Object-based classification of urban plant species from very high-resolution satellite imagery
Introduction
Global warming and air pollution are two major concerns affecting biodiversity, life quality, citizens well-being and human health (Anenberg et al., 2022, Malashock et al., 2022, Sicard et al., 2023, Southerland et al., 2022), in particular in cities where 56% of the world population lived in 2020 (United Nations, 2020). Following successive heatwaves and the COVID-19 pandemic, the European cities with more than 20,000 inhabitants have to establish “ambitious Urban Greening Plans” by including nature-based solutions and greening strategies by 2030 (European Union Biodiversity Strategy for 2030, COM (2020)380 final) to both mitigate the effects of air pollution and climate change and adapt to them. Urban trees are the most important elements of urban ecosystems (Ostberg et al., 2018), and contribute to reduce air pollution in cities (Baró et al., 2015, Salbitano et al., 2016, Nowak et al., 2018, Sicard et al., 2018, Pace et al., 2021), increase carbon stock (Proietti et al., 2016), mitigate the urban heat islands (Manes et al., 2012, Ren et al., 2022), provide cooling and shading (Rahman et al., 2020), regulate water runoff (Pataki et al., 2011), reduce noise (Klingberg et al., 2017) as well as provide social and psychological benefits, enhancing citizens’ well-being and human health (Samson et al., 2019, Ugolini et al., 2020).
Information about the individual tree, accurate data on location, species, and structural characteristic such as tree height and crown diameter are essential to quantify the environmental and social benefits of urban trees (e.g., Manes et al., 2016; Russo et al., 2016; Fusaro et al., 2017; Pace et al., 2018; Pu and Landry, 2019). Most of these data are derived from municipal tree inventories (e.g., Selmi et al., 2016), which include only public trees managed by the municipality representing e.g., about 36% of full stocking in 50 cities in the United States (McPherson et al., 2016). Tree inventories can also be obtained from field visits by visual observations (Nowak et al., 2013, Persson, 2016). This method is often expensive and time-consuming (Klingberg et al., 2017), and inefficient at larger scales when the tree species distribution is heterogeneous (Nowak et al., 2008, Westfall, 2015, Persson, 2016), making remote sensing more attractive (Klingberg et al., 2017).
Shojanoori and Shafri (2016) performed an extensive literature review on the use of remote sensing for forestry applications, particularly for tree species detection. During the last decades, the use of aerial photography and airborne laser scanning, such as LiDAR data, was widely used and considered as the most accurate remote sensing technique to estimate forest attributes (e.g., Singh et al., 2015; Alonzo et al., 2016; Persson, 2016; Song et al., 2016; Chen et al., 2017; Solano et al., 2019). The high cost coupled with the spatial, temporal, and spectral resolution of such techniques are the main limiting factors to using LiDAR to investigate individual trees (Key et al., 2001, Shojanoori and Shafri, 2016). In most of the previous studies, tree attributes were derived from satellite imagery at moderate spatial resolution such as Landsat (Manes et al., 2012, Silli et al., 2015, Marando et al., 2016) and sensors like Moderate Resolution Imaging Spectro-radiometer (Manes et al., 2016, Bottalico et al., 2017). Remote sensing, with high spatial and spectral resolution, and Geographical Information Systems (GIS) could be cost-effective and consistent techniques to get information about individual trees (Larondelle et al., 2014, Alonzo et al., 2016, Parmehr et al., 2016, Fusaro et al., 2017), when employed with useful tools in Mapping and Assessment of Ecosystem Services (Maes et al., 2016).
Since 2000, the spatial resolution of the optical very high resolution (VHR) satellite sensors allows detecting and estimating some tree attributes at landscape scale (Table 1S), e.g., Geoeye-1 (Qian et al., 2020), IKONOS (Pu and Landry, 2012), WorldView (Immitzer et al., 2012, Pope and Treitz, 2013, Nouri et al., 2014, Li et al., 2015, Pu et al., 2015, Choudhury et al., 2019, Kokubu et al., 2020), and Pléiades (Beguet et al., 2014, Lefebvre et al., 2016, Akbari and Kalbi, 2016, Effiom, 2018, Pu and Landry, 2019) at a lower cost compared to field observations or airborne laser scanning (Persson, 2016). Compared to IKONOS, the WV-2 sensor has a greater capability to classify tree species (Pu and Landry, 2012), and the Pléiades and WV-2 images perform equally well for estimation of forest attributes (Persson, 2016). Such steerable optical sensors allow fast and frequent stereo or tri-stereo acquisitions of cloud-free images over larger areas within a short interval of time (Persson, 2016).
