Prospects of photon counting lidar for savanna ecosystem structural studies

Abstract

ABSTRACT:Discrete return and waveform lidar have demonstrated a capability to measure vegetation height and the associated structural attributes such as aboveground biomass and carbon storage. Since discrete return lidar (DRL) is mainly suitable for small scale studies and the only existing spaceborne lidar sensor (ICESat-GLAS) has been decommissioned, the current question is what the future holds in terms of large scale lidar remote sensing studies. The earliest planned future spaceborne lidar mission is ICESat-2, which will use a photon counting technique. To pre-validate the capability of this mission for studying three dimensional vegetation structure in savannas, we assessed the potential of the measurement approach to estimate canopy height in a typical savanna landscape. We used data from the Multiple Altimeter Beam Experimental Lidar (MABEL), an airborne photon counting lidar sensor developed by NASA Goddard. MABEL fires laser pulses in the green ( 532 nm) and near infrared (1064 nm) bands at a nominal repetition rate of 10 kHz and records the travel time of individual photons that are reflected back to the sensor. The photons' time of arrival and the instrument's GPS positions and Inertial Measurement Unit (IMU) orientation are used to calculate the distance the light travelled and hence the elevation of the surface below. A few transects flown over the Tejon ranch conservancy in Kern County, California, USA were used for this work. For each transect we extracted the data from one near infrared channel that had the highest number of photons. We segmented each transect into 50 m, 25 m and 10 m long blocks and aggregated the photons in each block into a histogram based on their elevation values. We then used an expansion window algorithm to identify cut off points where the cumulative density of photons from the highest elevation resembles the canopy top and likewise where such cumulative density from the lowest elevation resembles mean ground elevation. These cut off points were compared to DRL derived canopy and mean ground elevations. The correlation between MABEL and DRL derived metrics ranged from R 2 = 0.70, RMSE = 7.9 m to R 2 = 0.83, RMSE = 2.9 m. Overall, the results were better when analysis was done at smaller block sizes, mainly due to the large variability of terrain relief associated with increased block size. However, the increase in accuracy was more dramatic when block size was reduced from 50 m to 25 m than it was from 25 m to 10 m. Our work has demonstrated the capability of photon counting lidar to estimate canopy height in savannas at MABEL's signal and noise levels. However, analysis of the Advanced Topography Laser Altimeter System (ATLAS) sensor on ICESat-2 indicate that signal photons will be substantially lower than those of MABEL while sensor noise will vary as a function of solar illumination, altitude and declination, as well as the topographic and reflectance properties of surfaces. Therefore, there are reasons to believe that the actual data from ICESat-2 will give poorer re...

Voxel-Based Spatial Filtering Method for Canopy Height Retrieval from Airborne Single-Photon Lidar

“…The histogram method was a typical approach used to process photon counting lidar in many recent studies [4,8,10,11]. In this approach, photon-counting data were converted into pseudo-waveforms by aggregating lidar returns in user-defined elevation bins and footprint sizes.…”

“…In this study, we created SPL histograms over 71 field plots using a 0.15 m vertical resolution and a footprint radius of 15 m. The applied resolution and scale was identical to both field survey and previous studies [4,8,10,11]. We first normalized each histogram between 0 and 1 to compensate for spatial difference in photon density.…”

“…It therefore allows enhanced 3D mapping with greater coverage, spatial resolution, and photon density, and reduced acquisition time. Despite these advantages, SPL data includes more solar background noise than conventional near-infrared lidar instruments because of its use of green wavelength lidar and high photon detecting sensitivity [4][5][6]. This complicates the retrieval of terrain and canopy structural information from SPL data.…”

“…Algorithms for efficiently handling noise in photon counting data are only beginning to be developed [7][8][9]. Existing methods largely rely on detecting maximum canopy height through histogram-based filtering algorithms [4,8,10,11]. This is achieved by aggregating point clouds into pseudo-waveforms at coarse spatial and vertical resolutions.…”

“…Traditional waveform processing algorithms are then applied to the waveforms to derive canopy structure metrics including canopy height and cover. Although these methods have proven effective in filtering solar noise [4,8,10,11], the 3D structural detail in the dataset is reduced. It means that these methods forfeit a key advantage of SPL data over other systems-the ability to provide fine scale topographic and structural details.…”