Potential of Internet street-view images for measuring tree sizes in roadside forests

Highlights

• Tree size census using street-view images is recent and vital to evaluate forests.

• Tree size was measured efficiently using fix-sized objects in street-view images.

• Precision levels obtained were high—92% for DBH and 80–87% for canopy height.

• Meters were tree limewhite, lane width, traffic line width, and road curb height.

• Tree limewhite was a more effective gauge for measurements than the rest.

Abstract

Tree size censusing is essential for evaluations of trees and forests, but traditional field surveys are both time- and labor-intensive. Here, we discuss the use of panoramic 360-degree street views available on the Internet for censusing of roadside trees in urban regions. Use of scale-independent, fixed-sized street objects as recalibrating meters in tandem with imagery software enabled street-view images to be used effectively in the remote measurement of diameter at breast height (DBH), tree height, underbranch height, and canopy projection size. Comparison of four independent meters determined that stem limewhite-related meters (used for tree disease and bark-freeze injury control; usually 1.3 in height throughout China) had greater precision than road curb height, lane width, and traffic line width meters. The limewhite meter’s precision was slightly lower than those of the meters in combination (i.e., when at least three of the abovementioned meters were used for the same tree measurement), but no statistically significant differences were detected between the limewhite and combined meters (p > 0.05). In contrast, the road curb height, traffic line width, and lane width meters all had significantly lower precision. The highest levels of precision were 92%, 87%, and 80% for DBH, height (tree height and underbranch height), and tree canopy size measurements, respectively. Empirical recalibration of the image-based measurements did not improve data precision with reference to field surveys (p > 0.05). Moreover, similar results were obtained regardless of individual users, and repeatability for DBH measurements (r² > 0.92), and maximum differences among individual users were 0.6–1.9 cm for DBH (averaged at 22 cm) and 8–50 cm for underbranch height (mean value at 8 m). Labor costs and time needed for this approach were one-thirtieth to one-tenth those required for field surveys. Thus, the use of street-view images represents a more resourceful approach to assess forest ecological services.

Introduction

Street trees play an integral role in supporting healthy urban communities, and provide a range of significant social benefits, including enhancing overall human health (James et al., 2016), increasing physical activities (Dalton et al., 2016), improving child birth rates and weights (Cusack et al., 2017; Ebisu et al., 2016), decreasing psychological stresses (Dimitrova and Dzhambov, 2016), reducing environmental problems (e.g. heat island effect) (Richards and Edwards, 2017), and increasing property values (Chang and Chou, 2010). Urban green spaces, such as urban parks, woodlands, street trees, trees in town squares, among others (He et al., 2017; Konijnendijk et al., 2005), have long been recognized for their role in improving urban environments (He et al., 2017; Li et al., 2016, 2015). Accurate measurements of trees is necessary for many applications, including resource appraisals, biometric modeling (Dittmann et al., 2017; Ren et al., 2018), and ecological service evaluations (Wang et al., 2018; Zhai et al., 2017; Zhang et al., 2017; Zhao et al., 2016). A field survey of street trees provides invaluable information for urban forest managers, but such surveys are costly, labor intensive, time consuming, and can pose safety risks to field crews (Alonzo et al., 2016).

An alternative tree-censusing method that uses existing internet-based images to estimate tree attributes and meets precision requirements (Nielsen et al., 2014) would greatly facilitate roadside tree evaluations and could potentially be used to estimate ecological services. Internet-based data have previously been used to evaluate ecological services provided by urban vegetation, such as the use of Google Street View to estimate the effects of tree shading (Li and Ratti, 2018; Li et al., 2018; Richards and Edwards, 2017), correlations between racial/ethnic minority communities and access to urban parks and tree coverage (Zhou and Kim, 2013), and the relationship between urban vegetation distribution and income levels (Li et al., 2016; Long and Liu, 2017). Street-view images have also been used in the auditing of built environments (Badland et al., 2010; Kelly et al., 2013; Rundle et al., 2011) and in systematic social observations of neighborhoods with high densities of children (Odgers et al., 2012). Street-view imagery has also been used in non-urban ecosystems, such as to identify vulture nesting sites on cliff faces (Olea and Mateotomás, 2013) and to provide useful occurrence data for mapping the distribution of the pine processionary moth (Rousselet et al., 2013). Currently, internet-based street-view images are available free of charge on sites operated by Google, Baidu, and Tencent, among others (Anguelov et al., 2010; Long and Liu, 2017). The diversity of research applications of street-view images clearly demonstrates their utility as sources of virtual information that can reduce the need for field work.

The potential for using street-view datasets for measuring tree sizes in urban environments should take different aspects into consideration. First, a meter of fixed size and the tree to be measured should ideally be in the same picture, thereby enabling size measurements via specialized software (Hou et al., 2014; Juujjrvi et al., 1998; Nielsen et al., 2014). The meter should have scale-independent/sight-depth-independent features (Abd-Elrahman et al., 2010; Clark et al., 2000), such as indices that must meet international or national standards for street construction (e.g., traffic line width on roads, lane widths, etc.) or urban forest management (e.g., limewhite on tree stems, approximate DBH height, urban and roadside poplar tree cultivation in China) (Ministry-of-Housing&Urban-rural-Development-of-China, 2012). Second, the image should be at an appropriate viewing angle, so as to minimize visual deformation of the trees to be measured (Brownlie et al., 2007; Hou et al., 2014); for instance, 360-degree panoramic street-view images enable viewing from numerous angles. Finally, software with measurement capabilities, such as ImageJ, have been shown to facilitate image analysis (https://imagej.nih.gov/ij/). However, a recent study that utilized street-view pictures to remotely estimate tree diameter (without software-based measurement) found that the level of personal experience greatly influenced tree size estimates (i.e., virtual survey tree diameter estimates were, on average, 38% of the field measurements initially, whereas later measurements approached 91% accuracy) (Berland and Lange, 2017), which made such an approach an unreliable and unfeasible alternative to in situ field tree censusing. To the best of our knowledge, the potential for using these abovementioned meters (i.e., comparing trees to objects of known sizes in street-view images) in assessments of tree sizes (diameter, tree height, canopy size) has been unexplored. As such, a systematic study examining the reliability of using street-view pictures to census tree sizes, and to identify appropriate measurement meters and evaluate software for picture measurements, is greatly needed.

In this paper, we addressed the following questions: 1) how closely do image-based measurements coincide with field measurements? How do the different scale-independent meters of traffic line width, road curb height, tree limewhite height, and lane width differ in their effectiveness for measuring four tree size parameters (DBH, underbranch height, tree height, and canopy size)? (2) do different people produce similar results when using these meters in image-based measurements, and is the precision the same regardless of who makes the estimates? and (3) is using an image-based tree size measurement approach more efficient in terms of time and costs than field censusing?