Identification and characterization of urban trees using VHR remote sensing and auxiliary data

Abstract

Trees play an important beneficial role within urban climate. Planning and maintenance of urban forest requires measurements of position, distribution, and characteristics of individual trees, which can be achieved with VHR remote sensing sensors. This work compares local maximum (LM) filtering and Laplacian of Gaussian (LoG) blob detector combined with marker-controlled watershed segmentation (MCWS), clustering with Voronoi-tessellation and region growing for individual tree detection and crown delineation (ITCD) in the urban area of Munich. Crown height model (CHM), lightness and distance transform of crown area from VHR aerial imagery are employed for ITCD. Region specific parameters are set with the inclusion of additional information of street green area. The trees are categorized by land-use into Street, Residential and Park trees. Tree height, crown area and diameter, and normalized difference vegetation index (NDVI) are measured for all delineated tree crowns. The ITCD result is validated against reference data from visual interpretation via stereophotogrammetry, then compared to a TLS street tree product and the tree classes of a VHR satellite landcover classification. The distribution and density of trees is analyzed over the urban area of Munich aggregated to 100 m INSPIRE GeoGitter, city districts and centrality. Tree detection with LM filtering on CHM with regional calibration based on street green yields highest results with an F-Score of 0.949, precision of 0.959, and recall of 0.939. Performance without region specific parameters decreases to an F-Score, precision, and recall of 0.900, 0.936 and 0.866 respectively. Tree crown delineation performs best with LM and CHM with overall accuracies between 73.5 % to 76.7 % for all delineation methods, regional restriction to street green resulted in overall accuracies of 86.9 % - 88.1 %. With less a priori information required, delineation with MCWS is preferred. The detection method LM CHM with MCWS results in 1.54 million trees with a combined crown area of 92.24 km2 for the administrative city area of Munich in 2017. Mean tree height and crown area of individual trees is 12.45 m and 60.3 m2 respectively. Categorization on land-use shows 9.1 %, 38.4 %, and 33.1 % of trees belonging to Street tree, Residential tree, and Park tree with relative crown area of 7.5 %, 30.4 %, and 45.5 %, respectively. The ITCD showed an accordance of 80.9 % of distinct trees with a TLS street tree product and high correlation to a VHR satellite landcover classification. The inner city of Munich displays low tree density compared to surrounding areas. The district Waldperlach has the highest tree density with an average of 83 trees per hectare. Crown coverage increases with distance to city center and levels off at a distance of 3 km with 30 % relative coverage.