Towards a first Time-Series of Land Surface Temperature from AATSR with Uncertainty Estimates
Ghent, Darren; Remedios, John
University of Leicester, UNITED KINGDOM

Land surface temperature (LST) is the radiative skin temperature of the land, and is one of the key parameters in the physics of land-surface processes on regional and global scales. Earth Observation satellites provide the opportunity to obtain global coverage of LST approximately every 3 days or less. The Advanced Along-Track Scanning Radiometer (AATSR) is an important source of satellite-retrieved LST data for the near 10-year window from July 2002 to April 2012, with this instrument achieving high accuracy in measurements with extraordinary stability - an important consideration for potential climate time series analysis. Being part of the ATSR-series of instruments the potential to develop a high quality 20-year LST time series further exists. Here we present first regional and global time-series of LST data from AATSR with estimates of uncertainty.

Although time-series across all three ATSR missions have previously been constructed (Kogler et al., 2012), the use of low resolution auxiliary data in the retrieval algorithm and non-optimal cloud masking resulted in time-series artefacts. As such, considerable ESA and NCEO supported development has been carried out on the AATSR data to address these concerns. This includes the integration of high resolution land cover, fractional vegetation and precipitable water data in the retrieval algorithm; subsequent generation of coefficients utilising 0.05° emissivity data and ECMWF ERA-Interim atmospheric profile data; optimisation of the non-linear water vapour and view angle sensitivity parameters; and the development of an improved cloud mask based on the simulation of clear sky conditions from radiative transfer modelling (Ghent et al., in prep.).

The uncertainty analysis takes into account the expected performance of the retrieval algorithm under varying surface and atmospheric conditions. We characterise the pixel-level uncertainties in terms of five components: radiometric noise; fractional vegetation cover as representative of surface emissivity; atmospheric water vapour; geolocation uncertainty; and uncertainties as a result of the coefficient fitting process. The uncertainty due to misclassification of cloudy pixels is difficult to propagate to LST uncertainty bars and has yet to be evaluated in the framework of the current study. The total uncertainty budget is thus a combination of these five components added together in quadrature; these uncertainty fields being a requirement for interpretation of time series of data.

The progress made here will allow other time series of LST to be compared with the record from AATSR with greater certainty and hence increases confidence in our knowledge of recent surface temperature changes over the land. Furthermore, these improvements are encompassed in the on-going preparation of the level-2 LST product (SL_2_LST) for the upcoming Sea and Land Surface Temperature (SLSTR) instrument on-board Sentinel-3 - the successor to AATSR.

References

Ghent, D., Corlett, G., and Remedios, J. Advancing the AATSR land surface temperature retrieval with higher resolution auxiliary datasets, in prep.