Soil Salinity Properties from Hyperspectral Sensing Across 17 Indian Coastal Districts

372 soil samples from 17 coastal districts in western India were analyzed for properties like pH and salinity using hyperspectral reflectance data in the 350-2500 nm range. Gopal Ramdas Mahajan of the Central Coastal Agricultural Research Institute calibrated spectral data with five multivariate techniques, achieving acceptable to excellent prediction performance. The study identified specific absorption wavelengths, such as 494 nm and 2207 nm, correlated with soil salinity levels.

Use Cases

• Train machine learning models for soil salinity prediction based on hyperspectral reflectance features.
• Compare the performance of linear (PCR, PLSR) and non-linear (MARS, RF, SVR) regression techniques on spectral data.
• Monitor soil degradation and salinity in coastal agricultural regions using remote sensing indicators.
• Develop calibration models for predicting soil properties like bulk density and available nitrogen from spectral signatures.

Strengths

• Data includes 372 soil samples with a defined 70/30 split for calibration and validation.
• Study tested five distinct multivariate modeling techniques and reported their comparative robustness.
• Prediction models achieved a Ratio of Performance to Deviation (RPD) ranging from 1.48 to 2.06 for seven soil properties.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the underlying spectral data table is unknown, which may limit suitability assessment.
• Data may reflect geographic bias inherent to its collection from a specific coastal region of India.

Provenance

Source

Central Coastal Agricultural Research Institute

Collection Method

Hyperspectral reflectance measurement of processed soil samples combined with laboratory-estimated soil properties.

Geography

17 coastal districts from the west coast region of India