Landform controls on SOC and calcite‐equivalent SIC concentrations in Caspian coastal soils: VNIR‐SWIR spectroscopy and Landsat‐8 upscaling

Abstract

Carbon partitioning between organic and carbonate‐associated forms remains poorly constrained in arid to semiarid coastal soils, where geomorphic position, salinity, and carbonate accumulation jointly structure surface‐soil carbon patterns. We evaluated landform controls on surface (0–20 cm) soil organic carbon (SOC), calcite‐equivalent soil inorganic carbon (SIC_eq), soil total carbon (STC), and SIC_eq:SOC across six landforms of the southeastern Caspian coastal plain. We analyzed 136 composited samples using conventional laboratory measurements, visible‐near infrared (VNIR)‐short wave infrared (SWIR) spectroscopy, Landsat‐8 reflectance, and linear and nonlinear predictive models, evaluated with a random calibration–validation split for laboratory spectra and leave‐one‐transect‐out spatial cross‐validation for Landsat‐scale models. Mean (± SD) SOC, calcite‐equivalent carbonate, SIC_eq, and STC concentrations were 1.03 ± 0.67%, 28.32 ± 8.93%, 3.40 ± 1.07%, and 4.43 ± 1.30%, respectively. Landform effects were significant for SIC_eq (Kruskal–Wallis H  = 57.258, p  < 0.0001), STC ( H  = 49.356, p  < 0.0001), and SIC_eq:SOC ( H  = 18.684, p  = 0.0022), but not SOC ( H  = 10.540, p  = 0.0613). Laboratory VNIR‐SWIR modeling showed target‐, algorithm‐, and preprocessing‐dependent performance, with first‐derivative spectra frequently improving validation and carbonate‐associated variables showing clearer spectral relationships than SOC. Direct laboratory‐to‐Landsat transfer, transfer‐learning adaptation, and empirical Landsat‐based digital soil mapping showed weak to near‐null predictive performance, indicating limited transferability from controlled laboratory spectra to 30 m satellite observations. Overall, coastal landforms organized present‐day surface carbon primarily through carbonate‐associated carbon and organic–inorganic carbon partitioning. These findings support joint consideration of SOC and carbonate‐associated carbon and highlight the need for depth‐resolved sampling, carbonate‐phase verification, paired dry‐combustion SOC calibration, and independent external validation before operational mapping.