**4. Discussion**

The spectral signatures created from the direct measurements performed with the hand-held spectroradiometer demonstrate the unique features of cacti. It is clear that a water absorption dip occurs around 972 nm and reflectance peaks occur around 862 nm and 1072 nm for barrel, cholla, and prickly pear cacti. The dip is also present in the data of drone-mounted and airplane-mounted hyperspectral sensors. Nuanced differences between the aerial spectra and the hand-held spectra could be caused by a variety of factors, including 1. spatial resolution, 2. radiometric sensitivity, 3. The reflectance calculation method, and 4. atmospheric water absorption of spectra around 972 nm [31].

We observed diminished sensitivity of the Cacti Indices within the NEON AVIRIS data. This was probably a function of the coarser spatial resolution of that imagery. Within 1 m pixels, instead of being pure cacti reflectance, the features are often a mix of vegetation, bare ground, and shadow. Prickly pear samples were especially prone to mixed signals due to their spreading structural form (Figure 5). However, prickly pears tended to be large, 2–4 m in diameter, making up for shadowing created by their structural composition. Barrel cactus and cholla usually had less mixed signals due to their more compact morphology with diameters as small as 50 cm. Despite reduced sensitivity from AVIRIS NEON data, the cacti indices still demonstrated the ability to separate cacti from non-cacti vegetation. Though both CI1 and CI2 show separability between cacti and non-cacti vegetation, CI1 shows a broader separation, making it the preferred index in our study area.

**Figure 5.** On the **left**: Example of a prickly pear cactus and a cholla cactus as captured by the DJI Mavic Pro Quadcopter. The nominal spatial resolution of the image is 1.6 cm. The difference in structure between the cholla and prickly pear is clear as seen by the influence of shadowing on the prickly pear cactus. On the **right**: The Cacti Index 1 visualized using the 1 m NEON AVIRIS reflectance data. It can be seen that cholla and prickly pear have higher cacti index values.

The Nano Hyperspectral sensor has diminished radiometric sensitivity near the edges of its range (near 400 nm and 1000 nm), which leads to a lower signal to noise ratio than other spectral bands in the sensor. This hardware limitation impacts CI1 produced from the Nano Hyperspectral sensor because 972 nm is near the edge of the silicon-based sensitivity. The implication is that in addition to identifying cacti, the index produces many false high values for low light shadowed areas (Figure 6). Mitigation strategies could include identifying and removing low radiance pixels found in shadowed areas, or collecting imagery with longer exposure time. Ideally, drone-based mapping of cacti should use a hyperspectral sensor with a wider range (i.e., >1000 nm) than the Nano Hyperspectral sensor provides.

**Figure 6.** Drone-mounted Nano hyperspectral imagery shown as true color (**left** panel) and Cacti Index 1 (**right** panel). Cacti were easily identifiable with the index; however, sensitivity problems near the edge of the sensor range created false high index values in low light shadowed areas of the study area.

The spectral slices captured by the AVIRIS NEON sensor (5.5 nm) and Nano Hyperspectral sensor (2.4 nm) played a role in cacti detection. The smaller the slices the better the sensor is at capturing the difference between the peak at 862 nm and the dip at 972 nm. This led to a smaller dynamic range in the CI1 values calculated with the AVIRIS NEON data compared to the CI1 values calculated with the Nano Hyperspectral data.

The water absorption dip (972 nm) observed in the drone-based and airplane-based imagery demonstrate that the Cacti Indices can be used for the identification and mapping of cacti species across larger extents. Depending on the application, the Cacti Indices could be combined with other sensor data (e.g., LiDAR height information), used within a supervised classification framework, or implemented with a user-defined threshold.

The Cacti Indices should be effective in identifying succulent cacti in other regions of the world if the plant species store water in their tissue similar to barrel, cholla, and prickly pear cacti. However, localized research should investigate the extent to which non-target species might exhibit high index values and confuse cacti identification. In our study area for example, we discovered that Palo Verde, with their photosynthesizing stems, exhibited CI1 values nearly as high as prickly pear samples.

Using the Cacti Indices with satellite imagery may be possible but will be challenging. Our methods require narrow spectral bands and high spatial resolution to identify individual cacti. Additionally, the spectral area of interest near 972 nm is greatly impacted by water vapor in the atmosphere [31]. As a result, many satellite sensors do not have bands sensitive to this spectral region, and if they do, the signal will be quite weak. The Earth Observing-1 (EO-1) Hyperion (2000–2017) and a few other orbital hyperspectral reflectance sensors (HISUI-Hyperspectral Imager Suite-onboard the International Space Station; EnMAP-Environmental Mapping and Analysis Program; PRISMA-Hyperspectral Precursor and Application Mission) could possibly leverage the Cacti Indices [32]. These hyperspectral sensors all have moderate spatial resolutions (e.g., 30 m) that are unable to detect individual cacti, but could be used to estimate percent cover of cacti per pixel. More research on these sensors and platforms is needed.
