1 A physically - constrained calibration database for 2 Land Surface Temperature using infrared retrieval 3 algorithms 4

2 Instituto Dom Luiz, University of Lisbon, IDL, Campo Grande, Ed C1, 1749-016 Lisbon, Portugal; 8 E-Mails: [email protected] (V.B.); [email protected] (C.C.) 9 * Correspondence: [email protected]; Tel.: +351 21 844 7055, Ext: 1555 10 11 12 Abstract: Land Surface Temperature (LST) is routinely retrieved from remote sensing 13 instruments using semi-empirical relationships between top of atmosphere (TOA) radiances and 14 LST, using ancillary data such as total column water vapor or emissivity. These algorithms are 15 calibrated using a set of forward radiative transfer simulations that return the TOA radiances given 16 the LST and the thermodynamic profiles. The simulations are done in order to cover a wide range of 17 surface and atmospheric conditions and viewing geometries. This work analyses calibration 18 strategies, considering some of the most critical factors that need to be taken into account when 19 building a calibration dataset, covering the full dynamic range of relevant variables. A sensitivity 20 analysis of split-windows and single channel algorithms revealed that selecting a set of atmospheric 21 profiles that spans the full range of surface temperatures and total column water vapor 22 combinations that are physically possible seems beneficial for the quality of the regression model. 23 However, the calibration is extremely sensitive to the low-level structure of the atmosphere 24 indicating that the presence of atmospheric boundary layer features such as temperature inversions 25 or strong vertical gradients of thermodynamic properties may affect LST retrievals in a non-trivial 26 way. This article describes the criteria established in the EUMETSAT Land Surface Analysis – 27 Satellite Application Facility to calibrate its LST algorithms applied both for current and forthcoming 28 sensors. 29