Mapping urban tree species using integrated airborne hyperspectral and LiDAR remote sensing data
Introduction
Urban tree species play an important role in the urban environment by providing a range of ecosystem services (Gu et al., 2015), such as isolating noise (Roy et al., 2012), sequestering carbon through photosynthesis (Davies and Edmondson, 2011), mediating urban temperature, mitigating the urban heat island effect (Armson et al., 2012), alleviating urban flood risk (Zimmermann et al., 2016), and providing shelters for wildlife (Goddard et al., 2010). Besides these many ecological and environmental services, urban trees also provide important psychological and social benefits to human societies (Young, 2010), such as potentially reducing crime (Kuo and Sullivan, 2001), encouraging people to build stronger social relationships (Peters et al., 2010), and improving residential property value (Anderson and Cordell, 1985).
The quality of services that urban trees can provide depends on the species, structure, and locational context of the trees (Escobedo et al., 2011, Escobedo and Nowak, 2009). For example, tree species differ significantly in their ability to reduce air and surface temperature, as well as to increase relative humidity (Gillner et al., 2015). In addition, tree species provide different habitats for wildlife, primarily birds and small mammals. Consequently, urban managers recognize the importance of diversifying urban forests to create a more sustainable urban environment (Raupp et al., 2006). Historically exotic tree species were introduced to aesthetically embellish urban environments. However, exotic species also bring additional challenges to urban managers due to their potential to increase pests and diseases and cause other issues (Watson and Adams, 2010), and as a result native tree species are emphasized and encouraged for use in urban sustainable development. Although there has been continuing debate about the role of native vs. exotic species in urban environments, it is clear that accurate identification and mapping of tree species is essential to develop strategies for city managers for sustainable urban development and planning (Li et al., 2015, Pu and Landry, 2012).
Traditionally, information about urban forest canopy and species composition has been obtained from field sampling and manual interpretation of aerial photography. In addition, many cities utilize inventory systems to collate tree location, species and condition information if trees are specifically planted. However, these methods are expensive, generally labor-intensive, time-consuming, and usually unable to provide complete coverage (Alonzo et al., 2014). Contrary to extensive field sampling, aerial or satellite remote sensing imagery have the advantage of providing spatially-explicit data, potentially with fine temporal resolution and lower cost (Boyd and Danson, 2005, Masek et al., 2015). Recent, very high spatial resolution optical satellite sensors such as QuickBird, GeoEye, and WorldView provide around one half-meter spatial resolution and have increased the potential to classify tree species within complex urban environments (Novack et al., 2011, Richardson and Moskal, 2014). In addition, intelligent image segmentation and object-based classification techniques have been shown to be highly useful for urban remote sensing applications (Myint et al., 2011). However, analysis of spectral data is limited in its capacity to quantify structural attributes, while shading, canopy illumination, and background effects present additional challenges to extracting useful information on urban trees (Zhou et al., 2009).
Airborne Light Detection and Ranging (LiDAR) systems represent one of the most accurate methods for measuring structural attributes and biophysical parameters of vegetation from local to regional scales. Based on either the range or intensity of individual pulse returns or the characterization of full pulse waveforms, LiDAR data can be used in conjunction with field data and models to estimate vegetation parameters, such as tree height, biomass, stand density, basal area, volume, and Leaf Area Index (LAI) at both the individual tree and the stand level (Edson and Wing, 2011, Hudak et al., 2006, Næsset, 2007, Riaño et al., 2004). Kim et al., 2011, Kim et al., 2009 used mean intensity values and structure variables (such as height percentiles, standard deviations (SD) of heights, and length to width ratios of a crown) of laser returns within individual tree crowns, derived from leaf-on and leaf-off airborne LiDAR data, for tree species differentiation. The accuracy was found to be higher using leaf-off data when using intensity values. Conversely, structure values performed better when leaf-on data were used. Additional intensity and structure parameters such as point distributions, intensity and tree crown shape index were proposed by Lin and Herold (2016). LiDAR waveform attributes such as peak power, shape, length, beam divergence, and within-footprint irradiance distribution also provide additional reflectance information and these attributes have proven beneficial for species classification in a number of studies (Cao et al., 2016, Hovi et al., 2016, Yao et al., 2012).
Due to the fine spectral resolution offered, airborne hyperspectral remote sensing is also well-suited to classifying tree species in regions with high species diversity. However, hyperspectral data come with the disadvantage of high data dimensionality due to the Hughes phenomenon (Richter et al., 2016). Many researchers have used a number of feature reduction methods such as InStability Index (ISI) (selecting a subset from the original bands, Somers et al., 2010) and Principal Component Analysis (PCA) (applying statistical transformations to create a reduced set of variables) (Bajorski, 2011) prior to species classification. Similarly, spectral vegetation indices (SVIs) are generally derived from a number of selected wavelengths to detect changes in plant physiology (Mahlein et al., 2013), and offer an alternative to feature extraction. These indices, based on information from a few significant spectral wavelengths, have been developed to measure a range of plant attributes, such as pigment content in the visible 0.4–0.7 μm range (Fassnacht et al., 2015), leaf area and canopy structure (da Luz and Crowley, 2010, Neinavaz et al., 2016), or water content (Zhang et al., 2015) in near infrared 0.7–1.1 μm range. For tree species classification, SVIs have potential to distinguish species with different phenophases in diverse urban environments (Naidoo et al., 2012, Shang and Chisholm, 2014).
