An automated deep learning convolutional neural network algorithm applied for soil salinity distribution mapping in...

Area of Science:

• Environmental Science
• Remote Sensing
• Data Science

Background:

• Traditional soil salinity studies are labor-intensive and costly, limiting large-scale assessments.
• Remote sensing offers a viable alternative for frequent and extensive soil salinity monitoring.

Purpose of the Study:

• To develop and validate an innovative deep learning convolutional neural network (DL-CNN) approach for soil salinity mapping.
• To assess the effectiveness of multi-spectral remote sensing data (Landsat series) for soil salinity detection.
• To compare the performance of the DL-CNN model against traditional remote sensing soil salinity indices.

Main Methods:

• Acquisition of Landsat 7 ETM+ and 8 OLI images for multiple years (2005, 2010, 2015, 2019).
• Collection and utilization of 704 soil surface samples for training (70%) and validation (30%) of the DL-CNN model.
• Training the DL-CNN model using remote sensing-derived variables (LST, SM, evapotranspiration) and geospatial data (NDVI, landuse).
• Employing ReLu, Cross-entropy, and ADAM as activation, loss, and optimizer functions, respectively.
• Applying Frequency Ratio (FR) and Weight of Evidence (WOE) models for geospatial assessment of classification results.

Main Results:

• The DL-CNN model achieved high overall accuracies (OA) ranging from 0.9399 to 0.9500 across the study years.
• The automated DL-CNN approach demonstrated superior performance compared to conventional remote sensing soil salinity indices.
• Landuse, Land Surface Temperature (LST), and NDVI were identified as key influential factors for soil salinity using FR and WOE models.

Conclusions:

• The proposed DL-CNN method provides an accurate, efficient, and automated solution for soil salinity mapping.
• Remote sensing data, combined with DL-CNN, enables frequent and large-scale soil salinity assessments.
• The methodology is recommended for spatial modeling of soil salinity in similar environmental conditions.