Surface temperature and emissivity separability over land surface from combined TIR and SWIR AVHRR data

This paper presents a method to recover land surface emissivity and temperature from radiance measurements without a priori assumptions on these parameters. The model combines spectral radiances in the short wave infrared (SWIR) domain (such as AVHRR channel 3: 3.55-3.93 μm) and the thermal infrared (TIR) domain (such as AVHRR channels 4: 10.5-11.5 μm or 5: 11.5-12.5 μm). The model assumes that the data have been previously accurately corrected for atmospheric effects and that emissivity separation from temperature can be decoupled from the atmospheric correction procedure. The approach is based on the estimation of the reflected component of the SWIR channel radiance after computing its thermal emitted component derived from a TIR channel brightness temperature. A correction factor is introduced to account for the emissivity difference between the spectral channels, and the authors propose two methods for estimating it. In the first method, referred to as the TS-RAM model, this factor is estimated using a linear regression model from the ratio of the atmospherically corrected radiances of the SWIR and TIR channels. The regression model is established from theoretical simulations computed with the acquisition conditions and using a reference emissivity database. In the second method (the Δday model), the correction factor is estimated from consecutive data sets acquired the same day, assuming that the emissivity remains constant between the two acquisition times. Knowing the surface emissivity in the SWIR channel, the land surface temperature (LST) and other channel emissivities can then be easily retrieved. The results obtained from simulations in the context of AVHRR data show that rms errors for the TS-RAM model are around ±0.005 for channel 3 emissivity, ±0,01 for channels 4 and 5 emissivities and ±0.5 K for surface temperature. The Δday model performs better, with rms errors of ±0.002 and ±0.3 K, respectively, for the channel 3 emissivity and LST as the basic model assumption is held. The performance and sensitivity of the model were assessed for a wide range of surface types and ground thermal conditions as well as for different measurement conditions, i.e., viewing and solar zenith angles and atmospheric conditions. This assessment shows that an uncertainty of ±0.5 g cm^-2 in atmospheric integrated water vapor content, under standard thermal conditions (280 K⩽LST⩽300 K), will introduce errors of the order of 2.5% in SWIR emissivity and 2.5 K in LST. The authors suggest that the model be applied to data acquired in near-nadir conditions (view angles up to 30°) to reduce the influence of directional effects. An application using some selected cloudless NOAA-AVHRR 10 and 11 images over Quebec shows that the proposed approach is promising for characterizing heterogeneous land surfaces