JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 101, NO. C5, PAGES 12,127-12,140, MAY 15, 1996 Global validation of the along-track scanning radiometer againstdrifting buoys A. R. Harris and M. A. Saunders Department of Space andClimatePhysics, University College London, Surrey, England Abstract. The along-track scanning radiometer (ATSR) was launched ontheEuropean Space Agency's first remote sensing satellite, ERS 1, on July 17 1991. ATSR is designed to retrieve sea surface temperature (SST) to an accuracy of 0.25 K rms,whichrepresents morethana factorof 2 improvement overany previously flown satellite radiometer. Early validation studies from limitedregions suggest thatATSR is capable of measuring SST to near thisdesign accuracy. We report a global validation study against quality-controlled drifting buoys by examining 280 matchups worldwide with ATSR measurements at theirfull (1 km) resolution. We investigate optimizing the precision of ATSR using four different SST algorithms derived using a theoretical atmospheric transmission model, combined with various techniques to reduce remnant noise and other errors. We find thata "low-noise" retrieval algorithm incorporating onlythe 3.7 and 11 gm nadirview channels gives the optimum precision, a global pixel precision of 0.26 K (or 0.25 K if 1/2 ø spatial averages are used). A standard deviation of 0.25 K against global driftingbuoydata approaches the geophysical limit set by theinherent variability of the skineffect andby the buoy bulk temperature accuracy. Further progress will require comparison against qualityin situ radiometer-derived skintemperatures, although theproblem of obtaining sufficiently largeand diverse data sets will need to be addressed. 1. Introduction With efforts steadilyincreasing to model climate change, the need for accurate global sea surface temperature (SST) measurementscontinues to grow. SST is a fundamental parameter for climate research as it controls the release of heat from the oceanto the atmosphere. The sheer size and remaining uncertainty in the annual heat loss from the ocean to the atmosphere demonstrate why accurate SSTs are important. The Gulf Stream, for example, has an annual net surfaceheat loss exceeding 200 W m -2 [ Oberhuber, 1988],this reaching 300 W m -2 for5 months of the year [Bunker, 1976]. These levels of heat exchange exceedthe globally averaged surface thermal heating from incoming solar radiation (~170 W m-2).However, of evengreater significance for improved climate modeling is the ~50 W m -2uncertainty in many regions forthe annual mean net ocean surface heat flux simulated by current atmospheric general circulation models [Gleckler et al., 1995]. This level of uncertainty is anorder of magnitude greater than the 4 W m -2 change in thermal forcing expected from doublinggreenhouse gasconcentrations [Intergovernmental Panel on Climate Change, 1994], so narrowing the ~50 W m -2 disagreement isclearly vital to the success of rigorous climate models. SST measurement error is a significant, though not dominant, contributor to this error. To examine the size of the SST error contribution we consider the latent heat flux term. For tropical regions, between latitudes 20øN and 20øS, the international Tropical Ocean-Global Atmosphere (TOGA) program of the World Climate Research Program(WCRP) hasspecified an SST accuracy over a 2ø latitude-longitude grid of 0.3 K every 15 days Copyright 1996 by theAmerican Geophysical Union. Paper number 96JC00317. 0148-0227/96/96JC-00317 $09.00 (WCRP, 1985). In the absence of other measurement errors an SST error of 0.3 K corresponds to an uncertainty in the total tropical ocean-atmosphere latent heat transfer of ~10 W m -2. With repeat samplingand the assumption of randomerrors,the error in themean SSTreduces as0.3n -1/2, where n is the number of samples. Thusin 15 days theWCRP requirement is easily met with the repeat sampling (several timesdaily) now achievable by current satellite instruments. This, of course, assumes that instrument biases due to calibration and aerosol contamination are <0.3 K. A more worthwhile goal for current numerical weather prediction and climate research is to achieve an SST point (rather thantime averaged) measurement precision of 0.3 K over a 1/2øxl/2 ø latitude/longitude grid. This precision would permit the dynamics and variability of ocean-atmosphere heat fluxeson timescales much less than 15 days to be studied with a contribution to theerror budget of only ~10W m -2due to SST error. Indeed, it may be argued that since two of the terms included in the bulk formula for latent heat are nonlinear (wind speed and specific humidity), latentheatshould be calculated on a point by point basis and then averaged to provide monthy means ratherthan averaging the parameters before applying the formula. Satellite imaging provides the best global synoptic view of SST as such measurements are largely free from the problems of data quality controlandlimited globalcoverage associated with "ships of opportunity" and drifting buoys. Measurements made by thermal infrared radiometers offer the most reliable SST estimates since they measure at wavelengths near 10 gm corresponding to the wavelength of peak radiative flux emission from the midlatitude ocean surface and the main thermal infrared atmospheric transmission window. For well over a decadethe main satellite instrument that hasbeenused for measuring global SST is the advanced very high resolution radiometer (AVHRR) flying on board the U.S. National Oceanic and Atmospheric 12,127