Speaker
Description
Physiological reactions of plants are used in bioindication as early warning indicators of environmental stress. As part of a larger project aimed at developing repeatable remote sensing methodologies for monitoring crop performance and identifying environmental stressors relevant to precision agriculture and environmental surveillance, this work presents a quantitative approach based on drone-assisted thermography (UAV) using maize (Zea mays L.) as a bioindicator. To test the ability to detect stress signatures from drone observations, two experimental tests were developed: (i) an open-field agronomic scenario in which different fertilizations were considered, which can induce different canopy physiology, and (ii) a controlled environmental monitoring scenario (in tanks) in which maize was grown under soil-contaminated conditions (i.e., heavy metals Pb, Zn, Cr, and PAHs). Thermal data were collected through two flights: one with a Foxtech quadcopter equipped with a long-wave infrared (LWIR) sensor (MicaSense Altum), and one with an additional thermal camera on the DJI Mavic 3 Thermal drone. These flights were also used to cross-validate the sensors used. The Micasense Altum sensor also acquired multispectral images for spectral context (VIS–Red Edge–NIR), allowing for a joint interpretation of (a) functional thermal responses related to transpiration and canopy energy balance, and (b) structural responses from spectral indices. An end-to-end workflow was applied, from flight planning, through the necessary radiometric corrections and generation of the orthomosaic from which multispectral indices and thermal maps were extracted, to the extraction of canopy temperature statistics for the various scenarios considered. The results demonstrate that UAV thermography provides a solid functional contribution to bioindication, integrating spectral indices related to pigment content, chlorophyll content, and canopy vigor. The UAV-derived thermal and multispectral data were validated using ground-based measurements, confirming the reliability of the workflow in detecting maize stress. To quantify the importance of thermal information in identifying different maize stress conditions, several machine learning algorithms and related ablations studies were tested. The main limitation of using this data, compared to multispectral data, is the lower native LWIR spatial resolution, which increases mixed pixel effects (different land cover) and can alter canopy temperatures where cover varies between treatments. Conservative vegetation masking and boundary erosion mitigated this bias. Overall, the integration of UAV thermography and multispectral imaging can be effectively used, thanks to the concept of bioindication, both for agronomic optimization processes and for contamination monitoring.
