Objective

This project aimed to model and visualize the spatial variation of soil pH using DeepKriging, a deep learning-enhanced geostatistical method. The goal was to support site-specific soil management by identifying patterns in soil acidity and the environmental factors that influence it.

Tools and Technologies Used

• Languages: Python
• Libraries: Pandas, Matplotlib, TensorFlow
• Other: Microsoft Excel (data prep and exploration)

Process and Methodology

• Collected soil pH data with geographic coordinates from field measurements and prepared it in Excel.
• Performed visual exploratory data analysis using scatter plots and density maps.
• Built and trained a DeepKriging model using TensorFlow to combine deep neural networks with traditional Kriging spatial interpolation.
• Compared results against baseline deep neural networks and traditional Kriging methods.
• Evaluated performance using Mean Squared Error (MSE) and Mean Absolute Error (MAE).

Key Findings and Insights

  fig1:

• DeepKriging outperformed both baseline DNNs and traditional Kriging, minimizing prediction error significantly.
• Figure 1 showed strong spatial heterogeneity in pH levels — regions with concentrated pH values (e.g., around latitude 40.15°N and longitude 21.38°–21.40°E) revealed unique soil characteristics likely influenced by vegetation, drainage, or human activity.

• Figure 2 visualized soil pH distribution using a scatter plot where dot size and color represented pH levels:

  Larger, darker dots = higher pH (alkaline)

  Smaller, lighter dots = lower pH (acidic)

• Clusters of alkaline soil were found in the northern and southwestern regions, indicating fertile zones favorable for agricultural use.