Accuracy of vegetation height and terrain elevation derived from ICESat/GLAS in forested areas

Abstract

This paper focuses on accuracy assessment of canopy top elevation, ground elevation and vegetation height (VH) derived from space borne full-waveform LiDAR (Light Detection And Ranging) data across forested areas. Computed height metrics from LiDAR data which were acquired by the GLAS sensor aboard the ICESat (Ice Cloud and land Elevation Satellite) are compared against airborne laser scanning (ALS) based digital elevation models. Due to the dynamic topography of the sites under investigation, a wide range of slope angles could be investigated. ICESat's raw waveform data (GLA01) and the land surface altimetry data (GLA14) products are used to determine height metrics with different methods. GLA14 based elevation and vegetation heights are computed from range offset information. Values are provided for signal begin, signal end, land range and up to six Gaussian peaks for each received waveform. GLA01 based terrain heights are computed by locally weighted polynomial regression and peak detection on the received waveform itself. A range of different smoothing spans and noise threshold values on the original waveform, which is represented by 544 single values (bins), were tested. A new method based on the unsmoothed waveform was developed for the detection of the signal begin. By detecting the location above the noise threshold, where the signal rises at least for 5 bins (75cm), achieved more precise results, than the given signal begin in the GLA14 product. For ground peak detection by smoothing of the waveform it was found that noise thresholds of 4 and 4.5 times the standard deviation plus the mean noise level give the best performance. For VH computation in areas of up to 10° terrain slope, a smoothing span of 10 bins achieved r2=0.58, whereas the GLA14 based method achieved r2 of 0.75 in flat terrain (0–5°). For these flat areas, best results in VH computation (r2=0.91) were achieved by using the new method for canopy top detection and the GLA14 based ground elevations. Determination of terrain elevations was observed to be best by computation with GLA14 based parameters. The stronger of second last two peaks was found to be closest to the ground level elevation.