Using Sentinel-1 Imagery for Soil Salinity Prediction Under the Condition of Coastal Restoration

Soil salinity is a major cause of land degradation in coastal environments and arid lands; in the first case due to sea water, and the second case due to precipitation/evaporation relationship. In coastal wetlands, soil salinity is very sensitive to plant invasion. In this context, it is necessary to obtain a better understanding of soil salinity variation to improve the management of coastal land resources. In this study, we explored the potential of Sentinel-1 data in predicting electrical conductivity (EC) at three depths. Also, we assessed the usefulness of the knowledge of the invasion process in EC prediction by comparing structural equation modeling (SEM), that included such knowledge, and linear regression model (LM), that simply modeled the relationships between EC and predictors. The case study was conducted in an invaded coastal wetland dominated by Spartina alterniflora Loisel in the east-central China coast. Before modeling, principal component analysis was used to reduce the multidimensionality of time series images. In SEM, the model explained 82% of EC variation in 0-30 cm, 99% in 30-60 cm, and 71% in 60-100 cm. The cross validation showed the SEM model provided good accuracy, with RPD (a ratio of performance to deviation) values of 1.41 in 0-30 cm, 1.51 in 30-60 cm, and 1.43 in 60-100 cm. In comparison to the poorer accuracy of LM models, we argued that modeling the relationships between the exotic plant and EC at different depths can be treated as a substantial advantage of the approach. These results provided useful indications about the strong potentials of Sentinel-1 imagery in quantitative prediction of soil salinity.