Estimating Soil Salinity with Different Levels of Vegetation Cover by Using Hyperspectral and Non-Negative Matrix...

Area of Science:

• Remote Sensing
• Soil Science
• Spectroscopy

Background:

• Hyperspectral technology is a valuable tool for monitoring soil salt content (SSC).
• Partial vegetation cover on soil surfaces significantly limits the accuracy of hyperspectral SSC estimation.

Purpose of the Study:

• To quantify the impact of varying fraction vegetation coverage (FVC) on hyperspectral SSC estimation.
• To explore the potential of non-negative matrix factorization (NMF) to mitigate FVC influences on SSC estimation.

Main Methods:

• Simulated mixed soil scenes with controlled SSC and FVC were created in a laboratory setting.
• Hyperspectral data were acquired from these mixed scenes.
• Non-negative matrix factorization (NMF) was applied to extract soil spectral signals, followed by partial least squares regression (PLSR) for SSC estimation.

Main Results:

• Accurate SSC estimation was achieved using original mixed spectra for FVC up to 25.76% (R²cv = 0.68, RMSEcv = 5.18 g·kg⁻¹, RPD = 1.43).
• NMF-extracted soil spectra significantly improved SSC estimation accuracy compared to original mixed spectra.
• NMF-extracted spectra from scenes with FVC below 63.55% yielded acceptable SSC estimation (e.g., R²cv = 0.69, RMSEcv = 4.15 g·kg⁻¹, RPD = 1.8).

Conclusions:

• NMF is a promising algorithm for reducing vegetation interference in hyperspectral SSC estimation.
• The NMF-extracted soil spectra retain crucial spectral information for accurate SSC monitoring, even under partial vegetation cover.
• A combined Spearman correlation and variable importance projection analysis strategy effectively evaluates model performance.