*2.1. Study Areas*

We selected 18 FluxNet sites as our research areas. Among them, 15 sites were used for estimating and verifying the single-point time series albedo for five different land cover types (cropland; deciduous broadleaf forest; evergreen needleleaf forest; grassland; evergreen broadleaf forest). Five sites were selected as the central pixel for estimating and verifying regional time series albedos for each land cover type.

Due to the long satellite reentry cycle and cloud contamination, satellite data and ground data did not always match temporally. We also needed to match ground data with TM data in the time series to conduct this study, so the year of data used for each site in the experiment was inconsistent . Table 1 provides further information.


**Table 1.** Ground stations used for validation.

Cropland (CRO); deciduous broadleaf forest (DBF); evergreen needleleaf forest (ENF); grassland (GRA); evergreen broadleaf forest (EBF). Sites with \* are used for estimating and verifying single-point time series albedo, sites with + are selected as the central pixel for estimating and verifying regional time series albedo.

## *2.2. Ground Verification Data*

FluxNet is a global network of micrometeorological flux measurement sites that measure the exchanges of carbon dioxide, water vapor, and energy between the biosphere and atmosphere. It was established based on other observation networks including AmeriFlux, CarboEurope, AsiaFlux, OzFlux, and a few independent sites. At present, over 140 Flux tower stations are operating on a long-term and continuous basis. Data and site information are available online at the FluxNet website, http://fluxnet.fluxdata.org/. Land surface types include temperate conifer and broadleaf (deciduous and evergreen) forests, tropical and boreal forests, crops, grasslands, chaparral, wetlands, and tundra. Sites exist on five continents and their latitudinal distribution ranges from 70 ◦N to 30 ◦S. The FluxNet site records upwelling and downwelling shortwave radiative flux data with a half-hour observation period. For our purposes, we selected observation data from 0.5 hours before

and after noon. The shortwave albedo of the site was calculated as the ratio of upwelling radiation and downwelling radiation. After eliminating invalid observations (filled data) and data with albedos less than 0 or greater than 1, each selected site was associated with sufficient time series ground albedo data for validation.

## *2.3. Landsat Satellite Data*

Landsat sensors have continuously imaged the land surface since the 1970s [28]. The TM onboard Landsat 4 and 5 satellites with seven spectral bands sampled the shortwave range at a spatial resolution of 30 m from 1984 to 2011. TM images are an important remote sensing data source for earth resources and environments as per their high spatial resolution, spectral resolution, and positioning accuracy. We downloaded all available Landsat data during the ground measurement period. The distribution of available high-quality TM data corresponding to each FluxNet site within one year is shown in Figure 1. Abundant cloud-free TM data can be obtained for individual pixels; TM data are prone to contamination over larger areas. As shown in Figure 1, all of the sites have more than five available cloud-free TM images within one year. This provides reliable observation data for the effective operation of the EnKF algorithm.

**Figure 1.** Distribution of available Thematic Mapper (TM) data corresponding to Flux sites (one year).

### *2.4. MCD43A3 BRDF/Albedo Product*

MCD43A3 (Version 6) (V006) is the latest version of the MODIS Bidirectional Reflectance Distribution Functions (BRDF)/Albedo product. It includes bi-hemispherical reflectance (white-sky albedo) and directional-hemispherical reflectance (black-sky albedo) with a 500 m spatial resolution. The daily albedo is composed of 16-day multi-angle observations, where the Julian date of each specific file represents the 9th day of the 16-day retrieval period. MCD43A3 was produced based on the Algorithm for Model Bidirectional Reflectance Anisotropies of the Land Surface (AMBRALS), which uses all atmospherically corrected, high-quality, cloudless surface reflectance over the course of 16 days to achieve the best-fit surface bidirectional reflectivity via Ross–Li kernel models [29,30]. The Ross–Li kernel model is expressed as follows:

$$R(\theta, \theta, \phi, \Lambda) = f\_{\text{iso}}(\Lambda) + f\_{\text{vol}}(\Lambda)K\_{\text{vol}}(\theta, \theta, \phi) + f\_{\text{geo}}(\Lambda)K\_{\text{geo}}(\theta, \theta, \phi) \tag{1}$$

where *<sup>R</sup>*(*<sup>θ</sup>*, *ϑ*, *φ*, Λ) is the frontal reflectivity with a solar zenith angle *θ*, observed zenith angle *ϑ*, relative azimuth *φ*, and wavelength band Λ. *fiso*(Λ) is the proportion of uniform scattering in all directions, *fvol*(Λ) is the proportion of body scattering, *fgeo*(Λ) is the proportion of geometric optical scattering, *Kvol*(*<sup>θ</sup>*, *ϑ*, *φ*) is the RossThick kernel, and *Kgeo*(*<sup>θ</sup>*, *ϑ*, *φ*) is the LiSparse kernel.

MODIS V006 products provide shortwave black-sky albedo and white-sky albedo [31], which must be converted into blue-sky shortwave albedo according to the proportion of sky scattered light [32].

$$a(\theta\_{\rm i}, \lambda) = (1 - \mathbf{s}(\theta\_{\rm i} \tau(\lambda)))a\_{\rm bs}(\theta\_{\rm i}, \lambda) + \mathbf{s}(\theta\_{\rm i} \tau(\lambda))a\_{\rm ws}(\theta\_{\rm i}, \lambda) \tag{2}$$

where *<sup>α</sup>*(*<sup>θ</sup>*i, *λ*) is the blue-sky albedo of the band *λ* at a solar zenith angle of *θ*, *<sup>α</sup>*bs(*<sup>θ</sup>*i, *λ*) is the black-sky albedo, *<sup>α</sup>*ws(*<sup>θ</sup>*i, *λ*) is the white-sky albedo, and <sup>s</sup>(*<sup>θ</sup>*i, *τ*(*λ*)) is the fraction of diffuse skylight when the solar zenith angle is *θ*, which is a function of aerosol optical depth and can be calculated using a predetermined look-up table (LUT) based on the 6S atmospheric radiative transfer code [33].

There was some data missing in MCD43A3 due to retrieval failure, so we used a gap filling algorithm to construct a continuous albedo data set. Missing data was filled with the average of all pixel values of the same land type in a given window (10 × 10 pixels) centered on the target pixel.
