Identification of soil type in Pakistan using remote sensing and machine learning

Area of Science:

• Agricultural Science
• Remote Sensing
• Soil Science

Background:

• Accurate soil type identification is crucial for crop productivity.
• Conventional methods are time-consuming, costly, and require expert analysis.
• Digital soil mapping offers rapid, low-cost, high-resolution soil information.

Purpose of the Study:

• To develop and evaluate a model for identifying soil types using remote sensing data.
• To compare the performance of machine learning algorithms for digital soil mapping.
• To provide data for improved agricultural planning and crop cultivation.

Main Methods:

• Acquisition of Landsat 8 spectral data from the Upper Indus Plain, Pakistan (June-August 2020).
• Identification of soil types (Silt Loam, Loam, Sandy Loam, Silty Clay Loam, Clay Loam) using bare soil images.
• Utilized spectral data, reflectance values, and vegetation indices.
• Applied Random Forest, Support Vector Machine, and Logistic Model Tree with 10-fold cross-validation for classification.

Main Results:

• Random Forest achieved an overall accuracy of 86.61% for soil type classification.
• Logistic Model Tree achieved an overall accuracy of 84.41%.
• Spectral data proved effective for multi-class soil type classification.

Conclusions:

• Remote sensing-based digital soil mapping is a viable and accurate method for soil type identification.
• Machine learning algorithms, particularly Random Forest, show strong performance in classifying soil types.
• The findings support enhanced agricultural planning and decision-making through precise soil information.