Shadow fading prediction at 18 GHz through physics guided learning in vegetative corridors.

Area of Science:

• Agricultural Engineering
• Wireless Communications
• Machine Learning

Background:

• Precision agriculture relies on robust wireless connectivity.
• Existing radio propagation models struggle with variable orchard geometries.
• Accurate models are essential for reliable sensor networks and automation.

Purpose of the Study:

• To develop and validate an improved radio propagation model for precision agriculture.
• To assess the impact of orchard corridor geometry on wireless signal propagation at 18 GHz.
• To compare a hybrid physics-ML model against traditional models.

Main Methods:

• Extensive radio propagation measurements (N=17,269) in a custard apple orchard.
• Systematic variation of corridor width and transmitter height across nine configurations.
• Development of a Hybrid Linear+XGBoost framework combining physical insights with machine learning.

Main Results:

• The standard Close-In (CI) model showed significant geometry-dependent variability (RMSE=3.91 dB).
• The hybrid model achieved a 24.0% improvement (RMSE=2.97 dB) over the CI model.
• The hybrid model demonstrated superior generalization to unseen geometric configurations compared to pure ML methods.

Conclusions:

• Orchard geometry, particularly corridor width, is a critical factor in radio propagation and shadow fading.
• Explicit incorporation of geometric parameters is necessary for accurate channel models in agriculture.
• Hybrid physics-ML architectures offer robust generalization for wireless channel modeling in complex environments.