Forest canopy height estimation based on ICESat/GLAS data by airborne lidar

Abstract: The forest vertical structure parameters can reflect the growth status of the forest and the species diversity of the forest to a certain extent. Forest canopy height is a significant part of vertical structure parameters, and quantification of its distribution is an active academic research focus in recent years due to its important significance for the forest ecosystem research. Spaceborne lidar system ICESat/GLAS, with laser sensor placed on the satellite, is the full waveform lidar altimeter system with a 1064 nm laser operated at 40 Hz. The return GLAS waveforms can describe the vertical distribution of the landscape. However, large footprint diameter can reshape the vertical extent of waveforms for topographic change. Airborne lidar system whose laser sensor is placed on the airplane can transmit a short duration laser pulse. There may be several secondary returns as the light from a single pulse is reflected from within canopy layers of vegetation. The property of high-density laser point cloud (0.5-10 points/m2) makes some airborne lidar metrics to be suitable predictors of either canopy height or even individual tree height. In this study, combining with the SRTMGL1 terrain data and airborne lidar point cloud data, GLAS-based canopy height correction model based on physical equation was proposed. The effect of slope on the estimation accuracy was analyzed and estimated result was validated in in situ data. An irregular triangulation algorithm was used to filter the point cloud to extract the ground point. The initial triangulation model was optimized by setting the angle threshold and the height threshold. Point cloud normalized processing which eliminates the terrain effect could reflect the actual canopy height. The range of the GLAS footprint was used to extract the local maximum value of the normalized point cloud and analyzed canopy profile. This value was modeled as a true value for model parameters fitting by considering the slope, the footprint size and the surface roughness. GLAS-based canopy height indirectly was calculated from waveform parameters, it was necessary to use in situ data to analyze the estimated result. The result showed: airborne points cloud data can accurately reflect the distribution of forest canopy in the range of the GLAS footprint, but there were obvious differences in the distribution of points cloud in different forest types. By comparing uncorrected GLAS-based canopy height RH100 with airborne lidar canopy height, we found that the canopy height from RH100 was overestimated, which led to the positive bias. For the flat terrain cases, canopy and ground peak can be accurately identified from the waveform. GLAS metric RH100 can be approximately considered as forest canopy height because the last Gaussian peak was assumed to represent the ground peak. However, for the slope terrain case, the slope stretched the waveform, leading to increased waveform extent, and decreased canopy and ground peaks. For heterogeneous land surface cases, single slope information can not accurately reflect changes in the surface. The undulating surface made the waveform more complex. After topographic correction, not surprisingly, the fitting result was closer to the 1:1 fitting line. Reflecting on the root mean square error RMSE, uncorrected RMSE was 6.43 m greater than the corrected RMSE was 3.54 m. The effect of the topography was alleviated to a certain extent. As slope level increased, the RMSE of uncorrection increased from 3.55 m to 10.25 m, whereas the RMSE of correction had stabilized at between 3.26 m and 3.88 m. The effect of the topography was alleviated to a certain extent. Comparing with arithmetic average height, it can be seen that the estimation result was close to Lorey's height. In addition, the accuracy of canopy height estimation in different forest types was different, the accuracy of mixed forest was the highest (r and RMSE were 0.84 and 1.06 m), and the accuracy of broad-leaved forest was the lowest (r and RMSE were 0.44 and 2.56 m).