Analysis of ALS-based prediction errors:an implication for mapping tropical forests
Remarkably relevant to the global greenhouse gas emissions,tropical forests are of substantial meaning in battle against the climate change.In line with the Three-Tier accuracies of carbon inventory proposed by the International Panel on Climate Change,airborne laser scanning (ALS) has shown potential in favoring tropical forest mapping with top ranking accuracies.However,this promise is discounted when it comes to a comparison with those derived under boreal conditions as the mapping accuracy in tropics falls well behind.As a result,this study aims to investigate the causing factors that lead to such poor predictions in certain types of sample plots,at the field observation,point-cloud and modeling levels with data from Laos.ALS metrics were extracted as a function of vegetation height thresholds,due to difficulties to coerce an empirical threshold that may be respected by all plots.The one leading to the most predictive power in predicting plot-level stem volume (m3/ha) went to further analyses.According to the residuals,three groups of plot-level field attributes and ALS metrics were selected to represent the good,the over-predicted and the under-predicted.Nonparametric tests are thereafter conducted for investigating differences between groups,indicating the attributes and metrics that contribute to the group partition.Influencing the prediction qualities,physical condition or structural pattern of sample plots was qualitatively analyzed through comparing tally-tree records with the point-clouds.The results indicate that either the over- or the under-predicted plots are of own properties that are traceable and useful for improving mapping quality.Two prediction strategies are accordingly recommended,one by modeling for respective groups via post-stratification according to the detected properties,the other through robust regression.Analyses also show that the conventional canopy density metrics as proportional values are by definition insensitive to depict plots of complex canopy structures.A plot of touching top canopies with a dramatic height variation may possess similar metrics to those under homogenous conditions.In response,the “count of first returns above,or even below,corresponding height percentiles is proposed as complementary canopy density metrics.Implications of the analyses are of potential interest to REDD-related communities.
Zhengyang Hou Qing Xu Timo Tokola
European Forest Institute(EFI), Joensuu, Finland;University of Eastern Finland, Faculty of Science a University of Eastern Finland, Faculty of Science and Forestry, School of Forest Sciences, Joensuu,
国际会议
北京
英文
35-44
2013-10-09（万方平台首次上网日期，不代表论文的发表时间）