*2.5. Albedo Data*

The albedo data employed in fitting and validation were based on the MODIS Bidirectional Reflectance Distribution Function (BRDF) and Albedo Product (MCD43) algorithm utilizing directional surface reflectance values [62,63] from both the Aqua and Terra satellites [64,65] and a semi-empirical kernel-driven BRDF model (Ross-thick, Li-sparse Reciprocal) [66]. Specifically, we employed the latest version (v6) of the MCD43A3 BRDF/Albedo product with a nominal spatial resolution of 500 m in a sinusoidal projection (tiles v2 h18 & v3 h18), which is now provided at a daily resolution [67]. Although they are now provided at a daily resolution, the v6 MCD43A3 product retains the 16-day observation window where the observations closest to the composite date (9th day) are given a greater weight in determining the appropriate reflectance anisotropy model (i.e., BRDF) [68]. The accuracy of the v6 product is generally higher for snow-free and full inversion retrievals, with RMSEs < 0.02 for most land cover types. For snow-covered conditions, RMSEs are typically < 0.05 for most land cover types [69]. Furthermore, the v6 product obtains more retrievals than the v5 product at higher latitudes from the use of all available observations (as opposed to four observations per day in v5) [69].

The albedo data were downloaded for a temporal extent spanning 1 January 2006–31 December 2010. This temporal extent was chosen because it falls in between the earliest satellite imagery underlying the original AR5 product (2002) and the latest aerial remote sensing data used to produce the SR16 product (2014). The quality flags of the MCD43A2 [70] companion product were used to filter and discard all non-full BRDF inversions, which includes those with solar zenith angles greater than 70◦. Data were then averaged into interannual monthly means, and composite dates and locations for the good quality data were stored for subsequent temporal synchronization with the snow cover and temperature data (next sections).

### *2.6. Snow Cover Data*

The snow cover data for the same temporal extent as the albedo data were based on the latest version (v6) of the MOD10A1 Snow Cover product [71], which is also provided at a daily resolution and has a nominal spatial resolution of 500 m in a sinusoidal projection. Unlike the previous product (v5), only the Normalized Difference Snow Index (NDSI) Snow Cover [72] is provided. The NDSI Snow Cover represents the visible fraction of a grid cell covered in snow and rarely exceeds 0.75 in forests. The detailed descriptions of the v6 product, including other important changes from v5, may be found in refs. [73,74]. Recent evaluations sugges<sup>t</sup> notable improvements in the v6 product over v5, which is largely due to refinements in the snow detection algorithm [75,76]. Evaluations at three locations after the conversion of NDSI Snow Cover to Fractional Snow Cover using linear regressions with global parameters sugges<sup>t</sup> an accuracy (RMSE) of 0.2–0.35 [75]. Prior to aggregating these to multi-year monthly means, the temporal signature of the snow cover data was synchronized to match that of the albedo dataset. In other words, the quality flags of the albedo dataset were applied to filter and discard snow cover retrievals for dates that did not correspond to the dates of the retained good quality albedo retrievals. NDSI snow cover (henceforth referred to as simply snow cover; *SC*) values outside the 0–100 range were also discarded prior to the application of the albedo quality filter.
