Hyperspectral and LiDAR data of the Botanical Garden of Rio de Janeiro

Description

Technical info (English)

On 9 March 2022, we collected hyperspectral and LiDAR data over the Botanical Garden of Rio de Janeiro. Hyperspectral images were acquired with the Nano-Hyperspec camera (Headwall Photonics, Inc.). This camera operates in the visible to near-infrared wavelength range (400-1000 nm) with a full width at half maximum (FWHM) of 6 nm, resulting in an image cube with 270 spectral bands. We used an 8 mm focal length lens to acquire the hyperspectral images. LiDAR data was acquired with the VLP-16 sensor (Velodyne Lidar, Inc.), which emits and receives 600,000 laser pulses (wavelength of 905 nm) in dual return mode. The sensor has a measurement range of up to 100 m and a vertical field of view of 30º(±15º). Nano-Hyperspec and VLP-16 are combined to form a single data acquisition unit (DAU) assembled with an Applanix APX-15 internal measurement unit (IMU) (Applanix, Inc.). APX-15 provides high-quality global navigation satellite system (GNSS) data and information on the DAU's velocity, pitch, row, and yaw during data collection. The DAU is mounted on the DJI M600 Pro (DJI, Inc.) using the DJI Ronin three-axis gimbal, ensuring unit stability during flights.

We manually delineated insdividual tree crowns (ITCs) visible in the hyperspectral image and adjusted their boundaries using the LiDAR-derived canopy height model (CHM). We delineated 205 ITCs using the QGIS software and identified them to the species level with field data. The ITCs corresponded to 90 species. For each ITC, we computed the aboveground biomass using the allometric model proposed by Velasco and Chen (2019).

Technical info (English)

LiDAR pre-processing

We first performed post-processing kinematics (PPK) procedures to improve the positional accuracy of LiDAR points. We used the POSPac-UAV software (Applanix, Inc.) to compute the smoothed best estimate of the trajectory (SBET). For SBET calculation, we used raw GNSS observables from the Trimble GNSS Smart Target Base Station (Applanix, Inc.). The base station recorded GNSS data at least one hour before and during the UAV flights. The LiDARTools software (Headwall Photonics, Inc.) uses SBET data to improve LiDAR positional accuracy. Moreover, this software automatically merges the .pcap files generated by the VLP-16 sensor and generate a single .las file for each flight. The final planimetric and altimetric accuracy of LiDAR points were ±4 cm and ±8 cm, respectively. Finally, we combined the .las files of each flight into a single .las using the lidR R package.

Hyperspectral data pre-processing

Hyperspectral pre-processing starts by converting raw cubes to radiance  W·m⁻²·sr⁻¹·nm⁻¹ using the SpectralView software (Headwall Photonics, Inc.). Then, radiance cubes are converted to surface reflectance using a calibration tarp with known diffuse reflectance. The reflectance of each pixel is computed by the ratio of its radiance to the mean radiance of the pixels of the calibration tarp. The illumination conditions did not change considerably during the 45 minutes of data acquisition. The surface reflectance cubes were orthorectified using a LiDAR-derived digital surface model (DSM). The orthorectification process is initiated by computing each flight line's input geometry (IGM) files. These IGM files contain essential georeferencing information for every original raw pixel in the image. Subsequently, we mosaicked the reflectance images of all 18 flight lines and resampled them to a pixel size of 40 cm, aiming to enhance the signal-to-noise (SNR) ratio.