Remote sensing of crop parameters using UAV-based multispectral imaging and radiative transfer models

Abstract

In the face of climate change and a growing global population, it's important to increase the output of agricultural systems and improve crop resilience to challenging environmental conditions. Achieving these objectives requires monitoring crop health in the field, as well as breeding more resilient crop varieties. However, these tasks are labor-intensive and not economically sustainable. Remote sensing tools, such as uncrewed aerial vehicles (UAVs), have demonstrated their potential to successfully measure crop parameters while minimizing the need for human and financial resources. <br/> <br/> The mapping of crop parameters has been widely studied in both the satellite and UAV research communities. The UAV community typically uses data-driven and parametric models to predict crop parameters, while satellites rely on physical models called radiative transfer models (RTMs) to retrieve these variables. However a key drawback of using satellites is their low spatial and temporal resolution for applications in precision agriculture. <br/> <br/> This thesis explores the use of radiative transfer model PROSAIL to retrieve crop variables with UAVs and multispectral imaging. First, we examine the reflectance calibration workflows of the optical sensor, vital for time-series image analysis. We propose a multi-panel approach for calibrating reflectance of a multispectral sensor, which our analysis shown to perform better than the one-point calibration. Next we address the challenges of retrieving structural and biochemical variables in complex and homogeneous crop canopies. Our findings confirm that the higher spatial resolution provided by UAVs doesn't disrupt the fundamental assumptions of the PROSAIL, which was originally developed for simpler canopies. Finally, we investigate the sensor synergies for crop stress detection. In one study we explore the synergy between terrestrial laser scanner, multispectral imaging, and RTMs to track drought-induced leaf movement in soybean. We show that it's possible to track leaf orientation using just multispectral cameras. Another study discusses the challenges associated with using multiple sensors together to detect crop stress.