Performance of a thermal-infrared radiosity and heat-diffusion model for estimating sub-pixel radiant temperatures over the course of a day

Temperature/emissivity estimation from remotely measured radiances generally assumes that scene elements represented by pixels in fact have a single emissivity spectrum and are isothermal. Thus, estimated temperatures and emissivities are the effective values that would be found if these simplified assumptions were met. In reality, the physical scene is neither homogeneous nor isothermal, and the effective values are not strictly representative of it. How much in error are they? In order to estimate this, a radiosity model used for predicting effective emissivity spectra and radiant temperatures for rough surfaces has been developed. Here we compare the computer model results to analytic model results in order to verify that the computer model is working properly. We validate the model against spectra measured in the field using a hyperspectral imaging spectrometer and a centimeter-scale DTM of the test scene acquired using a tripod-based LiDAR. The discrepancies between analytical and modeled values are less than 0.01%. Modeled emissivity spectra deviate from the measured by no more than 0.015 emissivity units. Modeled kinetic temperature on average deviates from measured by less than 1 K over the course of a day.

► Radiosity model is essential for predicting sub-pixel roughness effects in TIR images. ► 3-D radiosity model adapted for TIR coupled with a 1-D (downward) heat-diffusion model. ► Model performance verification and validation using hyperspectral TIR images. ► Model results for the set of natural surfaces.