Soil moisture monitoring by a drone-borne synthetic aperture radar

Author(s)

Marlon de Souza Alcântara

• Document type: Master's Dissertation
• Press: Campinas, SP.
• Institution: Universidade Estadual de Campinas (UNICAMP). Faculdade de Engenharia Elétrica e de Computação
• Defense date: 2020-12-18

Examining board members

Hugo Enrique Hernández Figueroa; Antônio Pires de Camargo; Lucas Heitzmann Gabrielli

• Advisor: Hugo Enrique Hernández Figueroa; Barbara Janet Teruel Mederos

Abstract

The use of high-resolution aerial images acquired using drone-borne optical, multispectral or thermal sensors for crop development and soil conditions monitoring has become very popular in precision agriculture. The monitoring of soil moisture using these sensors become difficult as crop develops. This limitation can be overcome with the use of images acquired by synthetic aperture radars, which have the ability to penetrate the plantation. In this work, soil moisture is monitored throughout the sugarcane development cycle, using high resolution images acquired with a drone-borne SAR system, that operates in the P-, L- and C-bands. For this purpose, SAR data acquisitions, soil moisture and biometric measurements of sugarcane were carried out in an experimental area located at FEAGRI¿Unicamp. From these data, two cases were explored: on bare soil and with the presence of sugarcane. For the case of bare soil, an empirical model was developed based on linear regression for each operating frequency. In addition, an artificial neural network was trained with synthetic data generated with Dubois semi-empirical model, using the three operating frequencies, and later validated with real measurements. To obtain soil moisture in the sugarcane crop, an empirical model based on artificial neural network was developed using data measured in the field, such as sugarcane height and soil moisture and SAR images acquired with the P and L bands. Results demonstrate the high sensitivity of the microwave signal to changes in soil moisture without the presence of vegetation, reaching an RMSE of 5.20, 3.60 and 4.54 vol.%, using empirical models developed with the P-, L- and C bands, respectively. The developed neural network with Dubois model presented a RMSE of 5.70 vol.%, however with a poor correlation. For the case with the presence of sugarcane, a greater signal penetration capacity was observed using the P and L band, making it possible to obtain soil moisture with a RMSE better than 10.0 vol.%, in all phenological stages of the sugarcane (AU)

• FAPESP's process: 18/12726-0 - Drone-borne radar for sugar cane precision agriculture
• Grantee: Marlon de Souza Alcântara
• Support Opportunities: Scholarships in Brazil - Master