ASABE Technical Library - Abstract

Click on “Download PDF” for the PDF version or on the title for the HTML version.
If you are not an ASABE member or if your employer has not arranged for access to the full-text, view access and subscription options

Improving Ecosystem Phenology Monitoring in Salt Marsh Using Vision Foundation Models

Published by the American Society of Agricultural and Biological Engineers, St. Joseph, Michigan www.asabe.org

Citation:  2026 ASABE Annual International Meeting  2600608.(doi:10.13031/aim.202600608)
Authors:   Puranjit Singh, Himal Acharya, Rodrigo Vargas, Yin Bao
Keywords:   Carbon Phenology, Climate Modeling, DepthAnythingV2, DINOv2, DINOv3, Machine Learning, Salt Marsh Ecosystem, Vision Foundation models.

Abstract. Salt marshes are critical coastal ecosystems that sequester carbon, support biodiversity, and facilitate nutrient cycling. Accurate monitoring of gross primary production (GPP), which represents the total amount of carbon fixed through photosynthesis, is essential for understanding coastal carbon dynamics and assessing climate mitigation potential. Traditional measurement approaches rely on spatially limited eddy covariance towers, necessitating development of scalable remote sensing techniques. This study evaluates the effectiveness of vision foundation models for predicting GPP in salt marshes using proximal PhenoCam imagery. Fixed-position PhenoCam images from the St. Jones Reserve were analyzed to extract feature embeddings from vision foundation models of DINOv2, DINOV3, and DepthAnythingV2. Extracted features were used to train machine learning regressors identified through LazyPredict library and optimized using the GridSearch algorithm. DINOv3 combined with Bayesian Ridge regression achieved strong predictive performance (R² = 0.95, MAE = 0.75 µmol CO_2 m^(-2) s^(-1) ), demonstrating the effectiveness of vision foundation models in capturing spectral and temporal patterns relevant to ecosystem productivity. These findings advance near-surface remote sensing applications for coastal ecosystem monitoring and highlight the potential of vision foundation models for scaling ecosystem carbon exchange assessments.

(Download PDF)    (Export to EndNotes)