Mapping inundation in the heterogeneous floodplain wetlands of the Macquarie Marshes, using Landsat Thematic Mapper

•Water and vegetation indices effectively detected inundation in a large heterogeneous wetland.•We accurately mapped flooded area using three classes: water, mixed and vegetation.•Inundation maps revealed high variability in extent and regional connectivity.•Our results were essential for measuring flooding regimes and managing environmental flows. Flood dependent aquatic ecosystems worldwide are in rapid decline with competing demands for water. In Australia, this is particularly evident in the floodplain wetlands of semi-arid regions (e.g. the Macquarie Marshes), which rely on highly variable flooding from river flows. Environmental flows mitigate the impacts of river regulation, inundating floodplains, thereby rehabilitating degraded habitats. Mapping flooding patterns is critical for environmental flow management but challenging in large heterogeneous floodplains with variable patterns of flooding and complex vegetation mosaics. We mapped inundation in the Macquarie Marshes, using Landsat 5 TM and Landsat 7 ETM+ images (1989–2010). We classified three inundation classes: water, mixed pixels (water, vegetation, soil) and vegetation (emergent macrophytes obscuring inundation), merged to map inundated areas from not-inundated areas (dry land). We used the Normalised Difference Water Index (NDWIB2/B5), masked by the sum of bands 4, 5, and 7 (sum457), to detect water and mixed pixels. Vegetation was classified using an unsupervised classification of a composite image comprising two dates representing vegetation senescence and green growth, transformed into two contrasting vegetation indices, NDVI and NDIB7/B4. We assessed accuracy using geo-referenced oblique aerial photography, coincident with Landsat imagery for a small and large flood, producing respective overall accuracies of inundated area of 93% and 95%. Producer’s and user’s accuracies were also high (94–99%). Confusion among inundation classes existed but classes were spectrally distinct from one another and from dry land. Inundation class areas varied with flood size, demonstrating the variability. Inundation extent was highly variable (683–206,611ha). Floods up to 50,000ha were confined to the north and south wetland regions. Connectivity to the east region only occurred when flooding was greater than 51,000ha. Understanding the spatiotemporal dynamics of inundation is critical for quantifying the environmental flow requirements across the suite of biota in the Ramsar-listed Macquarie Marshes.