The inherent heterogeneity of coastal wetlands and the small size of halophytic plants present challenges in accurately sensing plant species, even with very high-resolution satellite imagery. This study used sub-pixel imagery classification methods on high spectral and spatial resolution imagery from Worldview-3 to predict plant species distribution in a mesotidal coastal wetland system. The predicted sub-pixel fractional abundance of plant species is discussed for three targeted wetland categories in the Ria Formosa lagoon: naturally evolving patches, patches modified by human activities, and patches affected by coastal squeeze. The Random Forest Regression algorithm was proven to be highly effective in unmixing the spectral signal of halophytic vegetation, enabling the retrieval of plant species distribution (7 plant species). To train the algorithm, field observations were used to classify satellite images. Differences in band feature importance for key species and bare soil were observed across the various sites. The comparison of species distribution between sites suggests that, in addition to biotic factors, other environmental influences likely affect ecological succession; therefore, large-scale mapping approaches based on remote sensing should be undertaken with caution. The results are important for understanding the diverse ecological behavior of marsh plants within the same system and highlight the variability in plant reflectance and the need for ground truthing when sensing plant cover from satellite data.