Leveraging the Historical Landsat Catalog for a Remote Sensing Model of Wetland Accretion in Coastal Louisiana

Abstract

AbstractA wetland's ability to vertically accrete—capturing sediment and biological matter for soil accumulation—is key for maintaining elevation to counter soil subsidence and sea level rise. Wetland soil accretion is comprised of organic and inorganic components largely governed by net primary productivity and sedimentation. Sea level, land elevation, primary productivity, and sediment accretion are all changing across Louisiana's coastline, destabilizing much of its wetland ecosystems. In coastal Louisiana, analysis from 1984 to 2020 shows an estimated 1940.858 km2 of total loss at an average rate of 53.913 km2/year. Here we hypothesize that remote sensing timeseries data can provide suitable proxies for organic and inorganic accretionary components to estimate local accretion rates. The Landsat catalog offers decades of imagery applicable to tracking land extent changes across coastal Louisiana. This dataset's expansiveness allows it to be combined with the Coastwide Reference Monitoring System's point‐based accretion data. We exported normalized difference vegetation index (NDVI) and red‐band surface reflectance data for every available Landsat 4–8 scene across the coast using Google Earth Engine. Water pixels from the red‐band were transformed into estimates of total suspended solids to represent sediment deposition—the inorganic accretionary component. NDVI values over land pixels were used to estimate bioproductivity—representing accretion's organic component. We then developed a Random Forest regression model that predicts wetland accretion rates (R2 = 0.586, MAE = 0.333 cm/year). This model can inform wetland vulnerability assessments and loss predictions, and is to our knowledge the first remote sensing‐based model that directly estimates accretion rates in coastal wetlands.