**4.2 Photonic point cloud denoising**

The extremely high sensitivity of photon-counting LiDAR detection leads to relatively poor data signal-to-noise ratio, and although a narrow-band filter (0.15 nm) is

**Figure 8.**

*Accuracy assessment of ICESat-2 using airborne data. (a) Study area; (b) horizontal offset, and red point is best-fitted.*

installed, there is still a large amount of background sunlight noise. In some high solar angle and high ground reflectivity scenarios, the background light noise rate reaches about 10 MHz (i.e., 10 million/s, which translates to 1 noise point per 3 m in the elevation direction), so point cloud denoising is critical.

The extremely high sensitivity of photon counting LiDAR detection also leads to a lot of noise and relatively poor signal-to-noise ratio of the data. Although the ICESat-2 receiver is fitted with a narrow-band filter that limits the band range to 532.272 0.15 nm, there is still a large amount of background sunlight in that range. In some high-sun angle and in high ground reflectivity scenarios, the background light noise rate reaches about 10 MHz (i.e., 10 million/s, which translates to 1 noise point per 3 m in the elevation direction) [37], so point cloud denoising is critical.

Most of the currently available photon-counting LiDAR devices record data only along the direction of flight (a swing scan is less common) and therefore are usually processed in a two-dimensional profile. Two denoising algorithms, including the histogram and spatial density methods, are provided in the ICESat-2 basic theory algorithm documents ATL03 [46] and ATL08 [47], respectively: the histogram method considers that the location with the highest number of points in the vertical direction is more likely to be the signal [37]; the spatial density method considers that the signal points are more densely distributed in space, and the density histogram will show the distribution characteristics of "noise on the left, signal on the right" and "high noise and narrow signal" [36]. Based on the "double-peak" distribution of the density histogram, the DRAGANN (Differential, Regressive and Gaussian Adaptive Nearest Neighbor) algorithm is proposed, which uses two Gaussian functions to fit the noise and signal separately and calculates the noise removal threshold adaptively by computing the optimal parameters. The comparison of the two denoising effects is shown in **Figure 9**. The ATL03 algorithm works better in the flat ice cover area, but in the vegetated area, there will be obvious signal point leakage, and ATL08 is more suitable for the vegetated terrain area [48]. In addition, the targeted design on the issues of search kernel shape, terrain correlation, and directional adaptivity can further improve the algorithm performance and obtain better than 98% denoising accuracy [49].

### **4.3 LiDAR parameters calibration**

For a LiDAR measurement satellite similar to ICESat-2 with a 500 km orbital altitude, a pointing angle error of 1 arc second results in a horizontal geolocation error *Spaceborne LiDAR Surveying and Mapping DOI: http://dx.doi.org/10.5772/intechopen.108177*

**Figure 9.** *Comparison of photon counting LiDAR point cloud denoising effects. (a) ATL03 algorithm; (b) ATL08 algorithm.*

of about 2.4 m, and an elevation error of 8.3 cm if the ground slope is 2°. Therefore, the rigorous in-orbit calibration of LiDAR work parameters is crucial to the accuracy of elevation measurements.

The on-orbit calibration of satellite-based LiDAR has some similarities with the traditional optical satellite and airborne LiDAR, mainly based on the ground calibration field method or the nature terrain method [50, 51]. The ground-based calibration field directly measures the footprint point through the laser receiver, which has the highest accuracy, but it is necessary to estimate the location of the footprint point and select a suitable location to build the calibration field. The natural terrain method realizes the laser calibration parameter solution by profile alignment with the satellite laser data through the local terrain measured accurately in advance. The rigorous geometric location equation for satellite-based LiDAR is adopted to the basic calibrated model. The iterative pointing angle calibration method was proposed based on the least elevation difference matching criterion. The main steps include two steps: firstly, terrain matching is used to obtain the common terrain feature points whose coordinate deviations are used as input observations, and the second step is to calculate the system calibration parameters by laser beam adjustment. As shown in **Figure 10**, the specific solution can be simplified to calibrate two pointing angular parameters (θ0, β0) and 1 range parameter r. It has been shown that the angular calibration accuracy is better than 0.3 arc second using 1 km length laser line matched

**Figure 10.** *Diagram of laser on-orbit calibration based on natural terrain.*

with high-precision terrain, and the angular calibration accuracy is better than 0.1 arc second when the line length is increased to 2.5 km [52]. In addition, ICESat-2 adopts a similar strategy to ICESat for attitude maneuvering in the oceanic region, and the attitude and distance are calibrated separately by conical scanning, and the long-term drift of the calibrated range values is less than 1 mm/year [53].
