Proceedings of the 2022 Mississippi Water Resources Conference

Year: 2022 Authors: Senyurek V., Farhad M.M., Gurbuz A.C., Kurum M.

Soil moisture (SM) is one of the essential factors for agricultural science and related studies. Low or saturated SM negatively affects crop growth. Traditionally SM content is measured in-situ using SM probes. Although in-situ measurements provide accurate and high temporal SM information in a single point, it is costly, inefficient, and time-consuming for high spatial resolution. Spaceborne remote sensing platforms are practical for a global scale. However, their low spatial resolution (10-40 km) and low temporal resolution (2-3 days) are too coarse for many site-specific precision agriculture applications.

This study proposes a low-cost and practical technique based on Global Navigation Satellite Systems-reflectometry (GNSS-R) utilizing a small drone. A random forest machine algorithm is used to develop an SM estimation model that uses reflected GPS signals, vegetation indices from a multispectral camera, and crop size information from LIDAR as input features. The field experiment was conducted on a 2.48 ha corn and cotton field at MSU's heavily instrumented North Farm. The needed data for the machine learning model was acquired from April to October 2021.

The study results showed that the 2.48 ha field could be covered by GPS reflectometry with about 12 mins flight time, and SM can be mapped with 5m x 5m spatial resolution with high accuracy. We will present the data collection, machine learning based soil moisture estimation approach and the obtained results over the test area.