Spatio-temporal mapping of Phragmites australis across Great Salt Lake wetlands, Utah, USA, using probabilistic multi-sensor random forest approach

Mapping Phragmites australis in dynamic wetlands requires models that are explicit about area uncertainty. We built annual probabilistic Random Forest (RF) models for Phragmites presence along the Great Salt Lake, Utah. Predictors derived from Sentinel-2 and backscatter from Sentinel-1 were combined with field labels compiled from mechanized treatment activities. Annual probability surfaces for 2018–2024 were thresholded by F1 maximization, and the mapped extent was corrected for classification errors. Model performance was strong (overall accuracy ≈ 0.90–0.95; F1 for Phragmites ≈ 0.86–0.92; area under the receiver operating characteristic curve (AUC-ROC) ≈ 0.98–0.99; area under the precision–recall curve (AUPRC) ≈ 0.95–0.98; Brier ≈ 0.04–0.07). Within the 35,463 ha analysis extent, Phragmites occupied a 24–35% range across years. Change-detection analysis showed that 8,710 ha transitioned to Phragmites between 2018 and 2024, indicating substantial turnover. Inter-annual dynamics showed the largest increase between 2021 and 2022 (+3,742 ha; +42.0%), and the largest decline between 2022 and 2023 (−2,021 ha; −16.0%).