Flooding in Landsat across tidal systems (FLATS): An index for intermittent tidal filtering and frequency detection in salt marsh environments

Remote sensing can provide critical information about the health and productivity of coastal wetland ecosystems, including extent, phenology, and carbon sequestration potential. Unfortunately, periodic inundation from tides dampens the spectral signal and, in turn, causes remote sensing-based models to produce unreliable results, altering estimates of ecosystem function and services. We created the Flooding in Landsat Across Tidal Systems (FLATS) index to identify flooded pixels in Landsat 8 30-meter data and provide an inundated pixel filtering method. Novel applications of FLATS including inundation frequency and pattern detection are also demonstrated. The FLATS index was developed to identify flooding in Spartina alterniflora tidal marshes. We used ground truth inundation data from a PhenoCam and Landsat 8 pixels within the PhenoCam field of view on Sapelo Island, GA, USA to create the index. The FLATS index incorporates a normalized difference water index (NDWI) and a phenology-related variable into a generalized linear model (GLM) that predicted the presence or absence of marsh flooding. The FLATS equation for predicting flooding is 1-1/(-1.6+20.0*NDWI)(+68.6*Pheno)(e)(4,6)(3,4), and we found that a cutoff 0.1 was the optimized value for separating flooded and non-flooded pixel classes. FLATS identified flooded pixels with an overall accuracy of 96% and 93% across training data and novel testing data, respectively. FLATS correctly identified true flooded pixels with a sensitivity of 97% and 81%, across training and testing data, respectively. We established the need to apply FLATS when conducting vegetation time-series analysis in coastal marshes in order to reduce the per-pixel reflectance variations attributed to tidal flooding. We found that FLATS identified 12.5% of pixels as flooded in Landsat 8 tidal marsh vegetation time-series from 2013 to 2020, after traditional quality control and preprocessing steps were conducted, which could then be filtered out or modeled separately in order to conduct remotely sensed vegetation assessments. Therefore, in tidal wetlands, we recommend incorporating FLATS into Landsat 8 preprocessing prior to vegetation analysis. We also demonstrated innovative applications for the FLATS index, particularly in detecting flooding frequency and flooding patterns relevant to the broader biophysical modeling framework, including mapping marsh vulnerability due to fluctuation in inundation frequency. The FLATS index represents advancements in the understanding and application of inundation indices for coastal marshes.

Automated summary

Remote sensing is crucial for understanding coastal wetland ecosystems, but tidal inundation affects the reliability of remote sensing-based models. To address this issue, we developed the Flats index to identify and filter flooded pixels. We also demonstrated innovative applications of the index in detecting flooding frequency and patterns.