Most of the previous studies, using very high spatial resolution satellite imagery, were limited to rural and forested areas (e.g., Immitzer et al., 2012; Pu and Cheng, 2015; Akbari and Kalbi, 2016; Modica et al., 2016; Solano et al., 2019). The detection of individual trees and species differentiation are challenging in cities, as trees can be isolated, lined up or grouped in patch, with a wide range of plant species, high spectral similarity of vegetation types, and high-density stands, trees in the shade, trees with low spectral contrast, and proximity of neighboring buildings (Alonzo et al., 2015, Parmehr et al., 2016, Klingberg et al., 2017, Choudhury et al., 2019). In addition, private gardens are an important component of urban landscape and urban green infrastructure, for instance contributing to 35-47% of the total urban green spaces in the United Kingdom (Loram et al., 2007). The potential contribution of residential areas and yards to overall urban sustainability is recognized worldwide (Owen, 2010, Cameron et al., 2012, Vila-Ruiz et al., 2014, Zhang and Jim, 2014), and could make significant contributions to urban biodiversity and ecological richness in cities (Smith et al., 2006, Müller et al., 2010, Cameron et al., 2012). However, their relative value within the wider urban green space is difficult to quantify and hence their measurable benefits is rarely assessed (Cameron et al., 2012, Zhang and Jim, 2014).
For a realistic and proper quantification of the benefits of urban vegetation in terms of providing ecosystem services such as mitigation of air pollution and urban heat island, and above-ground carbon storage at city scale, a consistent inventory of vegetation within private and public areas, is needed to avoid a large underestimation and to establish an efficient Urban Greening Plan and carbon footprint (Sicard et al., 2018, Pretzsch et al., 2021). As tree species diversity is a major factor in providing ecosystem services in cities (Grote et al., 2016, Galle et al., 2021), the large variety of tree species cannot be classified as a single vegetation category (e.g., conifers, broadleaves, and mixed species) as previously performed in studies (e.g., Manes et al., 2016; Bottalico et al., 2017).
For individual tree classification in urban areas, previous studies have reported that the classification algorithm produced better results by using object-based (overall accuracy: 77%) than pixel-based approach (overall accuracy: 73%) due to high spectral variability (Pu et al., 2011). By using pixel-based classification, the spectral response of individual trees can be influenced by canopy exposition (sunlit/shaded) and shaded canopy due to nearby buildings, reducing the overall accuracy (Quackenbush et al., 2000). The object-based image analysis includes spectral, textural, and spatial features, suitable for individual tree species classification in urban environments (Choudhury et al., 2020, Lefebvre et al., 2016, Puissant et al., 2014, Shojanoori and Shafri, 2016, Zhang and Qiu, 2012, Zhou, 2013). In literature, the object-based classification was performed using two machine learning approaches: Random Forest (e.g., Boukir et al., 2015, Puissant et al., 2014; Huesca et al., 2019, Immitzer et al., 2016, Lefebvre et al., 2016) and Support Vector Machine (e.g., Adam et al., 2014, Deur et al., 2020). The Random Forest classifier showed a performance of 89% to classify different textures with the co-occurrence matrix (Boukir et al., 2015).
Several studies used VHR images for some tree species identification in urbanized areas (Choudhury et al., 2020, Li et al., 2015, Pu, 2011, Pu et al., 2011), but none focused on identifying vegetation in both private and public lands. The structural diversity of the vegetation in residential yards can be a good predictor of biological diversity in the urban environment (Müller et al. 2010). However, identifying plant species in private areas is challenging due to i) spatial complexity in courtyards with building and shadows presence, and ii) high plant species diversity compared with trees in the public space such as street trees. Previous studies have reported 60-70% of non-native species in private residential gardens in temperate cities (e.g., Smith et al. 2006; Cilliers et al. 2012; Vila-Ruiz et al., 2014).
