Optical Imaging Combined with AI Reveal the Role of Root Biomass in Wetlands

The Science

Coastal wetlands help protect coasts from storms and store significant amounts of carbon in the land-water interface. Assessing belowground processes critical to coastal wetland functioning, however, is challenging as traditional sampling techniques to assess plant root health are destructive and labor-intensive, contributing to gaps in fundamental nutrient cycling understanding. This multi-institutional study used a non-destructive, light-based imaging tool called optical coherence tomography (OCT) to penetrate deep into live roots of soil core samples from wetlands to determine root vitality. By combining these images with a newly developed machine learning (ML) model now called RhizoAI, the team was able to distinguish healthy vs. decaying roots where traditional other imaging methods fail. This new combination of techniques reduces sample processing times from days to a few minutes. The non-destructive nature of OCT also allows for repeated measurements on the same specimen and can be integrated into future field-deployable or large-area scanning approaches, greatly enhancing scalability and spatial coverage. This approach provides a novel, efficient, and non-intrusive way to monitor wetland health and resilience over time.

The Impact

Understanding how plant roots grow and decay is key to predicting soil stability, carbon storage, and the resilience of coastal wetlands. Current methods to characterize such processes, however, are slow, destructive, and difficult to use for large-scale or long-term studies. Conventional root sampling and characterization typically rely on subjective visual classification and oven drying to quantify biomass and tissue condition—processes that can take days per sample and require substantial labor. A multi-institutional team of researchers found that by combining OCT with the RhizoAI ML model, they can quickly and accurately identify live and dead roots without disturbing the soil—all within a span of a few minutes. OCT imaging acquires high-resolution, 3D volumetric data within minutes per sample, and the AI-driven workflow automatically classifies live and dead tissues. This provides a strong foundation for the further development of real-time, non-destructive monitoring of root health and provides new tools to guide wetland restoration and improve predictions of ecosystem functioning and evolution. Fundamental insights from this research have potential applications for the growth of bioenergy crops on marginal lands, in phytomining, as well as for energy-related land-use management practices and mitigating operational impacts stemming from changing environmental conditions.

Summary

Coastal wetlands play a vital role in protecting shorelines, storing carbon, and supporting ecosystems. Plants grown in these areas serve as important sources for bioenergy development and critical minerals extraction through phytomining. However, understanding how plant roots in these areas grow, decay, and contribute to the stability of soils at the land-water interface has been limited by destructive and time-consuming field methods. This study used OCT, a light-based imaging technique, to capture high-resolution, three-dimensional images of wetland roots collected from the Terrebonne Basin in Louisiana. The Environmental Molecular Sciences Laboratory (EMSL), a Department of Energy Office of Science user facility, partnered with a team from Texas A&M University to develop a ML model now called RhizoAI, which distinguishes between live and dead root tissues with high accuracy. The team successfully met their proof-of-concept goal threshold of about 70 percent accuracy, which is considered a strong initial result for demonstrating feasibility. The OCT imaging was performed at EMSL using the Ganymede spectral-domain OCT system, which allowed precise visualization of root microstructures at a few microns of resolution. EMSL's expertise in advanced imaging, data processing, and high-performance computing enabled the development and testing of an automated classification model. The results demonstrate the feasibility of combining OCT with ML to analyze root health in a non-destructive, repeatable, and scalable way. This work paves the way for future integration of OCT with x-ray computed tomography and deep learning tools, ultimately advancing coastal monitoring and restoration strategies. Additionally, it broadens investigations of important processes controlling productivity of bioenergy crops and phytomining.

Contacts

Mohamed Hassan 
Texas A&M University 
mhassan@tamu.edu

Maruti Mudunuru 
Pacific Northwest National Laboratory
maruti@pnnl.gov

Navid Jafari 
Texas A&M University 
njafari@tamu.edu

Funding

This research was supported by a limited scope project from the Environmental Molecular Sciences Laboratory, a Department of Energy Office of Science user facility sponsored by the Biological and Environmental Research program. Additional support was provided by the Department of Civil and Environmental Engineering and the Department of Oceanography and Coastal Sciences at Louisiana State University through the Anticipating Threats to Natural Systems project, funded by the U.S. Army Engineer Research and Development Center under Award ID W912HZ2020070.

Publication

M.O. Hassan; Truong, A.; Mudunuru, M.; Butler, L.G.; Rovai, A.; Larimer, C.; Daddona, J.; Ahkami, A.H.; Bardhan, J.; Varga, T.; Stratton, K.; Karra, S.; Twilley, R.; and Jafari, N.H. "From pixels to patterns: Coupling Optical Coherence Tomography and machine learning for monitoring coastal wetland root systems." Science of The Total Environment, 999, 180315 (2025). [DOI: 10.1016/j.scitotenv.2025.180315]