NEW METHODOLOGY FOR AN IMPROVED THERMAL CALIBRATION OF LANDSAT 5 THROUGH FUSION OF ENVIRONMENTAL DATA SOURCES Francis P. Padula, John R. Schott, Nina G. Raqueño Digital Imaging and Remote Sensing Laboratory Center F. Carlson Center for Imaging Science Rochester Institute of Technology 1. INTRODUCTION The Landsat program is the National Aeronautics and Space Administration’s (NASA) longest operating earth imaging program [1]. The current operating satellites are Landsat 5 and Landsat 7. Landsat 5 was launched in March 1984 carrying the Multispectral Scanner (MSS) and the Thematic Mapper (TM) instruments. This work focuses on the post launch calibration of the TM thermal band (band 6). The historical calibration period spans the period from 1984 to 2000. It is through fusion of environmental data sources (i.e. buoy observations, surface observations, radiosonde observations) that a vicarious calibration approach will be implemented to construct the complete historical calibration of the Landsat 5 satellite’s TM thermal band. The need for such a study is particularly important because NASA has just implemented a correction to the calibration of the TM instrument for all data after 04/01/1999 based on vicarious calibration results [2]. This effort is aimed at evaluating whether a similar correction is needed for data during the period 1985 – 1999 when no vicarious calibration data are available. 2. METHODOLOGY The vicarious calibration process has two main thrusts: 1) convert buoy derived water temperature at depth z (T b ) to skin temperature (T s ) (for correlation with remotely sensed radiometric temperature); 2) capture the state of the atmosphere at the time of image acquisition. The Great Lakes have been chosen to serve as the study domain of this work. After examining factors such as availability and proximity of upper-air data, surface data, and buoy data to potential ground truth sites, as well as, the known dynamic and thermodynamic properties of these large water bodies, Lake Huron was selected as the best fit for all considered factors. In regards to the focus of this study, radiance reaching an orbiting satellite originates from the top tens of microns of the water body. Buoy measurements of the bulk water temperature at depth T b must be extrapolated to the water surface providing T s . This study investigated two T b to T s models: 1) Zeng et al. (1999) use an empirically derived approach that attempts to capture T s from near-surface wind speed and the diurnal variation of T b [3], 2) a simplified treatment derived from Fairall et al. (1996). This approach modeled the thermal response to surface heating generating tabulated results for peak increase in temperature and depth as a function of wind speed and clear sky conditions [4]. The second modeling effort Fairall et al. (1996) has been proposed to examine the validity of the preferred Zeng et al. (1999) approach. Due to the historical nature of the problem, one is limited by the availability of data. Consequently, spatial and temporal interpolations are needed to account for the state of the atmosphere for the near noon image acquisition time over the calibration site. To propagate the ground leaving radiance through the atmosphere to the sensor, the MODTRAN (MODerate resolution atmospheric TRANsmission) [5] tool was used to model the state of the atmosphere at the time of image acquisition. This allows for the computation of an effective radiance (L eff ) at the sensor from the known calibration site. Finally, image region of interests (ROIs) are taken about the buoy location(s) and the digital count values are converted to radiance values using the current gain and offset values prescribed to each image. A comparison is then made between the image derived radiance and the ground truth derived radiance, revealing the potential bias of the sensor. 3. CONCLUSIONS Calibration points were derived from scenes from 1984 - 2000 between the months of March - November (based on the availability of ground truth buoy data). In total 21 scenes have been investigated and processed, resulting in 36 calibration