**1. Introduction**

Surface albedo is a key parameter in the surface energy balance, and it therefore plays an important role in land–climate interactions. As a key biophysical property of land ecosystems, surface albedo can change throughout the season, due to changes in the vegetation morphology, and it can also be affected by sky conditions [1]. Quantification of the surface albedo at the landscape scale is still subject to many sources of uncertainty, especially over urban land [1,2].

Satellite remote sensing has been widely used for the determination of land surface albedo [3–6]. An advantage of satellite monitoring is that it provides global coverage. New satellites can provide albedo measurements at reasonably high frequencies (2–3 days in the best case for Sentinel 2) and spatial resolutions (pixel size 10 m in the case of Sentinel 2, and several cm in the case of DigitalGlobe) to provide useful information for studies on ecosystem (tens of meters) to landscape (several kilometers to tens of kilometers) scales. However, all satellite measurements are biased towards cloud-free sky conditions. In urban landscapes with heavy haze pollution, retrieval of the true surface albedo from satellite imageries must remove signal contamination caused by particle scattering. Lightweight unmanned aerial vehicles (UAVs) as an alternative for albedo monitoring may be able to overcome these limitations. UAVs can cover areas ranging from 0.01 km<sup>2</sup> to 100 km2, depending on battery life and type of UAV [7]. They provide measurements at sub-decimeter spatial resolutions, and they can be used to obtain data under both clear sky and cloudy conditions [8,9]. UAV experiments can be conducted at almost any time, and at any locations [10–12]. Furthermore, the labor and financial expense of UAVs are much lower than those of aircraft [13]. Finally, UAVs can measure albedos at locations that are not accessible by ground-based instruments, such as steep rooftops in cities.

In a typical UAV experiment, consumer-grade digital cameras are utilized as multispectral sensors, similar to their counterparts on board satellites, to measure the spectral radiance reflected by ground targets, typically in the red, green, and blue wavebands and occasionally with modification to include a near-infrared (NIR) waveband. These at-camera radiance data are stored as digital numbers (DN), usually with an 8-bit resolution ranging from 0 to 255, to represent the brightness of the targeted object [14]. Vegetation indices derived from the spectral information [15] allow for the monitoring of vegetation growth status [16–18], and the estimation of crop biomass [19]. Because the UAV flies below cloud layers, cloud interference is no longer a problem. Also, because of the low flight altitude in typical UAV missions, the at-sensor radiance is a direct measure of the actual surface reflected radiance, but this is not true for satellite monitoring (Even with UAVs flying at higher flight altitudes, atmospheric interference is still much less severe than with satellite monitoring). Alternatively, a UAV can be deployed as a platform to carry pyranometers to measure albedo. For example, Levy et al. measured surface albedo over vegetation using a ground-based pyranometer paired with a pyranometer mounted on a quadcopter [20].

Although increasingly being used to produce image mosaics and to construct 3D point clouds of surface features, consumer-grade cameras on board of UAVs have rarely been used to study land surface albedo. The land surface albedo here refers to blue sky albedo, which means that it is the albedo under ambient illumination conditions. By using a Finnish Geodetic Institute Field Goniospectrometer, which includes a fisheye camera on a fixed-wing drone, Hakala et al. [21] estimated the bidirectional reflectance distribution function (BRDF) of a snow surface. Ryan et al. [22] measured snow albedo in the Arctic region with pyranometers on board a fixed-wing UAV. Since consumer-grade digital cameras are not calibrated for radiance measurements, many internal factors, such as Gamma correction, color filter array interpolation (CFA), and the vignetting effect, can contribute to radiometric instability [23]. Gamma correction is a nonlinear transformation of the electro-photo signal received by the detector at the focal plane of the camera to a DN output. The DN values and their corresponding raw data are not exactly 1-to-1 matched [22]. Because there is only one band value (e.g., green) for each pixel on a charge-coupled device (CCD) or a complementary metal-oxide-semiconductor (CMOS), the other band values (i.e., red and blue) have to be estimated by using the CFA interpolation method. The vignetting effect refers to the phenomenon whereby objects farther away from the image center will appear darker. All of these factors must be accounted for if the DN value is to be converted to true reflectance. Some researchers argue that using raw images from the camera instead of the compressed JEPG or TIFF images can avoid the Gamma correction and CFA [22]. Applying a vignetting mask on the images can effectively alleviate the vignetting effect [24].

Determination of the albedo with UAV data consists of three steps. First, accurate albedo determination requires radiometric calibration of the camera's DN values, to represent physically meaningful surface reflectance. Currently, radiometric calibration methods fall into two categories: absolute and relative. In absolute calibration, the calibration function that relates the DN value to the reflectance is based on the measurement of an accurately known, uniform radiance field [25]. Relative calibration is especially needed for sensors having more than one detector per band. By normalizing the outputs of different detectors, a uniform response can be obtained [25–28].

The next step after the radiometric correction involves the conversion of the spectral reflectance at the camera viewing angle to the hemispheric reflectance or spectral albedo [29,30], ideally using a BRDF of the target. Because the determination of the BRDF requires measurements at multiple illumination and viewing angles, a process that requires elaborate preparation and post-processing [21], it is commonly assumed that all the objects are Lambertian [22]. The albedo values that we retrieved in this study should be considered as "Lambertian-equivalent albedos". Without the consideration of the BRDF effect, spectral reflectance is essentially the same as the spectral albedo [29,30].

The third step is to convert the spectral albedo of discrete bands to the broadband albedo, that is, the total reflectance in all directions in either the visible band (wavelength 380 nm to 760 nm) or the shortwave band (250 nm to 2500 nm). Wang et al. [30] have established an empirical method to convert the spectral albedo measured by Landsat 8 to the broadband albedo. Here, we propose that the same conversion equation can be applied to UAV albedo estimation.

The objective of this study is to develop a UAV method for determining the landscape albedo. The method was tested at two sites typical of urban landscapes, and consisting of impervious and vegetation surfaces. The visible and shortwave band albedo derived from our method were compared with those of Landsat 8. This method can save labor cost, and it can be applied to the landscape albedo estimation where direct field measurement may be difficult.
