*2.3. Landsat-Retrieved Albedo*

The TM onboard the Landsat 4 and 5 satellites with seven spectral bands covered the shortwave range at a resolution of 30 m from 1984 to 2011. It is an optimal data source for regional and global land surface change research because of its long period of operation [48–52]. The high-spatial-resolution (30 m), long-term albedo product based on Landsat data has been derived [43,45,53]. In this study, we adopt the algorithm of He et al. [45] to derive the global long-term land surface albedo product from Landsat TM data. L1T data that were temporally consistent with AmeriFlux site observations were downloaded from the U.S. Geological Survey (USGS) website. To minimize the influence of cloud coverage, only the highest data quality (quality flag is 9) and a maximum cloud cover of 10% were used. In this paper, 2034 scenes of TM data were downloaded, covering all land cover types and seasons, where 1820 scenes corresponded to high-quality MODIS data. TM data were first rectified to sinusoidal projections and resampled to 30 m resolution using the nearest neighbor algorithm.

The accuracy of meso-scale data needs to be verified and calibrated before they are used to bridge field measurements and coarse-resolution products. Landsat TM albedo data was first calibrated with the field data. The Landsat TM overpassed at local time 10:00–11:00 AM. To guarantee the temporal consistency between TM data and field observation, field observations 15 min before and after the TM passing over were averaged. The field observation covers an area larger than 30 m, so we assume that the land surface at 30 m scale is homogeneous and field observation can compare directly with TM data. However, for each AmeriFlux site, the tower-located TM pixel was picked out according to the tower location. Sometimes, the tower was not located at the center of the pixel and possible geometric correction error may also exist, so albedo of 3 × 3 TM pixels around the tower were averaged and compared with the field observations.

Figure 1 shows a scatter plot of the field measurements and the TM albedo. The correction coefficient was derived from least squares regression between TM data and ground measurement. The determination coefficient was 0.877, RMSE and bias were 0.033 and 0.00035, respectively. The calibrated

TM data with the determined correlation were then used as a bridge for comparison with the MODIS albedo product as well as for further analysis to determine the spatial representativeness of the later.

**Figure 1.** Field data comparison with Landsat Thematic Mapper (TM) data, which is the average of 3 × 3 pixels. The fitting function is then used for TM albedo data calibration.
