A comparison of four relative radiometric normalization (RRN) techniques for mosaicing H-res multi-temporal thermal infrared (TIR) flight-lines of a complex urban scene

High-spatial and -radiometric resolution (H-res) thermal infrared (TIR) airborne imagery, such as the TABI-1800 (Thermal Airborne Broadband Imager) provide unique surface temperature information that can be used for urban heat loss mapping, heat island analysis, and landcover classifications. For mapping large urban areas at a high-spatial resolution (i.e., sub-meter), airborne thermal imagery needs to be acquired over a number of flight-lines and mosaiced together. However, due to radiometric variations between flight-lines the similar objects tend to have different temperature characteristics on the mosaicked image, resulting in reduced visual and radiometric agreement between the flight-lines composing the final mosaiced output.To reduce radiometric variability between airborne TIR flight-lines, with a view to produce a visually seamless TIR image mosaic, we evaluate four relative radiometric normalization techniques including: (i) Histogram Matching, (ii) Pseudo Invariant Feature (PIF) Based Linear Regression, (iii) PIF-Based Theil-Sen Regression, and (iv) No-Change Stratified Random Samples (NCSRS) Based Linear Regression. The techniques are evaluated on two adjacent TABI-1800 airborne flight-lines (each ∼30 km × 0.9 km) collected ∼25 min apart over a portion of The City of Calgary (with ∼30% overlap between them). The performances of these techniques are compared based on four criteria: (i) speed of computation, (ii) ability to automate, (iii) visual assessment, and (iv) statistical analysis. Results show that NCSRS-Based Linear Regression produces the best overall results closely followed by Histogram Matching. Specifically, these two radiometric normalization techniques: (i) increase the visual and statistical agreement between the tested TIR airborne flight-lines (NCSRS Based Linear Regression increases radiometric agreement between flight-lines by 53.3% and Histogram Matching by 52.4%), (ii) produce a visually seamless image mosaic, and (iii) can be rapidly automated within an operational multi-flight-line, multi-temporal mosaic workflow.