BayCEER Workshop 2017

A look beneath - LiDAR based classification of forest structure in the Black Forest National Park

P 4.6 in BayCEERversity: Across scales, compartments & communities

Forests constitute habitat for more than 75% of the world’s terrestrial biodiversity. In the Black Forest National Park diverse mosaic forest structures are supposed to supersede large-scale structures under conservation of natural dynamics with considerable effects on biodiversity. A comprehensive ecosystem monitoring is planned to document these processes.

The study aims to adapt a stratification of forest structure to local, conifer dominated forests and classify the forest strata across the national park by the means of LiDAR remote sensing. Six strata, differentiated by tree dimensions and heterogeneity, and a complementing opening stratum were defined. A discrete return LiDAR data set (30 returns/m²) was used for an area-based forest structure analysis. LiDAR metrics were calculated for a gridded national park raster of a 20 m x 20 m resolution. For model training and test purpose 86 forest observations across the six forested strata were selected. Additional 15 opening plots were located on meadows. Stratum classification by Random Forests modelling was chosen due to its positive results in comparable studies.

Random Forests integrated recursive feature elimination reduced the metrics used in the final model to 13, mainly density and height related predictors, which were partially highly correlated. The classification reached an overall OOB accuracy of 81.47% (K = 0.78) and a test accuracy of 90.00% (K = 0.88). The intra-stratum accuracies varied. Low vegetated strata were perfectly classified whereas multi-story strata showed moderate accuracies. Stratum membership analysis confirmed an on average good stratum identification.

The study proved that the introduced stratification defined well differentiable forest structure strata. It revealed well strata predictions with respect to accuracy measures and to spatial distribution. In conclusion, a reliable forest strata classification was shown, providing a comprehensive approach for repeated monitoring assessments.

 

Keywords: LiDAR Forest structure Forest stratification Remote sensing monitoring Black Forest National Park Random Forests