Trees at public roadside and green spaces have been studied, but those in private properties were largely ignored (Cameron et al., 2012, Zhang and Jim, 2014). Reliable information on domestic gardens is lacking, mainly due to difficulties in acquiring systematic data (Cameron et al., 2012, Zhang and Jim, 2014). Therefore, studying the vegetation characteristics within private areas has become a research priority in cities (Vila-Ruiz et al., 2014, Zhang and Jim, 2014). The aim of this study was to assess the potential of VHR satellite images for classifying plant species and mapping geo-located urban vegetation and greenspaces in both public and private land over two study areas in two cities, Aix-en-Provence (France) and Florence (Italy). We developed a object-based classification using the spectral and textural characteristics and image segmentation for detection, delineation, and classification of urban tree canopies and herbaceous areas. To our best knowledge, this is the first work using VHR satellite images for detecting, extracting, and classifying more than 20 individual dominant plant species, among hundreds of thousands tree canopies detected in both private and public areas, over large urban areas (50-80 km²) at high classification accuracy.
Section snippets
Aix-en-Provence, France
Southern France is characterized by strong urbanization pressures and increasing vulnerability to climate change (Sicard and Dalstein-Richier, 2015). The population of Aix-en-Provence, located near Marseille, is estimated at approximately 143,000 people over a total surface area of 186 km². The municipality Aix-en-Provence has a Mediterranean climate (Köppen-Geiger classification). The mean annual precipitation and temperature are 568 mm and 13.6 °C, respectively.
Florence, Italy
The municipality of Florence,
Results and discussion
In the present study, given the very-high spatial and spectral resolutions (e.g., PAN band at < 0.5 m resolution) and additional bands, we examined the suitability of WV-2 satellite imagery to identify urban vegetation in both private and public areas, and map the green cover by differentiating woody and herbaceous components over two large study areas in Aix-en-Provence and Florence. The images are of excellent quality, i.e., < 2% of the area is covered by clouds or haze. A pan-sharpening
Conclusions
Green spaces within private areas provide important contributions to the sustainability of urban systems. Despite that private areas occupy a large proportion of cities; domestic gardens and private areas are seldom included in the development and management of urban greening programs by local authorities to generate and sustain urban biodiversity while the trees in private areas amount to about 85% of the trees’ population in cities. For urban climate change resilience, it is essential to
CRediT authorship contribution statement
Conceptualization, P.S., F.C., and E.P.; methodology, P.S. and F.C.; data curation, P.S., M.L. and F.C.; writing - original draft, P.S. and F.C.; writing - review & editing, P.S., J.M., Y.H., V.A., A.D.M., and E.P.; supervision, P.S. All authors have read and agreed to the published version of the manuscript.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
This work was carried out with the contribution of the LIFE financial instrument of the European Union (LIFE19 ENV/FR/000086) in the framework of the AIRFRESH project “Air pollution removal by urban forests for a better human well-being”, and the Research Council of Lithuania (agreement No. S-LZ-21-3), and the Franco-Lithuanian Programme Hubert Curien (PHC) GILIBERT 2021 (46429UB). Project funded under the National Recovery and Resilience Plan (NRRP), Mission 4 Component 2 Investment 1.4 - Call
References (108)
- et al.
Mapping urban forest structure and function using hyperspectral imagery and lidar data
Urban For. Urban Green
(2016) - et al.
Mapping urban forest leaf area index with airborne lidar using penetration metrics and allometry
Remote Sens. Environ.
(2015) - et al.
Long-term trends in urban NO2 concentrations and associated paediatric asthma incidence: estimates from global datasets
Lancet Planet Health
(2022) - et al.
Mismatches between ecosystem services supply and demand in urban areas: a quantitative assessment in five European cities
Ecol. Indic.
(2015) - et al.
Random Forest in remote sensing: A review of applications and future directions
ISPRS J. Photogramm. Remote Sens.
(2016) Object based image analysis for remote sensing
ISPRS J. Photogramm. Remote Sens.
(2010)- et al.
A spatially-explicit method to assess the dry deposition of air pollution by urban forests in the city of Florence, Italy
Urban For. Urban Green.
(2017) - et al.
The domestic garden: Its contribution to urban green infrastructure
Urban For. Urban Green.
(2012) An improved dark-object subtraction technique for atmospheric scattering correction of multispectral data
Remote Sens. Environ.
(1988)- et al.
