**1. Introduction**

Land surface albedo describes the fraction of incoming solar radiation reflected by the surface of the land. It influences the surface energy budget [1], and it is essential for global and regional estimation of energy and mass exchanges between the Earth's surface and the atmosphere [2–6]. An accuracy of 0.02–0.05 is required by climate models for global surface albedo [7,8]. To monitor the spatio-temporal changes in land surface albedo, albedo products are routinely generated from satellite data, such as the Polarization and Directionality of the Earth's Reflectance (POLDER) [9–12], the Medium Resolution Imaging Spectrometer (MERIS) [13], the Clouds and the Earth's Radiant Energy System (CERES) [11,12], the Visible Infrared Imaging Radiometer Suite (VIIRS) [14,15], and the Airborne Visible Infrared Imaging Spectrometer (AVIRIS) [16]. The Moderate Resolution Imaging Spectroradiometer (MODIS) onboard the Earth Observation System (EOS) Terra and Aqua satellites routinely provided data to derive the land surface shortwave and visible albedos [17–19] used to calibrate and improve albedo parameterizations for land, weather, and climate models [20–26]. Assessment of the accuracy of these products is important because it is critical to the scientific community for various applications. Feedback from this activity will help improve the generation of these products [27].

Direct comparison with the ground-based observations of albedo values is the commonly used method to assess the accuracy of remote sensing albedo products [28–30]. The tower observation is compared directly with the satellite product [31–33] based on the assumption that the satellite and tower observation have the same footprint or the landscape is homogeneous. Researchers subsequently recognized that the land surface albedo varies strongly in space and across seasons, because of which land surface homogeneity is now examined prior to the "point-to-pixel" comparison. Only sites whose representativeness is adequate for satellite pixel scales are selected in the direct validation and "heterogeneous" sites are excluded [34]. The method most suited to a comparison for all kinds of landscapes is using higher spatial resolution data as intermediate data [27,35,36]. Burakowski, et al. [37] used airborne hyperspectral imagery to validate MODIS albedo product in snow-covered areas; Mira, et al. [10] used the convolved albedo onboard the Formosat-2 Taiwanese satellite as a reference to evaluate the newly released MCD43D product. However, since no global high-resolution albedo product (at a level of tens of meters) is available, validation using intermediate data has been conducted at a limited number of locations.

Prior to evaluating the remote sensing product, identifying the spatial representativeness of the products is essential. It can help ground sampling point settings and match remote sensing data from multiple sources. MODIS albedo products are retrieved using observations covering a large area that depends on the view zenith angles. Although observations are weighted by angular coverage before albedo retrieval, the actual coverage of the pixel is always larger than the nominal spatial resolution [38,39]. Efforts have been made to characterize the effective resolution of the MODIS gridded product [40]. Mira, et al. [10] used Formost-2 data at a resolution of 8 m to characterize the equivalent point spread function of MODIS albedo at a 1 km pixel. The Full Width at Half Maximum (FWHM), recognized as the effective resolution, has been confirmed to represent the footprint of MODIS data for accurate validation. Campagnolo et al. [41] used extensive time series data (2003–2014) at a large size of the linear natural target in the Netherlands to analyze the effective spatial resolution of the MODIS albedo product (the spatial scale the pixel represents). They verified that the spatial representativeness of the MODIS daily albedo product approximately varied from 606 m to 843 m. Campagnolo and Montano [40] estimated the point spread function (PSF) of nominal 250 m MODIS gridded surface reflectance products, and discovered that the spatial representativeness varied from 344 m to 835 m along the rows, and between 292 m and 523 m along the columns. Their work helps users understand the spatial properties of the satellite product, but their work was based only on a single area, and a general adaptable result that is based on different kinds of land surfaces is needed.

The MODIS BRDF/albedo product was derived with a semi-empirical, kernel-driven BRDF model. Data for 16-day, multi-angular, cloud-free, atmosphere-corrected surface reflectance was compiled to apply the retrieval procedure. To better characterize the rapid change of the land surface, the daily albedo product was retrieved using the same semi-empirical algorithm, but with the single day of interest emphasized by being weighted more heavily [39,42] in the 16-day composite period. The spatial representativeness and accuracy of the newly released daily albedo product (MCD43A3, V006) have not been tested for different types of land cover for a long time series.

The high-resolution (30–80 m) satellite albedo product is essential in understanding the climatic consequences of land cover change and medium-to-fine scale applications [43]. It is also a key bridge to the assessment of coarse-resolution products. He et al. [44] developed a method to estimate both snow-covered and snow-free albedo from the Chinese environment and disaster monitoring and forecasting small satellite constellation (HJ) satellite data. Zhou, et al. [15] then derived 30 m albedo from Landsat 7 and Landsat 8 using a similar algorithm. He et al. [45] evaluated 30 m albedo product estimated from Landsat Multispectral Scanner (MSS), Thematic Mapper (TM), Enhanced Thematic Mapper Plus (ETM+), and Operational Land Imager (OLI) at Surface Radiation (SURFRAD), AmeriFlux, Baseline Surface Radiation Network (BSRN), and Greenland Climate Network (GC-Net) sites, with results indicating that the direct estimation approach can generate reliable albedo estimates with accuracy of 0.022 to 0.034 in terms of the root mean square error (RMSE). The derivation of global land surface albedo product using Landsat sensors makes it possible to better understand energy transfer between the land surface and the atmosphere at global and regional scales. Furthermore, it makes the assessment of the coarse-resolution albedo product possible, as well as scale transformations for different land cover types and landscapes.
