*2.1. UAV Experiments*

We conducted two UAV flights along predefined routes (Table 1). One took place in a playground on the Yale University campus (41.317◦N, 72.928◦W) on 30 September 2015, and the other in Brooksvale Recreational Park, in Hamden, Connecticut, USA (41.453◦N, 72.918◦W) on 9 October 2015. The sky condition was clear on 30 September, 2015, and overcast on 9 October, 2015. For the Yale Playground, a quad-rotor drone equipped with a GARMIN VIRB-X digital camera was used for the image acquisition (Figure 1a). A fixed-wing drone designed and assembled by CielMap [31] equipped with a Sony NEX-5N camera was used in the second experiment (Figure 1b). Both cameras had fixed aperture and automatic shutter speed. The spectral sensitivity of the Sony NEX 5N camera can be found in Ryan et al. [22]. However, the spectral sensitivity of GARMIN VIRB-X was not released by the manufacturer. Although Ryan et al. [22] used raw images for retrieving ice sheet reflectance, the study by Lebourgeois [23] showed little improvement from using the raw images over JPEG images. They demonstrated that JPEG and RAW imagery data have a linear relationship for the range of DN values needed for crop monitoring. In this study, we used the compressed JPEG images.

**Figure 1.** The unmanned aerial vehicles (UAVs) before launch. (**<sup>a</sup>**,**b**) are the fixed-wing and rotor wing UAVs for Brooksvale Park and Yale Playground, respectively.


**Table 1.** Information about the study sites and the drone experiments. Here, image overlap is defined as the number of photos that were sampled in the same pixel. For example, an overlap of nine means that each pixel is seen by at least nine photos.