Mapping the diversity of street tree inventories across eight cities internationally using open data
Urban For. Urban Green.
(2021)
Variable selection using random forests
Pattern Recognit. Lett.
Synergistic use of QuickBird multispectral imagery and LIDAR data for object-based forest species classification
Remote Sens. Environ.
A comparison of multispectral and multitemporal information in high spatial resolution imagery for classification of individual tree species in a temperate hardwood forest
Remote Sens. Environ.
Influence of urban vegetation on air pollution and noise exposure - A case study in Gothenburg, Sweden
Sci. Total Environ.
Mapping the diversity of regulating ecosystem services in European cities
Glob. Environ. Chang.
An indicator framework for assessing ecosystem services in support of the EU Biodiversity Strategy to 2020
Ecosyst. Serv.
Regulating ecosystem services of forests in ten Italian metropolitan 31 cities: Air quality improvement by PM10 and O3 removal
Ecol. Indic.
“Structure, Funct. Value Str. trees Calif., Usa”. Urban For. Urban Green.
Air pollution removal by urban forests in Canada and its effect on air quality and human health
Urban For. Urban Green.
Estimation of urban tree canopy cover using random point sampling and remote sensing methods
Urban For Urban Green.
A multi-sites analysis on the ozone effects on Gross Primary Production of European forests
Sci. Total Environ.
Evaluating seasonal effect on forest leaf area index mapping using multi-seasonal high resolution satellite Pleiades imagery
Int. J. Appl. Earth Obs. Geoinf.
Mapping Forest leaf area index using reflectance and textural information derived from WorldView-2 imagery in a mixed natural forest area in Florida, US
Int. J. Appl. Earth Obs. Geoinf.
A comparative analysis of high spatial resolution IKONOS and WorldView-2 imagery for mapping urban tree species
Remote Sens. Environ.
Object-oriented mapping of urban trees using random forest classifiers
Intern. J. Appl. Earth Obs. Geoinform.
A new index to differentiate tree and grass based on high resolution image and object-based methods
Urban For. Urban Green.
Tree cooling effects and human thermal comfort under contrasting species and sites
Agric. For. Meteor.
Air pollution removal by trees in public green spaces in Strasbourg city, France
Urban For. Urban Green.
Trends in urban air pollution over the last two decades: a global perspective
Sci. Total Environ.
Should we see urban trees as effective solutions to reduce increasing ozone levels in cities?
Environ. Pollut.
Health and vitality assessment of two common pine species in the context of climate change in Southern Europe
Environ. Res.
Effects of LiDAR point density and landscape context on estimates of urban forest biomass
ISPRS J. Photogramm.
Urban domestic gardens: composition and richness of the vascular plant flora, and implications for native biodiversity
Biol. Conserv.
A methodology based on GEOBIA and WorldView-3 imagery to derive vegetation indices at tree crown detail in Olive Orchards
Int. J. Appl. Earth Obs. Geoinf.
Estimation of broad-leaved canopy growth in the urban forested area using multi-temporal airborne LiDAR datasets
Urban For. Urban Green
Urban vegetation classification: benefits of multitemporal RapidEye satellite data
Remote Sens. Environ.
Land-use/cover classification in a heterogeneous coastal landscape using RapidEye imagery: Evaluating the performance of random forest and support vector machines classifiers
Int. J. Remote Sens.
Determining Pleiades satellite data capability for tree diversity modeling
iForest
Seasonal variations in the spectral reflectance of deciduous tree canopies
Int. J. Remote Sens.
Analysis of effect of pan-sharpening in change detection on VHR Images
IEEE Trans. Geosci. Remote Sens. Lett.
Tree crown delineation algorithm based on a convolutional neural network
Remote Sens
Random forests
Mach. Learn
Urban tree species identification and carbon stock mapping for urban green planning and management
Forests
Social aspects of urban ecology in developing countries, with an emphasis on urban domestic gardens
Appl. Urban Ecol.: A Glob. Framew.
Tree species classification in mixed deciduous forests using very high spatial resolution satellite imagery and machine learning methods
Remote Sens
Cited by (4)
Measuring the 3-30-300 rule to help cities meet nature access thresholds
2024, Science of the Total EnvironmentAnalyzing the dynamic changes and causes of greenspace landscape patterns in Beijing plains
2024, Ecological Indicators