**4. Discussion**

Evaluation of earth system models requires surface albedo estimates with an absolute accuracy of between 0.02-0.05 units of albedo [24,42,46]. In this study, UAV flights over a mixed northern hardwood forest produced measurements within 0.01 units of each other across five flights and five hours. This albedo fell well within the range of site variation in albedo for fixed tower measurements of mixed forest albedo in Bartlett, NH; Durham, NH; and Petersham, MA, differing by no more than 0.02 units despite small cross-site differences in field of view, topography, and species composition. The tight grouping of UAV and tower estimates over a common, well–surveyed land use type supports the idea that this methodology will be able to provide accurate albedo measurements over land use types that have been poorly surveyed by conventional methods. Measurements made over a Norway spruce monoculture and cropped willow field show that the UAV easily distinguishes between land use types across multiple flights. These flights demonstrate how UAV may be used to sample sites that are not well captured by either satellite or tower measurements due to their small footprint or high rate of turnover for harvesting.

Although ground measurements (UAV and tower) were tightly grouped within a small range of albedo, each site was significantly different from the others, indicating site to site variation likely caused by species, land use, and topographic differences. This variability was poorly captured by satellite measurements, which were more tightly grouped and generally not significantly different from each other. Due to this, satellite data poorly estimated variation on the ground; UAV and tower data were generally lower than satellite data by 6 to 16%, although Bartlett forest reported 9% higher albedo than the satellite. UAV data did not differ from tower data in this way. This finding provides

additional support for the use of ground measurements, such as can be obtained by UAV, to validate satellite measurements and refine model predictions. Satellites have a larger and differently distributed field of view which may be confounded by different land use types at the pixel boundaries [4,24]. UAV can capture albedo with greater precision, and with the flexibility to take fine-scale measurements across the entire landscape.

Certain caveats of this method should be considered; uneven cloud cover means that a homogenous down-welling flux of incoming solar radiation cannot be assumed over the UAV's entire flight path. All flights in this study had to be recorded on days with no to very minimal cloud cover. UAVs must be flown such that unaided line of site is maintained, which can limit the range and altitude possible for flights over high canopy (Supplementary Table S5). Finally, continuous albedo measurements over the course of an entire day would be dependent either on the capacity to make many serial flights over the time horizon desired.

Satellites are often insufficient to capture the fine-scale landscape heterogeneity caused by local variation in canopy density, vegetative community, terrain, and other local scale properties [4,24,42]. Fine-scale point measurements, however, often lack the range needed to assess these properties on a global scale [3]. It is our belief that using UAVs to measure albedo will improve our ability to determine sources of variability in albedo measurements. Payload mass reduction through a split upwards and downward-facing sensor widens the range of UAVs available for these types of measurements and maximizes flight time (Table 2). The method described here provides a simple method of albedo assessment accessible to the typical researcher.



\* Based on MK HiSight SLR1 and Gaui Crane Gimbal.
