Best practices for processing thermal UAV imagery from DJI platforms: limitations of JPEG-encoded temperature data and implications for mosaicking

Commercial DJI UAVs, , democratize thermal remote sensing , yet their proprietary data formats present significant challenges for scientific applications. Standard photogrammetric software treats DJI’s radiometric JPEGs (R-JPEGs) merely as visual images, losing quantifiable temperature data. This study defines processing best practices by evaluating these proprietary limitations, the critical role of meteorological metadata, and the computational consistency across different software environments. We compared two DJI Thermal SDK conversion workflows : one utilizing immutable metadata fixed at acquisition and another allowing retrospective adjustment of environmental parameters. Using a DJI Mavic 3Twe tested workflow resilience against incorrect flight parameters. The results reveal that while both workflows provide consistent data when correctly parameterized with in-situ meteorological metadata (mean difference < 0.4 °C), the inability to correct metadata post-flight results in substantial systematic errors, resulting in a 4.7 °C temperature discrepancy. We also highlight software-specific technical constraints related to 16-bit data scaling and potential signal clipping by ThermoConverter in high-temperature scenes. Consequently, we demonstrate that direct mosaicking of R-JPEGs is scientifically invalid. To ensure radiometric integrity, SDK-based conversion to TIFFs must be utilized, prioritizing tools that enable the post-hoc harmonization of emissivity and atmospheric variables for generating reliable thermal orthomosaics.