Relative radiometric normalization of H-res multi-temporal thermal infrared (TIR) flight-lines of a complex urban scene

Useful biophysical information such as surface temperature and surface energy flux provided by thermal infrared (TIR) remote sensing sensors are commonly used for studying urban temperature variations and urban heat islands. However, an important limitation of TIR imagery is the influence of local microclimatic variability (i.e., wind, precipitation and humidity) on sensor observations. This can cause the same scene components (e.g., roads and buildings) to exhibit different thermal states (i.e., temperatures) when exposed to varying microclimate conditions. In the case of airborne TIR imagery, the ambient sensed temperature also naturally changes between flight line acquisitions, resulting in an image mosaic with different temperatures for the same scene components, making detailed analysis non-trivial. In an effort to mitigate this problem and to produce a ‘seamless’ TIR scene mosaic, we evaluate three different relative radiometric normalization methods on two adjacent flight-lines of TABI1800 data (each ~35km x 0.9km, at 50 cm spatial resolution, and 0.05 °C thermal resolution) that were acquired in May 2012 over The City of Calgary, Alberta, Canada. We then describe their effects on the resulting mosaic. The evaluated methods include: (i) Histogram Matching, (ii) Linear Regression based on Pseudo Invariant Features (PIF), and (iii) Theil-Sen Regression based on PIF. Based on visual assessment, results show that Histogram Matching produces is the best. Such radiometric normalization (i) increases the visual agreement between the thermal airborne flight lines, (ii) produces a seamless mosaic, (iii) improves radiometric agreement, (iv) improves hot-spot detection, and (v) provides accurate data for thermal-based energy models. Background and Relevance While thermal remote sensing provides important biophysical information (i.e., temperature and surface energy flux) about the earth’s surface, the applicability of these data still remains challenging due to difficulty in calibration and appropriately identifying atmospheric attenuation (Quattrochi and Luvall, 1999). In the case of urban surfaces, additional challenges are imposed by the composite and heterogeneous nature of the surface itself, as well as the surrounding environment (Voogt and Oak, 1997). Therefore, to use thermal remote sensing data to accurately measure urban thermal characteristics, it is necessary to thoroughly understand the limitations of such data. Furthermore, current satellite platforms only acquire moderate to low resolution (60 m to 1 km) thermal imagery that are not suitable for detailed thermal mapping of urban surfaces. As a result, airborne TIR imagery are increasingly used for urban mapping exercises (Hay et al., 2011; Weng, 2011). However, to thermally mapping large urban areas at a high spatial resolution (~1m), airborne imagery need to be acquired in numerous flight lines and mosaiced together; which induces geometric and radiometric variations between flight paths (Rahman et al., 2012). Due to these radiometric differences, the same class of scene objects tend to exhibit different spectral characteristics within a single mosaic, making image analysis and classification difficult (Tuominen and Perkkarinen, 2004). In an effort to reduce similar concerns, relative radiometric normalization techniques have been used for decades to normalize multitemporal multispectral remote sensing data (Salvaggio, 1993; Hall et al., 1991; Schott et al., 1988). However, the applicability of these techniques on multitemporal thermal datasets has yet to be adequately assessed. With the increased demands for thermal data by the remote sensing community (Hay et al., 2011), we recognize an emerging need to evaluate the applicability of relative radiometric normalization techniques on high spatial resolution (H-res) TIR imagery. Based on these ideas, this paper focuses on the evaluation of three different relative radiometric normalization techniques applied to a H-res TABI-1800 (Thermal Airborne Broadband Imager) dataset collected over a portion of the City of Calgary in May 2012.