphorus, and carbon interactions in the E3SM Land Model: Development and global benchmarking

Global evaluation of the nutrient enabled version of land surface model ORCHIDEE-CNP (v1.2)

Nutrient Dynamics in a Coupled Terrestrial Biosphere and Land Model (ELM‐FATES‐CNP)

Global evaluation of the nutrient-enabled version of the land surface model ORCHIDEE-CNP v1.2 (r5986)

GOLUM-CNP V1.0: a Data-Driven Modeling of Carbon, Nitrogen and Phosphorus Cycles in Major Terrestrial Biomes

Ecosystem C and N cycle interactions – diverse model representations and divergent model predictions versus collective empirical constraints

GOLUM-CNP v 1 . 0 : a data-driven modeling of carbon , nitrogen and 1 phosphorus cycles in major terrestrial biomes 2

Simulating Global Terrestrial Carbon and Nitrogen Biogeochemical Cycles With Implicit and Explicit Representations of Soil Microbial Activity

Benchmarking Carbon Fluxes of the Isimip2a Biome Models

Representation of the Terrestrial Carbon Cycle in CMIP6

Coupling of Phosphorus Processes With Carbon and Nitrogen Cycles in the Dynamic Land Ecosystem Model: Model Structure, Parameterization, and Evaluation in Tropical Forests

The DOE E3SM v1.1 biogeochemistry configuration: description and simulated ecosystem‐climate responses to historical changes in forcing

Plant responses to elevated CO2 under competing hypotheses of nitrogen and phosphorus limitations

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

The Impact of Climate Forcing Biases and the Nitrogen Cycle on Land Carbon Balance Projections

Modelling of land nutrient cycles: recent progress and future development

Development of a plant carbon–nitrogen interface coupling framework in a coupled biophysical-ecosystem–biogeochemical model (SSiB5/TRIFFID/DayCent-SOM v1.0)

Global 1 km land surface parameters for kilometer-scale Earth system modeling

A Framework for Benchmarking Land Models

Taking off the training wheels: the properties of a dynamic vegetation model without climate envelopes, CLM4.5(ED)

Simulating dynamic roots in the Energy Exascale Earth System Land Model