Mapping Soil Organic Matter, Total Carbon, and Total Nitrogen in Salt Marshes Using UAS-Based Hyperspectral Imaging

Abstract

Wetland ecosystems are critical to global carbon and nitrogen cycles. This study leverages unmanned aerial system (UAS)-based hyperspectral imaging to quantify soil organic matter (SOM), total carbon (C), and total nitrogen (N) in moderately to densely vegetated salt marshes at the Virginia Coast Reserve Long-Term Ecological Research (VCR-LTER) site. We utilized elastic net (ENet) regression and gradient-boosted regression trees (GBRT) within a hybrid modeling framework to predict these soil properties using features from the visible to near-infrared (VNIR) and shortwave infrared (SWIR) spectral ranges. Validated through a 1,000-iteration bootstrap analysis, the hybrid model demonstrated robust predictive capabilities. The model achieved mean normalized root mean square error of 0.118 for SOM, 0.127 for C, and 0.138 for N, with corresponding mean $R^2$ values of 0.874, 0.865, and 0.822, respectively. These outcomes highlight the efficacy of integrating advanced statistical methods with high-resolution remote sensing data to enhance soil property estimation in ecologically sensitive areas.