Work

Satellite-based surface water change detection and mesohabitat classification at Big Blue—Kansas Rivers confluence

Year: 2026

Type: article

Abstract: Mapping dynamic riverine habitats presents a particular challenge, often requiring high resolution bathymetric measurement and advanced sediment transport and hydraulic computational methods. To assess satellite-based screening under these constraints, this study links three analysis steps in sequence: satellite image classification, change detection against a fixed time-stacked baseline, and comparison against hydraulic-model output, then extends the analysis to direct prediction of hydraulically defined aquatic mesohabitats. This study evaluates two open-source workflows for mapping surface-water change and fluvial mesohabitats from optical satellite data: a cluster-assisted pipeline and a class-balanced supervised random forest. Increasingly complex targets were tested: (i) inundation dynamics, defined as surface water versus land within the river corridor, and (ii) direct prediction of eight hydraulically derived mesohabitat classes. Using PlanetScope (~3 m), Sentinel-2 (10–20 m), and Landsat-9 (30 m) imagery, binary inundation mapping was reliable for both workflows (mean balanced accuracy 0.810 cluster-assisted and 0.823 random forest across seven date-matched hydraulic conditions), whereas direct eight-class mesohabitat classification was not (mean leave-one-date-out accuracy 0.100 and 0.149, respectively, against a 0.125 random-guess expectation). Cluster-assisted surface-water area was positively associated with both Wamego discharge and hydraulic surface-water area (Spearman ρ = 0.893, p = 0.012), while the random-forest associations were weak (ρ = −0.214, p = 0.662). These results suggest that lightweight, open workflows may support screening of surface-water change, while also highlighting that direct image-based mesohabitat prediction remains experimental at this site. Future gains will likely require incorporating texture and morphological context, and integrating hydraulic predictors. Research Domain: Remote Sensing, Machine Learning, Ecosystem Modeling.

Source: Ecological Informatics

DOI: 10.1016/j.ecoinf.2026.104110

Authors: Elvin Cordero, Aubrey Harris

Institutions: University of Puerto Rico-Mayaguez, United States Army Corps of Engineers, U.S. Army Engineer Research and Development Center

Language: English

Topic: Remote-Sensing Image Classification

Subfield: Computer Vision and Pattern Recognition

Field: Computer Science

Domain: Physical Sciences

SDGs: Clean water and sanitation, Life on land

Open Access status: gold

Estimated APC paid: $2,770