Abstract: The spatial distribution of salt marsh vegetation in Yellow River Delta are highly heterogeneous. Accurate information on the historical distribution of salt marsh is of great significance for regional ecological stability and sustainable development. We constructed a long-series temporal-spatial-spectral multidimensional elicitation based on multi-source data, and accurately extracted information on the spatial distribution of typical salt marsh in the Yellow River Delta from 1996 to 2022 using a random forest (RF) model with recursive feature elimination, and further analyzed the succession of the native/invasive salt marsh communities since the diversion of the Yellow River in 1996. Compared to the single temporal spectral feature, the use of a temporal-spatial-spectral multidimensional feature set for extraction improved the overall accuracy of salt marsh vegetation classification by 8.4%. The classification effect of the sparse Suaeda salsa and the mixed area of Phragmites australis and Spartina alterniflora was optimized based on the temporal and spatial features of optical and SAR images. The distribution of salt marsh on the tidal flats after the Yellow River was diverted was obvious. The cover area of S. salsa communities decreased from 91.67 km² in 1996 to 38.11 km² in 2022, with the successional trend being influenced by the invasion of S. alterniflora. S. alterniflora was rapidly expanded and then distributed in large areas on the tidal flats on both sides of the current river channel since 2008. The area of the community reached the maximum (51.25 km²) in 2020. The invasion and expansion of S. alterniflora had a certain impact on the habitat pattern of the tidal flats.

Key words: multi-source remote sensing, multidimensional feature extraction, salt marsh vegetation, community succession, Yellow River Delta