Using imaging spectroscopy and elevation in machine learning to estimate soil salinity in intermittently tidal wetlands

Coastal soil salinization patterns are changing due to drought, sea level rise (SLR), and changing freshwater inflow. These changes are expected to impact coastal wetland plant health and ecosystem function, such as changes to biomass and productivity. These impacts have led to greater interest in how we monitor soil salinization across spatial and temporal scales. Remote sensing is a promising tool for estimating soil salinity at the spatial scales required for decision making by land managers. However, the development of a remote sensing estimation approach for wetland soil salinity must account for two factors: (1) the high spatial and temporal heterogeneity of coastal wetlands and (2) the fact that soil salinity is the result of multiple historical land use, hydrological, and geomorphic processes. In spring 2022, a combined airborne-field campaign, known as SHIFT, collected a weekly time series of airborne visible to shortwave infrared (VSWIR) image spectroscopy data. This dataset provides a unique opportunity to assess the application of fine spatial (5 m) and temporal (weekly) resolution VSWIR data to estimate root zone soil salinity; when combined with environmental variables such as elevation, these data can account for some of these factors. In this study, we utilized VSWIR and elevation datasets in a random forest regression to predict and map soil salinity in an intermittently tidal estuary, Devereux Slough, located in Santa Barbara County, California. The final model combined spectral indices with elevation to better capture soil salinity dynamics despite lower correlation (r = 0.85) than solely using elevation (r = 0.92). This research demonstrates the utility of remote sensing datasets, namely, elevation and the modified Anthocyanin Reflectance Index (mARI), for predicting root zone soil salinity in intermittently tidal coastal wetlands. These findings are an important step in advancing coastal remote sensing by creating a gridded salinity dataset that can be used for salinity monitoring and other coastal applications, such as modeling change in vegetation communities or ecosystems facing the impacts of climatic variability and change.