Challenges and limitations of thermal infrared remote sensing with unoccupied aerial systems

Abstract

Unoccupied aerial systems (UAS) thermal infrared (TIR) remote sensing is emerging as a novel alternative to satellite remote sensing across many scientific applications, due to its unique advantages for spatial and temporal resolution. However, much is still unknown regarding its utility for geothermal resource assessment. In particular, the nascence of this technology means that technique development, especially for individual camera models, is still needed to generate reliable thermal data. Therefore, to evaluate whether UAS TIR remote sensing is a viable option for geothermal reconnaissance, the objectives of this study included: 1) assessing the accuracy of the FLIR Duo Pro R radiometric thermal camera, 2) testing the ability of various photogrammetric workflows to preserve the thermal signature of geothermal anomalies, 3) modeling the most influential explanatory variables of remotely- sensed temperatures using a machine learning (ML) model in ArcGIS Pro, and 4) determining whether our ML model can predict the thermal variation in our remotely sensed data, based on a curated set of explanatory variables. To accomplish these objectives, this study employed a three-phased approach: lab experiments, field surveying, and computational modeling. The results from our lab experiments indicate that the FLIR Duo Pro R overestimates stable temperatures at varying degrees (6-7 °C) when it is operating at steady- state conditions. Our results also suggest that it exhibits measurement instability when exposed to simulated changes in environmental conditions, i.e. wind and heat. For instance, our wind experiment found that the camera underestimated the blackbody temperature by 8 °C when exposed to wind and jumped nearly 16 °C after the transition from wind to steady- state conditions. We also found that photogrammetric techniques strongly influence the signature (presence and magnitude) of thermal anomalies associated with geothermal system in UAS TIR imagery. Lastly, our ML model cannot confidently identify the most influential explanatory variables of remotely-sensed thermal anomalies or reliably predict remotely- sensed ground temperature based on explanatory variables. We conclude that the current limitations associated with UAS TIR remote sensing hinder its ease-of-use for geothermal applications, especially where thermal anomalies are subtle and require accurate temperature readings to differentiate them from their surroundings. Despite the challenges discussed in this paper, these techniques may still serve as a tool for assessing the spatial distribution of thermal properties of geothermal systems if future work continues to improve and develop these methodologies. Most notably, this technique will benefit from advancements regarding the temperature drift of the sensor during image acquisition and the photogrammetric processing of thermal imagery.