Uncrewed Aircraft Systems (UAS)-based detection of organic carbon with hyper- and multispectral sensors in cropland soils

The quantification and monitoring of soil organic carbon (SOC) in croplands is crucial in a context of effective and precise agricultural management. Conventional methods to assess SOC changes pose some limitations on time and labor consumption to efficiently assess SOC distributions and dynamics in a spatial and temporal context. Recent developments in Uncrewed Aircraft Systems (UAS), combined with miniaturized visible-near infrared sensors bring opportunities for the rapid and low-cost field-scale SOC mapping. It shows the potential of UAS-soil sensing to bridge the gap between high-resolution plot-scale sensing at centimetre-level and moderate-resolution landscape-scale sensing at meter-level from airborne or satellite platforms. For the UAS sensing system, variety of sensor types enable the multiple dimensional monitoring (i.e., temporal, spatial and spectral) of soil properties, while it brings large uncertainties in choosing a desirable sensor specification for a monitoring purpose. Accordingly, it lacks a capability evaluation for different sensor types in the estimation of soil properties. Meanwhile, there is still room for improving the pre-processing approaches against the perturbing factors such as soil roughness and anisotropy. A comprehensive understanding of how these perturbing factors affecting soil reflectance spectra under different measurement conditions will assist in the use of SOC prediction models derived from an existing regional soil spectral library (SSL), which can be applied to UAS-based hyperspectral data to reduce the need for additional field sampling. The overall aim of this thesis was therefore to improve the spatial and spectral accuracy and effectiveness of UAS-based multi- and hyper-spectral measurement for the estimation of SOC content in croplands. In this thesis, we started with assessing the potential of a non-imaging spectrometer on UAS-borne SOC estimation. We showed that local calibration with soil sampling had good capability for SOC content prediction (RPIQ = 2.35, RMSE = 0.57 g kg-1). Meanwhile, we developed a transfer function to align UAS-based measurement with laboratory-based measurement to enable the use of an existing model based on a soil spectral library. In a second phase, we focused on imaging sensors to have spatial and structural information of the soil surface. We first investigated the effect of camera properties and georeferencing methods on the quality of imagery mosaicking and 3D reconstruction using Structure from Motion (SfM) algorithms. The result demonstrated the robustness and repeatability of image mosaicking with precise georeferencing and accurate camera calibration based on post-processed Global Navigation Satellite System (GNSS) data. Next, we proposed and assessed the pre-processing approaches for UAS-based multispectral images to reduce the error caused by soil roughness and anisotropy, and optimized the workflow of SOC mapping in bare croplands. Furthermore, we characterized soil surface roughness and illumination geometry using UAS-based imaging spectroscopy, and investigated their influence on soil reflectance measurement and SOC estimation in bare croplands. These studies help to understand the capabilities and limitations of UAS-based spectroscopy in soil mapping and monitoring in croplands, and the potential of improving its accuracy and efficiency in SOC estimation by comprehensively considering perturbing factors and mechanisms behind it.