IMPROVING THE AVHRR ESTIMATES OF THE ADRIATIC SEA SURFACE TEMPERATURE Igor Tomazic, Milivoj Kuzmic Satellite Oceanography Group, Center for Marine and Environmental Research, Rudjer Boskovic Institute, Zagreb, Croatia Giulio Notarstefano, Elena Mauri, Pierre-Marie Poulain Istituto Nazionale di Oceanografia e di Geofisca Sperimentale, Trieste, Italy ABSTRACT Sea surface temperature (SST) is an important parameter in oceanography and meteorology due to its major influence on the exchange processes at the air-sea interface. Ever-increasing complexity and resolution of global and local, oceanographic, meteorological and coupled models further stress the need for accurate estimates of this parameter. Satellite measurements offer synoptic view and repetitive coverage of vast areas, but globally derived SST algorithms potentially conceal large regional differences. Regional performance assessment of SST algorithms is therefore an important task. Two matchup databases of satellite and in situ SST were compiled at two locations, Rudjer Boskovic Institute (IRB) and Istituto Nazionale di Oceanografia e di Geofisca Sperimentale (OGS) with a view to validate the Adriatic Sea SST derived from locally collected NOAA-15, NOAA-16 and NOAA-17 AVHRR HRPT telemetry. Bulk SST data from 118 drifting buoys (mostly of the CODE type) have been used as the sea-truth. For the period between October 2002 and December 2003 165,252 matchup pairs were found satisfying 4-hours temporal and pixel-size spatial constraints. When more rigorous co-location constraints were set on atmospheric condition, temporal and spatial discrepancy (cloud-free, within a satellite pass, within a pixel) the number was reduced to 2560. The validation has been made for multi-channel SST (MCSST) algorithm, separately for the daytime and nighttime pairs. Our analysis suggests that the use of regionally optimised retrieval coefficients improves the accuracy and statistics of the satellite derived Adriatic SST. When global coefficients are used in Adriatic absolute bias may exceed 0.3°C and scatter is typically around 0.7°C. With introduction of local coefficients bias drops to almost zero, and the scatter reduces to less then 0.6°C, for all three satellites and for both daytime and nighttime algorithms. Differences arising from different receiving and processing hardware and software turned out to be comparatively small.