Eberswalde, 4.05.2012Accuracy comparisonMilosz MielcarekFIT 3rd semester studentResearch Colloquium Presentation
• 1977 - Potential utilization of LIDAR by Soloduchin• 1993 - establishment of first operational laser system -first ALS system dedicated to topographic mapping.• 1993-2000 – development of system...• +2000 - based on surface models of forested areas fromLIDAR data, it is possible to capture many different treeand forest parameters (Olsson, 2004, Hyyppä et al.,2004) with very accurate results.
• The study area is situated in theTable Mountains National Park(south-western Poland)• For research purposes two sampleplots were set:1. Common Spruce (687 trees)2. Larch sp. (315 trees)
• Photogrammetric measurementswere performed on DEPHOS PCphotogrammetric station• High accuracy of surveys• Due to the very accurate resultsobtained during the photogrammetricobservations, thesedata were used as reference data
• Canopy Height Model (CHM)was generated by subtractingthe height value- "Z", recordedin the corresponding pixel ofDigital Surface Model (DSM)and a Digital Terrain Model(DTM)• Utilization of median filter inorder to improve the accuracyDSMDEMCHMPicture source: www. mnr.gov.on.ca
Picture source: www.lidarcomm.com
• Crown delineation was performed automatically usingScanLas Software• Created contours of crownsthat were not clearly assignedto one top of the tree from thereference data have been removedand were not takeninto account during furtheranalysis
• The distance betweensubsequent laser beamsmight be large enoughto miss some of thetreetops• Increase of randomerrors- affecting finalresults• Median filter is reducing random errors (noise reduction)
• Determination of the Real Tree Height :Stereo Photogrammetric Surveys – DEM = Real Tree Height• Determination of the CHM based Tree Heights :- Establishing maximum pixel value of CHM in each segment- Substracting DEM from obtained values• Comparison of Real Tree Heights with CHM Tree Heights- Statistic calculations
Comparison of segmentation accuracy100908070605040302010079,6084,00Correctdetection [%]17,0015,00Not found [%]3,401,00Double peakoccurrence [%]Common SpruceLarch sp.
Paired Samples ComparisonSpecies Pairs Bias[m]CommonSprucePair 1real_heigh -CHM_medianStd.Deviation[m]RMSE [m]-1.028 .826 1.320 .988Pair 2 real_heigh - CHM -.338 .796 .860 .989RLarch sp.Pair 1real_heigh -CHM_median-.515 .579 .770 .973Pair 2 real_heigh - CHM -.021 .563 .560 .975
• Segmentation accuracyCorrect detection [%]• Height accuracyRMSE [m]100908081,8077,00 76,301,000,900,800,710,8270600,700,600,53500,50400,40300,30200,20100,1000,00Mielcarek (2012) Sterenczak (2011) Heurich (2004)
• Height estimation accuracy was considered to besatisfactory for typical forest inventory issues• A tendency to underestimate the height of tree usingLIDAR data was revealed (negative Bias errors)• Median filter applied to CHM did not match theexpected increase in accuracy• The mean accuracy of single tree segmentation (81.8%)is comparable to other studies conducted on uppercrown layers (Heurich and Weinacker 2004; Persson2004; Sterenczak and Zasada 2011)
• Airborne Laser Scanning is competitive with traditionaland photogrammetric methods of forest parametersmeasurements due to the automation and accuracy ofsurveys.• Median filter does not improve the accuracy of treeheight estimation.• ALS is an effective and reliable method of obtainingdata for forest inventory.