Empirical models on urban surface emissivity retrieval based on different spectral response functions: A field study

Thermal emissivity is a prerequisite for retrieving the land surface temperature (LST) and estimating the land surface budget based on data from remote sensing. Despite the availability of empirical models on emissivity retrieval on the meso-scales, their incompatibility for complex surfaces of the micro-scale leads to errors in emissivity retrieval, thus compromising thermal assessments of urban environments. To minimize such errors, this paper proposed a method to retrieve pixel-scale emissivity on the micro-scale. To measure reflectance spectra and emissivity spectra in the field, a PSR+3500 handheld spectrometer and a 102F portable Fourier transform infrared spectrometer were respectively used. Upon resampling these spectra to spectral response functions (SRF) of drone-derived and satellite-derived sensors, diverse reflectance and emissivity values were obtained to establish empirical models characterizing correlations between the normalized difference vegetation index (NDVI) and emissivity. Then, these models were applied to the low-altitude hyperspectral image from the Nano-Hyperspec imager on a drone. The results show that the model established by the SRF of a thermal camera achieved a root mean square error (RMSE) of 0.0129, and the accuracy of emissivity retrieval was within 0.010. For satellite applications, the model founded by the SRF of Aster was the most accurate for retrieving the emissivity of urban surfaces, with a RMSE of 0.0082 and an average accuracy of 0.003. The model based on SRFs of Landsat 8 registered a RMSE of 0.0155 alongside an average accuracy of 0.012, while that based on Modis registered a RMSE of 0.1210, alongside an average accuracy of 0.007.

Automated summary

This paper proposed a method to retrieve pixel-scale emissivity on the micro-scale by measuring reflectance spectra and emissivity spectra in the field using spectrometers. Empirical models characterizing correlations between the normalized difference vegetation index (NDVI) and emissivity were established and applied to low-altitude hyperspectral images from a drone. The accuracy of emissivity retrieval for different sensors was assessed, with the model based on the SRF of Aster found to be the most accurate.