چکیده (انگلیسی):

The Landsat 5 Thematic Mapper (TM) sensor provided the longest single mission terrestrial remote sensing data record but temporally sparse station keeping maneuvers meant that the Landsat 5 orbit changed over the 27 year mission life. Long-term Landsat 5 TM reflectance inconsistencies may be introduced by orbit change induced solar zenith variations combined with surface reflectance anisotropy, commonly described by the Bi-directional Reflectance Distribution Function (BRDF). This study quantifies the local overpass time and observed solar zenith angle changes for all the Landsat 5 TM images available at two latitudinally separated locations along the same north-south Landsat path (27) in Minnesota (row 26) and Texas (row 42). Over the 27 years the Landsat 5 orbit changed by nearly 1 h and resulted in changes in the Landsat 5 observed solar zenith angle of > 10°. The Landsat 5 orbit was relatively stable from 1984 to 1994 and from 2007 to 2011, but changed rapidly from 1995 to 2000, and from 2003 to 2007. Rather than directly examine Landsat 5 TM reflectance time series a modelling approach was used. This was necessary because unambiguous separation of orbit change induced Landsat reflectance variations from other temporal variations is non-trivial. The impact of Landsat 5 orbit induced observed solar zenith angle variations on the red and near-infrared reflectance and derived normalized difference vegetation index (NDVI) values were modelled with respect to different Moderate-Resolution Imaging Spectroradiometer (MODIS) BRDF land cover types. Synthetic nadir BRDF-adjusted reflectance (NBAR) for the Landsat 5 TM observed and a modelled reference year 2011 solar zenith were compared over the 27 years of acquisitions. Ordinary least squares linear regression fits of the NBAR difference values as a function of the acquisition date indicated an increasing trend in red and near-infrared NBAR and a decreasing trend in NDVI NBAR due to orbit changes. The trends are statistically significant but small, no more than 0.0006 NDVI/year. Comparison of certain years of Landsat 5 data may result in significant reflectance and NDVI differences due only to Landsat 5 orbit changes and cause spurious detection of “browning” vegetation events and underestimation of greening trends. The greatest differences will occur when 1995 Landsat 5 TM data are compared with 2007 to 2011 data; NDVI values could be up to 0.11 greater in 1995 than in 2011 for anisotropic land cover types and up to 0.05 greater for average CONUS land cover types. A smaller number of Landsat 5 TM images were also examined and provide support for the modelled based findings. The paper concludes with a discussion of the implications of the research findings for Landsat 5 TM time series analyses.