**6. Conclusions**

Land surface albedo is an important component of surface energy budgets. The validation of the satellite product is a precondition for its scientific use. Prior to evaluating the remote sensing products, identifying the spatial representativeness of the products is essential. It can help ground sampling point settings and match remote sensing data from multiple sources but it is difficult to determine the effective spatial representativeness of satellite albedo product from a physical perspective, as multi-temporal data are composed to derive the product.

In this study, we evaluated the spatial representativeness of MODIS daily albedo product (MCD43A3) at AmeriFlux sites. Among 1820 paired high-resolution (TM) and coarse-resolution (MODIS) albedo over different land cover types used in this study, around 74.5% pixels were found

to be heterogeneous pixels. In order to derive the most reliable spatial representativeness of MODIS albedo product, the land surface heterogeneity was first assessed by the field-calibrated TM albedo; semivarioagrams were then calculated from 30 m Landsat data at different spatial scales. Sill value and relative coefficient of variation were employed as key indices to determine the land surface heterogeneity. The 30 m Landsat albedo data was aggregated to 450 m–1800 m using direct average method and convolved with PSF method. The aggregated data was then compared with MODIS albedo product. The spatial representativeness of MODIS albedo product was determined according to the surface heterogeneity and the consistency of MODIS data and the aggregated TM value.

The results indicate that for most MODIS pixels their spatial representativeness tend to be larger than the 500 m nominal resolution. More specifically, for evergreen broadleaf forests, deciduous broadleaf forests, open shrublands, woody savannas, and grasslands, the effective spatial representativeness of the MODIS albedo was about 630 m; for mixed forest and croplands, the effective spatial resolution was about 690 m. The accuracy of the MODIS 500 m albedo product was high, with a correlation coefficient of 0.94 and RMSE 0.024 when compared with the calibrated TM albedo estimates. The choice of spatial aggregation method between simple spatial averaging and PSF-weighted averaging did not result in any significant difference in determining the spatial representativeness of MODIS albedo. It is also found that the spatial representativeness was difficult to determine at the sites where surface heterogeneity was very high (e.g., covered with evergreen needleleaf forest or partial snow).

In this study, long time period and large space data sets are used for spatial representativeness evaluation at 109 AmeriFlux sites with five land cover types when former works mainly focused on a specific research area. The availability of the 30 m Landsat albedo data set makes it possible for the analysis to be carried out at sites with different land cover types. There are many high-level remote sensing products, in this work, we only focus on evaluating the spatial representativeness of MODIS daily albedo product. Similar work is also worth for other products.

**Author Contributions:** H.Z. and S.L. conceived and designed the experiments. H.Z. performed the experiments and analyzed the data. J.W. and Y.B. helped the experiment and paper writing. T.H. and D.W. helped technology implementation of the albedo data processing. H.Z. wrote the paper. All the authors reviewed and provided valuable comments for the manuscript.

**Funding:** This research was supported by the National Natural Science Foundation of China under grants 41801242 and 41771379, the Key research and development program of China under grants 2016YFB0501404, 2016YFB0501502, the Chinese 973 Program under gran<sup>t</sup> 2013CB733403.

**Conflicts of Interest:** The authors declare no conflict of interest.