Combined airborne hyperspectral and LiDAR datasets provide both vertical and horizontal information and have shown great potential in improving tree species identification (Zhang et al., 2016). For example, Dalponte et al. (2012) increased classification accuracy from 74.0% to 83.0% using fused datasets. Hyperspectral and LiDAR data can be combined at either the pixel or object level. At the pixel level, the fusion of hyperspectral and LiDAR data increased both producer's (5.1–11.6%) and user's (8.4–18.8%) accuracies as found by Jones et al. (2010). Dalponte et al. (2012) found more structural features could be extracted from high density (8.6 points per m2) LiDAR yielding higher species classification accuracy (75.3%) than low point density (0.48 points per m2) LiDAR data (73.0% accuracy) (Dalponte et al., 2012). Object-oriented approaches, such as intelligent image segmentation (Koch et al., 2006) or LiDAR point cloud segmentation (Moradi et al., 2016) have been used to create discrete crown-objects from which individual tree characteristics can be derived. Object-specific metrics such as crown size, tree height or measures of the vertical distribution of LiDAR returns can be used as additional variables for species classification.
In this study, we evaluate the capacity of combining high density (25 points per m2) airborne LiDAR point cloud data with high-spatial resolution (1 m) hyperspectral imagery to map 15 tree species in the city of Surrey, British Columbia, Canada. We utilize data acquired under early spring conditions which results in some deciduous species experiencing leaf off conditions while others are in various stages of bud break and flowering. Tree crown structure information and spectral indices were derived from LiDAR and hyperspectral data respectively within crown objects. A random forest classifier was used to classify the major species in the city with the contribution of each input variable evaluated. Our specific objectives are: 1) to classify 15 common tree species using a fusion of hyperspectral and LiDAR data across the City of Surrey; 2) to evaluate the accuracy of the tree species classification under early spring conditions; and 3) to investigate what spectral and structural features are the most appropriate for distinguishing tree species under these early spring conditions.
Section snippets
Study area
The City of Surrey (49°11′N, 122°51′W) with in the Greater Vancouver regional district is locate on the west coast of British Columbia, Canada (Fig. 1) and covers 316 km2. More than 90,000 trees are actively managed within the city boundary, with an additional 3500–5000 planted every year (Plowright et al., 2016). A database containing the species, geographic location, and planting date for each of these trees is maintained by the city. While Surrey's tree stock is composed of over 300 different
Results
Fig. 4 shows the tree delineation using the watershed method (red line) and the manual tree delineation (blue line). The watershed method has improved performance when delineating trees in open areas compare to trees surrounded by other trees or other urban infrastructure.
Fig. 5 shows the average reflectance curves for each of the 15 species along the 72 hyperspectral bands. As expected, species occupy similar spectral ranges, especially in the visible portion of the spectrum. American
Discussion
In this study, we examined the capacity of airborne LiDAR and hyperspectral remote sensing data to accurately classify common urban tree species using a range of structural and spectral variables in isolation and in combination. Structural and spectral features were analyzed using a RF classifier which has been proved to be effective in tree species classification at the individual tree level (Naidoo et al., 2012). The analysis revealed that a combination of both LiDAR and hyperspectral
Conclusion
This study has demonstrated the application of airborne LiDAR and hyperspectral data to effectively identify and classify urban tree species in the city of Surrey at the individual tree level. The classification enabled accurate discrimination of 15 common tree species. Compared with previous studies, the accuracies in this study are similar, if not slightly lower, however they were obtained across a wider range of species. Importantly, iconic evergreen species had higher accuracies than other
Acknowledgements
Funding for this project was provided by an NSERC Discovery (RGPIN 311926-13) grant to Coops and NSERC Engage and Engage + and CRD (EGP 462042-13, EGP2 476350-14 and CRDPJ 488240-15 respectively) collaboration grants with Surrey City Energy. We thank Andrew Plowright, Txomin Hermosilla, Xuan Guo, Curtis Chance and Henry Flanagan for research insights and editorial assistance. We also thank the anonymous reviewers for helpful suggestions.
References (82)
- et al.
Urban tree species mapping using hyperspectral and lidar data fusion
Remote Sens. Environ.
(2014) - et al.
The effect of tree shade and grass on surface and globe temperatures in an urban area
Urban For. Urban Green.
(2012) - et al.
Random forest in remote sensing: a review of applications and future directions
ISPRS J. Photogramm. Remote Sens.
(2016) Quantifying chlorophylls and caroteniods at leaf and canopy scales: an evaluation of some hyperspectral approaches
Remote Sens. Environ.
(1998)- et al.
Tree species classification in subtropical forests using small-footprint full-waveform LiDAR data
Int. J. Appl. Earth Obs. Geoinf.
(2016) - et al.
Early detection of plant stress by digital imaging within narrow stress-sensitive wavebands
Remote Sens. Environ.
