Land surface brightness temperature retrieved from Landsat data

Abstract

Land Surface Temperature (Ts) is an important boundary condition in many land surface modelling schemes. It is also important in other application areas such as hydrology, urban environmental monitoring, agriculture, ecological and bushfire monitoring. Many studies have shown that it is possible to retrieve Ts globally using thermal infrared data from satellites. Development of standard methodologies that routinely generate Ts products would be of broad benefit to the utility of remote sensing data in applications such as hydrology and urban monitoring. AVHRR and MODIS datasets are routinely used to deliver Ts products. However, these data have a 1 km spatial resolution, which is too coarse to detect the detailed variation of land surface change of concern in many applications, especially in heterogeneous areas. Higher resolution thermal data from Landsat (60-120 m) is a possible option in such cases. To derive Ts, two scientific problems need to be addressed, the first of which is the focus of this paper: • Remove the atmospheric effects and derive surface brightness temperature (TB), • Separate the emissivity and Ts effects in the surface brightness temperature (TB). For single thermal band sensors such as Landsat 5, 7 and due to a stray-light issue on Landsat-8, the multi- band methods used to derive TB, such as the split window methods used for NOAA-AVHRR data (Becker & Li, 1990) and the day/night pairs of thermal infrared data in several bands used for MODIS (Wan et al., 2002) are not available for correcting atmospheric effects. The inputs used for retrieval of surface brightness temperature TB from Landsat data therefore need more attention, as the accuracy of the TB retrieval depends critically on the ancillary data, such as atmospheric water vapour data (precipitable water). In the past, it has been more difficult to retrieve a reliable TB product routinely from Landsat due to limited availability (time and space) of accurate atmospheric water vapour information. To test the possibility for retrieving a good quality TB product routinely from Landsat, in this paper, TB products are derived from four current and routinely available global atmospheric profiles, namely the NASA Modern Era Retrospective-Analysis for Research and Applications (MERRA); the National Center for Environmental Prediction (NCEP) reanalysis I, the National Center for Environmental Prediction reanalysis II and the European Centre for Medium-Range Weather Forecasts (ECMWF) Interim reanalysis. These products were evaluated against the TB derived from the near coincident ground-released radiosonde data (balloon data) using a physically based radiative transfer model MODTRAN 5. The results from this comparison have found: • The global data sets NCEP1, NCEP2, MERRA and ECMWF can all generally give satisfactory TB products and can meet the 1 K accuracy levels demanded by many practitioners. • The ECMWF data set performs best. The root mean square difference (RMSD) for the 9 days and 3 test sites are all within 0.4 K when compared with the TB products estimated using ground-released radiosonde measurements.