Software: Constrained carbon partitioning using a self-trained physics-informed machine learning

KGML Carbon Partitioning Reference:Sadegh Ranjbar, Ankur R. Desai, Sophie Hoffman, Einara Zahn, Elie Bou-Zeid, and Paul C. Stoy. (2025). Constrained carbon partitioning: a self-trained physics-informed machine learning model to partition carbon measurements into GPP and RECO from eddy covariance measurements. Constrained Carbon Partitioning using KGML This repository contains code for the study: Constrained carbon partitioning: a self-trained physics-informed machine learning model reduces GPP overestimation from eddy covariance measurements Sadegh Ranjbar1, Ankur R. Desai2, Sophie Hoffman1, Einara Zahn3,4, Elie Bou-Zeid3, Paul C. Stoy1 1Department of Biological Systems Engineering, University of Wisconsin – Madison, Madison, WI, USA. 2Department of Atmospheric and Oceanic Sciences, University of Wisconsin – Madison, Madison, WI, USA. 3Department of Civil and Environmental Engineering, Princeton University, Princeton, New Jersey, USA. 4Department of Earth and Environmental Science, University of Pennsylvania, Philadelphia, PA, USA Overview Gross Primary Productivity (GPP) is a key component of the carbon cycle but cannot be directly measured.This project implements a knowledge-guided machine learning (KGML) model to partition Net Ecosystem Exchange (NEE) into GPP and ecosystem respiration (RECO), constrained by physical laws and theoretical expectations. Features Self-supervised, physics-informed KGML model Monte Carlo dropout for uncertainty estimation Visualizations of energy balance, GPP, RECO, and WUE Reproducible workflow for NEON tower datasets