2.2.1. Field Spectroradiometer

We acquired outdoor clear-sky hyperspectral nadir measurements with a full-range (350–2500 nm) spectroradiometer, FieldSpec®3 by Analytical Spectral Devices Inc. (ASD) Boulder, CO, USA, with a 15-degree field of view during the summer of 2017. We used a calibrated Spectralon panel to compute component/surface reflectance data [24]. We took measurements from 5 chollas, 11 prickly pear cacti, 9 barrel cacti, 32 mesquite, 9 creosote, 26 patches of bare soil, and 28 plots of grass. We averaged the measurements for each type of sample.

From the field spectroradiometer data, we identified the same reflectance dip at 972 nm first reported by van Leeuwen. This reflectance dip is due to water absorption in succulent cacti and can serve as a distinguishing characteristic from non-cacti vegetation [24].

We measured the magnitude of the dip with a normalized difference approach between the reflectance values at the bottom of the dip and reflectance values immediately outside of the dip. This is similar to the concept of the normalized difference vegetation index (NDVI) [29].

A priori, it was unknown which specific bands outside of the dip would provide a robust index capable of distinguishing cacti from non-cacti vegetation. We tested two different spectral indices.

Cacti Index 1 (CI1) uses reflectance at 862 nm, which occurs immediately before the dip at 972 nm:

$$\text{Cacti Index} \, 1 = \frac{862 \,\text{nm} \, - \, 972 \,\text{nm}}{862 \,\text{nm} \, + \, 972 \,\text{nm}}$$

The second equation, Cacti Index 2 (CI2), uses reflectance at 1072 nm, which occurs immediately after the dip at 972 nm:

$$\text{Cacti Index} \, 2 = \frac{1072 \,\text{nm} \, - \, 972 \,\text{nm}}{1072 \,\text{nm} \, + \, 972 \,\text{nm}}$$

### 2.2.2. Taking the Cacti Index Airborne

Using the cacti index with aerial and drone-acquired imagery would greatly improve its utility for mapping and monitoring cacti populations across large tracts of land. We also wanted to examine the impact of spatial resolution on the uniqueness and clarity of the cacti signature. We tested the efficacy of the cacti index using both drone- (3 cm spatial resolution) and airplane-mounted (1 m spatial resolution) hyperspectral sensors.

To investigate the utility of the cacti index with drone scale imagery, we used a Nano Hyperspectral (https://www.headwallphotonics.com/products/vnir-400-1000nm (accessed on 23 May 2022)) Visual & Near Infrared (VNIR) sensor by Headwall Photonics gimbal mounted on a DJI Matrice 600 Pro 6 rotor copter, which is a push broom slit sensor with 640 linear array detectors. The data contain 270 bands ranging from 400 nm (blue) to 1000 nm (near infrared) at ~2.2 nm slices. In August 2021, we collected imagery over a 2 ha plot in SRER known to have a mix of cacti and non-cacti vegetation. We flew the drone ~65 m above ground level, yielding a spatial resolution of ~3 cm. Using Headwall software, we converted the raw imagery digital numbers to radiance and then to reflectance using a tarp with known reflectance values. Individual frames were then orthorectified and mosaicked into a stacked imagery product. On the stacked orthomosaic, we identified and extracted spectral signatures from 11 prickly pear, 7 mesquite, 10 barrel cacti, 10 patches of bare ground, and 10 samples of grasses. Due to known sensor noise in the near-infrared region of the spectrum, we employed a three-band moving average to smooth the spectral signatures. We calculated CI1 using bands 212 (864 nm) and 260 (970 nm) and extracted these values from the same vegetation samples. The sensor is not sensitive to radiation beyond 1000 nm, so we were unable to calculate CI2.

To investigate the utility of the Cacti Indices at an airplane imagery scale, we used NEON Airborne Observation Platform hyperspectral data collected across the SRER between August 24 and 29 of 2018. The sensor was a next-gen version of the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS-NG) (https://avirisng.jpl.nasa.gov/avirisng.html (accessed on 23 May 2022)). The data contain 426 bands ranging from 380 nm (blue) to 2500 nm (near infrared) at ~5.5 nm slices. The aircraft flew ~1000 m above ground level with a nominal spatial resolution of 1 m. NEON used ATCOR4r [30] to atmospherically correct the data and serve it as unitless surface reflectance values scaled by 10,000.

To facilitate the identification of individual cacti and non-cacti vegetation samples, we collected 35 ha of high-resolution (1.6 cm) RGB drone imagery using a DJI Mavic Pro multirotor. We found this method to be more efficient than locating samples on foot. We coregistered orthomosaics created from the drone imagery with the NEON AVIRIS imagery using ArcGIS Pro. We identified and extracted the spectral signature from many samples of cholla (N = 418), prickly pear (N = 325), mesquite (N = 105), palo verde (N = 100), creosote (N = 100), and bare ground (N = 300). We calculated CI1 and CI2 using AVIRIS bands 97 (862 nm), 119 (972 nm), and 139 (1072 nm).
