Monitoring Terrestrial Ecosystem Productivity Using Hyperspectral Satellite Data

Authors

• Serge Tuyambaze, University of Nebraska-LincolnFollow

First Advisor

Ran Wang

Committee Members

John Gamon, Brian Wardlow

Date of this Version

4-2026

Document Type

Thesis

Citation

A thesis presented to the faculty of the Graduate College at the University of Nebraska in partial fulfillment of requirements for the degree of Master of Science

Major: Natural Resource Sciences

Under the supervision of Professor Ran Wang

Lincoln, Nebraska, April 2026

Comments

Copyright 2026, Serge Tuyambaze. Used by permission

Abstract

Accurate estimation of terrestrial Gross Primary Productivity (GPP) is critical for quantifying global carbon sequestration and understanding the terrestrial carbon cycle. While eddy covariance (EC) towers provide standard flux measurements, upscaling these measurements to continuous global coverage remains challenging due to the sparse distribution of flux towers and their small footprints. Remote sensing (RS), especially coupled with eddy covariance data, has been used to estimate GPP at the global scale. Different RS derived productivity models have been developed, including vegetation indices, light use efficiency (LUE) models, solar-induced fluorescence (SIF), dynamic global vegetation models (DGVM), and machine learning. One widely adapted RS based productivity model is the LUE framework which calculates GPP as the product of absorbed photosynthetically active radiation (APAR) and vegetation light use efficiency (ε). While greenness vegetation indices (e.g., NDVI) estimate the fraction of absorbed photosynthetically active radiation (fAPAR), traditional LUE models have often used interpolated meteorological data to estimate light use efficiency (ε), which introduce significant uncertainties in GPP estimation.

This study explores the use of data from NASA’s latest hyperspectral mission, Plankton, Aerosol, Cloud, ocean Ecosystem (PACE), to directly parameterize the LUE model. We evaluated multiple models using NDVI and the near-infrared reflectance of vegetation (NIRv) as proxies for fAPAR, photochemical reflectance index (PRI) and chlorophyll-carotenoid index (CCI) as direct proxies for ε. Using data from 54 sites from AmeriFlux, we compared the LUE model predictions to both EC measured GPP and GPP from other RS-based models including NIRv and solar induced fluorescence (SIF). Our results illustrated that LUE models using PRI as proxy for ε (GPP_APAR,PRI and GPP_NIRv,PRI) emerged as the best proxy for GPP. Decomposing the LUE model components revealed distinct ecosystem specific drivers for GPP intra-annual variations. fAPAR and NIRv were main drivers for productivity changes in ecosystems that undergo seasonal structural changes (e.g., croplands, deciduous forests). In contrast, physiological proxies were the main drivers in ecosystems without pronounced seasonal pattern such as in water-limited environments (e.g., savannas). By re-evaluating the LUE model with recent hyperspectral data, this study demonstrates how next generation global hyperspectral data can improve modeling of terrestrial ecosystem productivity.

Advisor: Ran Wang