SUBPIXEL TEMPERATURE ESTIMATION FROM LOW RESOLUTION THERMAL INFRARED REMOTE SENSING C. Ottlé 1 , A. Kallel 1 , G. Monteil 1 , S. LeHégarat 2 , B. Coudert 3 (1) LSCE/IPSL, CEA-Orme des Merisiers, 91191 Gif sur Yvette, France (2) IEF/Université Paris-Sud, 91405 Orsay Cédex, France (3) CESBIO/IUT-A, 24 rue d'Embaquès , 32000 Auch, France ABSTRACT The paper presents a new methodology adapted to the downscaling of low resolution IRT signals, i.e. the estimation of subpixel temperatures. The approach is based on the inversion of subpixel variables by multilinear regressions constrained by a priori temperature estimates provided by a physical land surface model. The method was developed and validated against a synthetic database built on model simulations. The precision of the methodology was analysed in terms of errors on the subpixel temperature estimations according to model and observation uncertainties. The impact of the number of observations used (i.e. the number of low resolution pixels considered) as well as the influence of the pixel heterogeneity were studied. Index Terms— Thermal infrared, subpixels surface temperature, SVAT models, spectral mixture analysis 1. INTRODUCTION The estimation of surface fluxes (like evapotranspiration) from space presents a variety of applications in environmental monitoring but remains a major challenge for the remote sensing community. For that purpose, thermal infrared imagery is of great interest since it provides surface temperature measurements which are strongly linked to surface energy and water budgets. Land surfaces are characterized by a strong heterogeneity due to the variability of vegetation and soils, water availability, etc…, and consequently, the estimation of surface energy budgets or surface temperature is often required at high spatial resolution, for example at the scale of individual agricultural fields or at least, the scale of the individual biome composing the landscape. Such a fine scale is within the resolution of high resolution instruments like Landsat Thematic Mapper or ASTER but with the compromise of a lower temporal resolution incompatible with routine needed estimations. The higher temporal resolution instruments present, in return, kilometric spatial resolution, integrating generally landscapes composed of different ecosystems. In order to use such data acquired over pixels composed of various land use, it appears necessary to develop methods to estimate sub-pixels variables from large scale measurements. Downscaling technics are generally based on the following assumptions: the mixed pixels are composed of a few fundamental components called endmembers and the large scale measured signal can be modeled as a linear combination of endmember contributions weighted by the fraction of area covered by each endmember within the mixed pixel. These technics requiring the knowledge of the land use/land cover inside the pixels are therefore based on the use of high spatial optical imagery too. The number of unknowns being larger than the number of observations, the equation system is highly underdetermined and stationarity assumptions (in space and/or in time) are needed to solve the problem. The use of initial or a priori solutions is also commonly used to guide the search strategy (see for example [1] or [2]. In this paper, we present a methodology to estimate sub pixels temperatures from low resolution thermal infrared imagery based on the resolution of the linear mixing model constrained with a priori knowledge of the endmember temperatures provided by a land surface model. 2. METHODOLOGY The linear mixing model consists in assuming that the total radiance emitted by an heterogeneous surface, is the sum of the radiances emitted by the individuals. In the thermal infrared, the emittance M of an object is given by the Stephan-Boltzmann law and is related to its absolute temperature T as follows: 4 M T εσ = (1) ε is the broadband surface emissivity in the whole IRT spectral domain (3-100μm) and σ, the Stephan-Boltzmann constant. Then, the total irradiance of a composite surface composed of n homogeneous individuals may be written: