Estimation of Satellite Derived Flood Area and Volumes to Monitor Environmental Watering Events

Abstract

Wetlands are significant repositories of biological diversity and play an essential role in maintaining ecosystem services. The Macquarie Marshes in northern New South Wales is one of those significant wetlands recognised for its rich biodiversity and ecology. The Macquarie Marshes experienced extensive flooding events, especially in 2016-17 due to an increase in the water flow in Macquarie River related to a strong La Niña climate cycle. The area has also experienced drought conditions when there is insufficient water flowing in the river or due to the diversion of water to irrigated lands. Mapping the inundation patterns of the Macquarie Marshes is essential for understanding the impacts of climate variability on these important ecosystems. The study analysed 28 Sentinel-1 and 47 Sentinel-2 images covering the La Niña year 2022. The Sentinel-1 images were pre-processed and co-registered to delineate water. For mapping water bodies using Sentinel-2, multiple water detection algorithms were used to quantify flooded areas based on the Sentinel 2 satellite data. Flooded areas obtained using different water indices were compared with the river discharge at several gauge stations located at different reaches of the Macquarie River. Overall, the Fisher’s Water Index (WI) gave the best results relative to PlanetScope verification data, but all Sentinel 2 methods indicated some underestimation of overall water areas. Sentinel 1 appeared to strongly overestimate the flooded area. Therefore, the final step fuses the Sentinel-1 and Sentinel-2 data, using a layer-stacking technique; the random forest (RF) classifier was then applied to predict flooded areas using the Water Index (WI), other satellite variables and environmental variables related to slope, depressions, land use and rainfall. Finally, floodwater depth of the water inundated using the RF model was calculated using a 5m resolution LiDAR DEM dataset.