An optimization algorithm for separating land surface temperature and emissivity from multispectral thermal infrared imagery

Land surface temperature (LST) and emissivity are important components of land surface modeling and applications. The only practical means of obtaining LST at spatial and temporal resolutions appropriate for most modeling applications is through remote sensing. While the popular split-window method has been widely used to estimate LST, it requires known emissivity values. Multispectral thermal infrared imagery provides us with an excellent opportunity to estimate both LST and emissivity simultaneously, but the difficulty is that a single multispectral thermal measurement with N bands presents N equations in N+1 unknowns (N spectral emissivities and LST). In this study, we developed a general algorithm that can separate land surface emissivity and LST from any multispectral thermal imagery, such as moderate-resolution imaging spectroradiometer (MODIS) and advanced spaceborne thermal emission and reflection radiometer (ASTER) data. The central idea was to establish empirical constraints, and regularization methods were used to estimate both emissivity and LST through an optimization algorithm. It allows us to incorporate any prior knowledge in a formal way, The numerical experiments showed that this algorithm is very effective (more than 43.4% inversion results differed from the actual LST within 0.5/spl deg/, 70.2% within 1/spl deg/ and 84% within 1.5/spl deg/), although improvements are still needed.