Advancing Soil Moisture Mapping with Multi-Modal Drone Sensing and AI: Integrating GPR, Hyperspectral, and Thermal Imaging for Precision Agriculture

Abstract

Accurate soil moisture estimation at different depths is essential for sustainable agriculture, helping improve irrigation efficiency, crop productivity, and water conservation. However, conventional monitoring methods often struggle to provide reliable, scalable, and non-invasive soil moisture information, especially in vegetated areas or at deeper soil layers. This research addresses these challenges by developing advanced soil moisture estimation approaches that combine drone-mounted sensors with machine learning and deep learning techniques. The goal of this work was to integrate data from multiple sensors, including hyperspectral, RGB-thermal, and Ground Penetrating Radar (GPR), to improve depth-specific soil moisture estimation in cornfields. By combining surface, canopy, and subsurface information, this research aimed to capture a more complete picture of soil moisture dynamics throughout the growing season. In the first part of the research, the study used RGB-thermal imagery to estimate soil moisture across various depths. By combining structural canopy information and Land Surface Temperature (LST), a canopy-informed model was developed to track moisture variability throughout different growth stages. This approach proved especially effective for shallow soil layers where canopy development is closely linked to root zone moisture conditions. The second part of study explored hyperspectral imaging as a tool for monitoring plant canopy characteristics and estimating soil moisture in the root zone. By analyzing canopy reflectance and identifying key wavelengths linked to soil moisture, machine learning models were developed to predict moisture conditions at different depths. These findings highlighted the potential of hyperspectral data as an indirect indicator of root zone moisture, particularly under water stress. Recognizing the limitations of optical data for deeper soil moisture estimation, the research then integrated GPR data with RGB-thermal imagery. GPR provides subsurface moisture information that complements canopy and surface observations. Using one-dimensional Convolutional Neural Networks (1D-CNN), meaningful patterns were extracted from GPR signals, improving moisture estimation at both shallow and deeper soil layers. The research also introduced a deep fusion approach, combining hyperspectral imaging and GPR data to better capture interactions between canopy and soil conditions. This integration reduced estimation errors and improved model reliability across varying field conditions. Finally, GPR data alone, processed through 1D-CNN and Artificial Neural Network (ANN) models, provided robust soil moisture estimation at multiple depths, even under dense vegetation and in challenging environments. Together, these studies demonstrate how combining multi-sensor data with advanced modeling techniques significantly improves soil moisture estimation. This work offers practical tools for precision agriculture, supporting more efficient irrigation management and contributing to sustainable farming in the face of climate change and water scarcity. In conclusion, this research presents a significant step forward in the field of soil moisture monitoring, demonstrating the potential of multi-modal sensing and advanced machine learning for achieving accurate, depth-specific soil moisture estimation. The integrated methodologies developed through this work offer valuable insights for researchers, practitioners, and policymakers seeking to optimize irrigation management, enhance agricultural sustainability, and address the challenges posed by global water scarcity.