(1994) - et al.
Hyperspectral discrimination of tropical rain forest tree species at leaf to crown scales
Remote Sens. Environ.
(2005) - et al.
The role of spectral resolution and classifier complexity in the analysis of hyperspectral images of forest areas
Remote Sens.
(2009) - et al.
Tree species classification in the Southern Alps based on the fusion of very high geometrical resolution multispectral/hyperspectral images and LiDAR data
Remote Sens. Environ.
(2012) - et al.
Remote sensing of species mixtures in conifer plantations using LiDAR height and intensity data
Remote Sens. Environ.
(2007)
Spatial heterogeneity and air pollution removal by an urban forest
Landsc. Urban Plan.
Urban forests and pollution mitigation: analyzing ecosystem services and disservices
Environ. Pollut.
Non-destructive estimation of foliar carotenoid content of tree species using merged vegetation indices
J. Plant Physiol.
Role of street trees in mitigating effects of heat and drought at highly sealed urban sites
Landsc. Urban Plan.
Scaling up from gardens: biodiversity conservation in urban environments
Trends Ecol. Evol.
Hyperspectral vegetation indices and novel algorithms for predicting green LAI of crop canopies: modeling and validation in the context of precision agriculture
Remote Sens. Environ.
Detecting Tamarisk species (Tamarix spp.) in riparian habitats of Southern California using high spatial resolution hyperspectral imagery
Remote Sens. Environ.
Correction of laser scanning intensity data: data and model-driven approaches
ISPRS J. Photogramm. Remote Sens.
Identifying species of individual trees using airborne laser scanner
Remote Sens. Environ.
LiDAR waveform features for tree species classification and their sensitivity to tree-and acquisition related parameters
Remote Sens. Environ.
Assessing the utility of airborne hyperspectral and LiDAR data for species distribution mapping in the coastal Pacific Northwest, Canada
Remote Sens. Environ.
Tree species differentiation using intensity data derived from leaf-on and leaf-off airborne laser scanner data
Remote Sens. Environ.
Classifying individual tree genera using stepwise cluster analysis based on height and intensity metrics derived from airborne laser scanner data
Remote Sens. Environ.
GeoDMA—Geographic Data Mining Analyst
Comput. Geosci.
Tree species classification based on explicit tree structure feature parameters derived from static terrestrial laser scanning data
Agric. For. Meteorol.
A comprehensive but efficient framework of proposing and validating feature parameters from airborne LiDAR data for tree species classification
Int. J. Appl. Earth Obs. Geoinf.
Development of spectral indices for detecting and identifying plant diseases
Remote Sens. Environ.
The role of remote sensing in process-scaling studies of managed forest ecosystems
For. Ecol. Manag.
Per-pixel vs. object-based classification of urban land cover extraction using high spatial resolution imagery
Remote Sens. Environ.
Classification of savanna tree species, in the Greater Kruger National Park region, by integrating hyperspectral and LiDAR data in a random forest data mining environment
ISPRS J. Photogramm. Remote Sens.
Retrieval of leaf area index in different plant species using thermal hyperspectral data
ISPRS J. Photogramm. Remote Sens.
Effects of different sensors and leaf-on and leaf-off canopy conditions on echo distributions and individual tree properties derived from airborne laser scanning
Remote Sens. Environ.
Social interactions in urban parks: stimulating social cohesion?
Urban For. Urban Green.
Assessing urban tree condition using airborne light detection and ranging
Urban For. Urban Green.
A voxel-based lidar method for estimating crown base height for deciduous and pine trees
Remote Sens. Environ.
A comparative analysis of high spatial resolution IKONOS and WorldView-2 imagery for mapping urban tree species
Remote Sens. Environ.
Uncertainty in urban forest canopy assessment: lessons from Seattle, WA, USA
Urban For. Urban Green.
The use of airborne hyperspectral data for tree species classification in a species-rich Central European forest area
Int. J. Appl. Earth Obs. Geoinf.
A systematic quantitative review of urban tree benefits, costs, and assessment methods across cities in different climatic zones
Urban For. Urban Green.
Selecting trees for urban paved sites in Scandinavia – a review of information on stress tolerance and its relation to the requirements of tree planners
Urban For. Urban Green.
Tree species classification and estimation of stem volume and DBH based on single tree extraction by exploiting airborne full-waveform LiDAR data
Remote Sens. Environ.
Cited by (197)
Airborne data and machine learning for urban tree species mapping: Enhancing the legend design to improve the map applicability for city greenery management
2024, International Journal of Applied Earth Observation and GeoinformationMerging multiple sensing platforms and deep learning empowers individual tree mapping and species detection at the city scale
2023, ISPRS Journal of Photogrammetry and Remote SensingSpecies classification from hyperspectral leaf information using machine learning approaches
2023, Ecological InformaticsCarbon sinks in urban public green spaces under carbon neutrality: A bibliometric analysis and systematic literature review
2023, Urban Forestry and Urban GreeningA LiDAR biomass index-based approach for tree- and plot-level biomass mapping over forest farms using 3D point clouds
2023, Remote Sensing of Environment