Tidal marshes offer numerous ecological and economic benefits but are threatened by human development, sea-level rise, and invasive species. Routine mapping of their species composition is crucial for coastal management, yet current efforts are spatially and temporally limited and lack species identification. To address this, we used 3-m PlanetScope satellite imagery to classify four common marsh plants across Virginia’s Middle Peninsula in Chesapeake Bay, a region targeted for restoration given its ecological and economic significance. We developed a random forest classifier using May 2021 reference data from the Virginia Institute of Marine Science, delineating Spartina alterniflora , Phragmites australis , Spartina patens , and Juncus roemerianus at eight marshes. Balanced agreement was strong at 93%, ranging from 63%-98% across individual marshes. We then expanded to the broader Middle Peninsula, using the Mann-Whitney U test to compare satellite- and reference-derived coverage of P. australis , finding moderate agreement despite a large temporal offset ( r rb = 0.45; N = 152). Following suitable model performance, we generated annual assessments for May 2021-2024. S. alterniflora was most extensively distributed, covering half of our study area. We also analyzed random forest class probabilities to inform data collection and model interpretation. Probabilities were generally above 0.6, although those accompanying J. roemerianus were notably lower. We present methods for large-scale species mapping to inform resource prioritization and coastal management, including a framework for communicating classification certainty, which is adaptable to stakeholder needs. Our framework can be used to retrain our classification model for application elsewhere, where local field data is available. • Methods are presented for routine, large-scale tidal marsh species mapping • Four species were mapped across 40 km 2 of Chesapeake Bay’s Middle Peninsula • PlanetScope data and a random forest model delineated species with 93% agreement • Random forest class probabilities were used to communicate classification certainty • Results can inform resource prioritization and coastal management
Coffer et al. (Thu,) studied this question.