WSP11909 - Laserfiche WebLink

Home Browse Search

I 1502 I I "ground control" arcas within various river reaches and thus LIDAR has assumed one of the main disadvantages of the photogrammetric approach. The mean error in the photogrammetry data for bare sand is -3 em :i:. 33 em, whereas the mean absolute error (vertical accuracy) is 17 em :i: 28 em, which is close to the preferred vertical accuracy for GCMRC sediment monitoring and better than the 25-cm accuracy found in some previous studies (Blank. 2000; Gomez Pereira and Wichersoo, 1999). On cobble bars, we found the unadjusted venical accuracies in ALMS and RAMS data to be 29 em :::l: 25 em and 18 em:t 17 em, respectively. The adjusted vertical accuracies (using bare-sand offsets) are not appreciably different. The RAMS data for cobble bars appear to be as accurate as the data obtained for smooth, bare.sand surfaccs. Photogrammetric data on cobble bars were found to have a vertical accuracy of 11 cm .:t 11 em. This error is comparable to the photogrammetric errors found on bare.sand surfaces and is much lower than the L1DAR crrors on cobble surfaces. I Vertical accuracy within vegetation We examined photogrammctry and LIDAR elevation data within our vegetated terrain to determine the accuracies of these data for producing reliable bare-ground elevations. 475-700 surveyed ground elevations within vegetated parts of oUI four study areas were used in this evaluation. LIDAR elevations were adjusted for their bare-ground vertical offsets. Only one LIDAR survey (ALMS) in just one study area produced a vertical accuracy less than 20 em. ALMS LlDAR provided higher vertical accuracies in the vegetated terrain than the RAMS LlDAR; a larger fraction of the RAMS elevations points fall far above the bare-ground surfaces than do the ALMS elevations. The higher ALMS accuracies in our vegetated terrain can only be attributed to its March acquisition date, when leaf-off conditions prevail in the upper Colorado River ecosystem. The vertical accuracy of the photogrammetry data within the vegetated terrain (37 em) is not at our level of acceptable (~20 em) vertical accuracy. However, its vertical accuracy is 2-4 times better than that provided by the LIDAR data. The continuous, high-resolution imagery used in photogrammetry better samples bare ground within vegetated areas. Part of the photogranunetric error is due to a small, anomalous elevation mound erroneously produced by the contouring procedure. One reason we selected the RAMS system for our tests was the detector's purported ability to distinguish multiple returns from a single laser pulse. However, within our four study areas only 0.2% of the RAMS first returns had corresponding second returns, even though a large proportion of the vegetation is 4-6 m high. We examined the RAMS second-return elevations to determine their ability to map bare ground in our vegetated regions. Because we did not ground survey the second-return data, we had to compare the second-retum elevations with first-return elevations from nearby, unambiguous bare-ground. These data show that the average absolute difference between the second~ return elevations and nearby first-return "ground" elevations is 32 em. Although this error is greater than that found for the RAMS data on bare-sand surfaces (22 cm), it is significantly lower than the 147 cm error found for the RAMS first- return elevations in this vegetated terrain. This error is also comparable to the error found for the photogrammetry data (37 cm). This result might suggest LlDAR systems capable of recording multiple returns could compete with photogrammetry in vegetated terrain, however, many of the vegetation stands within our study areas average 5-6 m in height, but very few second returns were recorded within or even along the perimeters of these stands. The paucity of LlDAR second returns produced in this ecosystem suggests that LIDAR (at least the RAMS system) may not provide better ground topographic mapping than photogrammetry in this environment. In addition, the most ambiguous LlDAR results are obtained on sand surfaces with numerous, scattered willow and arrowweed bushes, which are common in this ecosystem. These bushes arc ~ m high, have a low branch density. and are difficult to discern in the 18-cm to 30- cm panchromatic imagery that is commonly collected with LJDAR data. As a result, LIDAR topography in such terrain appears quite irregular, similar to that produced by a boulder field. Removal of the LlDAR points thatprobab/y hit the bushes is ~xtremely time consuming due 10 the difficulty in unambiguously determining where the bushes are within low-resolution panchromatic image data. Photogrammetry generally collects very high resolution (about 5 em) CIR or natural-color imagery for GCMRC monitoring purposes, in which the scattered bushes can easily be discerned and mass points can be unambiguously identified. . Although more accurate ground elevations within riparian vegetation might be obtained by even higher resolution LlDAR data (e.g., spot spacing of s;'l m and spot diameter of S;O.25 m), such data will require a lower flight AGL, may preclude simultaneous image acquisition due to higher ground velocities, and will no doubt greatly increase collection.costs. Reproducibility Reproducibility of topographic data is important for monitoring changes within the Colorado River ecosystem. We assessed the reproducibility (precision) ofthe high resolution RAMS LIDAR data that were acquired for the study areas at two separate times using the same instrument and flight conditions. On bare.ground surfaces, the unadjusted mean absolute differences between the two data sets arc in the range of 10.34 em, whereas the adjusted mean :bsolute differences are in the range of 9.26 em. On vegetated surfaces, the unadjusted and the adjusted mean absolute EV ALUAll0N OF LIDAR AND PHOTOGRAMMETRY FOR MONITORING VOLUME CHANGES IN RIPARIAN RESOURCES WITHIN 11iE GRAND CANYON, ARIWNA Pecora tSfL.nd S.telllte Inrormation IVIISPRS Commission IIFIEOS 2002 Conrerence Proceedings