Abstract
Soil monitoring information is important to improve our understanding of the role of soil on global environment change such as invasion of foreign species. For regions with dense vegetation cover the use of remote sensing data provides an attractive solution to soil prediction through the relationship between soil and remotely sensed information of vegetation, especially considering the availability of multi-temporal series of synthetic aperture radar (SAR) data such as Sentinel-1. In this study, we used a structural equation model (SEM) to link soil organic carbon (SOC) and bulk density (BD) with temporal variation of SAR signals, taking into account possible interacting relationships of the soil-vegetation system. The test area is in the coastal wetlands of east-central China, where Sentinel-1 data were acquired during the vegetation growing season in 2017. A total of fifteen sites were sampled at three depths: 0-30 cm, 30-60 cm, and 60-100 cm. Predictive accuracy was assessed using leave-oneout cross-validation (LOOCV). Results showed that SE models successfully predicted SOC (RMSE = 1.63 g kg(-1), RPD = 1.22) and BD (RMSE = 0.14 g cm(-3), RPD = 1.25) at three depths. We found that SEM supported the idea that the interrelationships exist among soil, vegetation, and remotely sensed information, and improved our ability to investigate relationships between SAR backscatters and soil attributes. The use of time series Sentinel-1 data allowed capturing characteristics of vegetation dynamics and the possible relationships between soil attribute and vegetation. The findings from this study highlight the usefulness of dense temporal SAR data and SEM in soil prediction.
NSTL
Elsevier