JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 102, NO. D14, PAGES 16,697-16,713, JULY 27, 1997 A split-window algorithm for land surface temperature from advanced very high resolution radiometer data: Validation and algorithm comparison C6sar Coll and Vicente Caselles Departmentof Thermodynamics, Facultyof Physics, University of Valencia, Valencia, Spain Abstract. A split-window algorithm for deriving land surface temperatures (LSTs) from advanced very high resolution radiometer(AVHRR) channels 4 and 5 is proposed and validated with in situ measured temperatures. On the basis of the radiative transfertheory the algorithmdefines a set of surface-independent coefficients which are equivalent to the classical split-window coefficients for sea surface temperature(SST). These coefficients are calculated using SST matchups (coincident AVHRR and buoymeasurements) provided by the National Oceanic and Atmospheric Administration (NOAA)-NASA Pathfinder Databaseof worldwidemeasurements. Thus calibrationof the split-window coefficients is done using real data. The variabilityof atmospheric attenuationis represented in the proposed algorithmby a quadratic dependence on the brightness temperaturedifference. For LST determination the emissivity effect is modeled through an additivecoefficient which depends on surface emissivity in the AVHRR channels 4 and 5. The algorithmis validatedfor both SST and LST by usingindependent ground-based and AVHRR data. The database usedin the validationof LST was obtainedfor a wide range of surface types in a semiarid environment. The samedatabases are used to comparethe accuracies of other published split-window algorithms. The proposed algorithmyieldsstandard errors of temperatureestimate between _ 1.0 and +_ 1.5 K, and no significant biases are observed. Although results are encouraging, more validationis required principally for moist atmospheric conditions. 1. Introduction Land surfacetemperatures (LSTs) derived from the ad- vanced veryhighresolution radiometer (AVHRR) instrument onboardthe National Oceanicand Atmospheric Administra- tion (NOAA) satellites providea uniquetool for deriving re- gional fluxes of latent and sensible heat over land surfaces [Norman et al., 1995].For this purpose, LST needs to be cor- rected for the absorption and emission of the Earth's atmo- sphere and the nonblackness of natural emitting surfaces. Split-window methodsfor atmospheric correction,which use channels4 (10.3-11.3 /am) and 5 (11.5-12.5 /am) of the AVHRR 2, perform well when appliedto the recovery of sea surface temperature (SST) [McClain et al., 1985],giving accu- racies of ___0.5 K for mostatmospheric conditions [May, 1993]. The split-window technique takesadvantage of the differential absorption between AVHRR channels 4 and 5, whichis closely correlatedwith atmospheric conditions, mainly given by the water vapor and air temperature profiles[McMillin, 1975]. The effect of surface emissivity is superimposed onto the atmospheric attenuation in the case of land. Unlike sea sur- face,land surface thermalemission is highly variable because it depends on a large numberof factors: pixel composition, veg- etation cover, soil background, and surfacegeometry,among others.Moreover, there is evidence of a spectral variation of surface emissivity for different landsurface materials [Salisbury andDMria, 1992]. This spectral and spatial dependence means Copyright1997 by the American Geophysical Union. Paper number 97JD00929. 0148-0227/97/97JD-00929509.00 that the correlation between the radiances measured in differ- ent channels is modified in such a manner that is not correlated to the atmospheric properties, the modification being highly dependent on the nature of the surface. In the last years a number of papers have addressed the extension of the split- window technique to the recovery of LST, resulting in a con- siderable variety of approachesand theoretical algorithms [e.g., Price, 1984; Becker andLi, 1990; Sobrino et al., 1991; Ottl• and Vidal-Madjar, 1992; Prata, 1993; Collet al., 1994]. In gen- eral, LST split-window algorithms take the form of linear com- binations of the satellite brightness temperatures; however, different expressions and values have been proposed for the algorithmcoefficients. Both sea and land surface algorithms have been derived from the radiative transfer theory assuming certain approxi- mations.The theoretical development determinesthe func- tional form of the split-window algorithms. However, there is an importantdifference between SST and LST algorithms with regard to the methodology in which the coefficients are de- rived. For sea surfacethese coefficients are obtained empiri- cally using extensive data setsof in situ measuredSSTs and cotemporal, collocated AVHRR measurements (the so-called matchups). The coefficients are then obtained by statistical regression [Strong andMcClain, 1984]. The matchup database consists of high-quality, cloud-freemeasurements distributed over worldwide oceans.This methodology ensuresthat the algorithms are calibrated with real data and over a highvariety of atmospheric conditions. Each time new matchupdata are available,algorithms can be further validated and regularly updated. For land surface an empiricalmethodology cannotbe ap- 16,697