The Land Surface Temperature (LST) processing includes a split-window method, using radiances from two channels, the band centres of which are close in wavelength, to determine the effective radiometric temperature of the Earth's surface "skin" in the instrument field of view ("skin" refers to the top surface in bare soil conditions and to the effective emitting temperature of vegetation canopies as determined from a view of the top of a canopy). This method assumes that the linearity of the relationship between LST and BT results from linearisation of the Planck function and linearity of the variation of atmospheric transmittance with column water vapour amount.
The algorithm is soundly based on radiative transfer theory as applied to the exchange of radiation between the surface and atmosphere. The effects of land surface emissivity are implicitly taken into account in these algorithms via biome and fractional vegetation. The basic algorithm may be stated as:
where a0, b0 and c0 are classes of coefficients that depend on atmospheric water vapour, satellite viewing angle and land surface emissivity. T11 and T12 represent the brightness temperatures measured at 11 µm and 12 µm respectively.
The purpose of this algorithm is to derive, for each pixel, the appropriate indices associated with biome, fractional vegetation and atmospheric water vapour. Using these indices, the three coefficients can be derived and the LST can be calculated.
Accounting for all of the complex effects introduced by surface heterogeneity, shadowing, terrain variability (e.g. height variations) and atmospheric variability may be achieved under special experimental conditions, but will not be possible for global conditions. Consequently the approach adopted is to determine robust regression coefficients that can be used for classes of land cover conditions, atmospheric water vapour loadings and seasons. To include the effects of seasonal vegetation growth, coefficients are linearly combined and a time-dependency of both coefficients and auxiliary data is included. Special coefficients are used for snow and ice covered surfaces and lakes.
A specific probabilistic cloud masking module over Land pixels is implemented.
The following figure describes the successive processing sub-steps. Note that only major sub-steps are fully described.