Urban surface cover determined with airborne lidar at 2 m resolution – Implications for surface energy balance modelling
Introduction
Urban surface cover is strongly interrelated with several surface properties and processes in the surface–atmosphere interaction of an urban area. The replacement of pervious surfaces and vegetation by impervious materials alters the surface energy balance: most notably, the available energy is used for the vertical turbulent sensible heat flux rather than evaporation (e.g. Grimmond and Oke, 1991, Nordbo et al., 2012a, Loridan and Grimmond, 2012) which leads to the intensification of the inadvertent urban heat island effect, observed in many cities (Arnfield, 2003). The reduced evaporation and percolation are accompanied by a larger runoff, and together they affect the water balance of a city and cause a higher flood risk (Xiao et al., 2007); though this could give rise to opportunities for improving the urban environment (Coutts et al., 2013). The lack of vegetation also minimizes carbon dioxide (CO2) uptake via photosynthesis and reduces the buffering effect of air pollutants (Demuzere et al., 2014). Indeed, the fraction of natural area has been shown to be a very strong proxy for predicting CO2 emissions from cities (84% explanation power, Nordbo et al., 2012b).
The exact determination of the fraction of vegetation has received a lot of attention due to links to urban-heat-island mitigation. The simulated urban heat island in Tokyo was shown to be up to 1.5 °C weaker when patchy vegetation was taken into account in a simulation (Hirano et al., 2004). By patchy vegetation, we mean small and isolated vegetation surrounded by impervious surfaces (one example is street trees). Vegetation has been concluded to have a cooling effect in an urban environment due to shadowing effects and evaporative cooling, especially in summer (Simpson, 2002, Bowler et al., 2010, Lindberg and Grimmond, 2011, Coutts et al., 2013). Thus, there is a clear need for land-cover classification that adequately resolves patchy vegetation.
Urban surface cover is commonly available only with a coarse resolution (10–50 m) and without building and tree height information. At such a resolution, patchy vegetation is almost entirely disregarded. Airborne Light Detection And Ranging (LiDAR, hereafter lidar) measurements are based on active remote-sensing technology where visible or near-infrared (NIR) light is sent and received in high-frequency pulses (Beraldin et al., 2010). Lidar technology provides a solution to the resolution problem: it allows detailed scanning of a 3D complex urban structure with a high horizontal and vertical accuracy with a relatively low cost. Furthermore, the intensity of returned lidar pulses is an indicator for the colour of the target hit, which improves the ability for land-cover determination. Generally, the determination of urban morphology is comparatively simpler since it relies only on the height information of the returned lidar pulse; whereas intensity information is additionally needed for surface-cover classification.
The use of the lidar scanning technique is increasing its popularity in the determination of urban land cover (Table 1). Classification has been mainly done using object-based approaches (Blaschke, 2010), where pixels are first divided into segments which are then classified into land-cover classes using different methods (Brennan and Webster, 2006, Im et al., 2008, Matikainen and Karila, 2011, Buján et al., 2012, Zhou, 2013, O’Neil-Dunne et al., 2013 in Table 1). Classification trees, a machine learning algorithm presented in Breiman et al. (1984), or other related approaches, have been used in many studies, and overall accuracies over 90% have been reached (e.g., Im et al., 2008, Mancini et al., 2009, Matikainen and Karila, 2011 in Table 1). The number of surface-cover classes ranges from 2 to 10 among the research done within a decade, and the most commonly used variables used for classification are: height above ground and intensity or/and a colour channels from an orthoimage (geometrically corrected aerial photograph, Table 1). Segmentation and classification have typically been carried out using rasterized data. This is the most straightforward approach when an orthoimage is used together with lidar data. Classification of individual lidar returns has mainly been used in a pre-processing stage to group lidar points into ground points and non-ground points. Matikainen and Karila (2011) made a comparative surface-cover classification study in a suburban area using five techniques and concluded that a combination of lidar scanning and orthoimage data gives the best result. Lidar scanning can also be used for updating maps through an automatic detection of changes in buildings (Matikainen et al., 2010), for calculating solar irradiance in urban canopy (Tooke et al., 2012) as well as for predicting building energy demand (Tooke et al., 2014).
This study has two main aims: (i) to determine the land cover of a large area within Helsinki at a very fine resolution (2 m) for use in modelling of the surface–atmosphere exchange, in addition to creating an elevation map as a part of the land-cover classification, and (ii) to quantify the sensitivity of urban energy balance modelling on land-cover resolution at an urbanized micrometeorological measurement site, thus addressing errors caused by too-coarse maps commonly used to describe the surface in urban energy balance models.
Surface cover is determined by airborne lidar scanning data, together with a classification-tree algorithm. The classification is conducted initially for each individual lidar return, rather than for pixels or segments as has been done in several previous studies. This is done due to the large and very heterogeneous surface and the need for a generic method that could be applied at other micrometeorological stations. The study area of concern is the second largest study worldwide among urban areas for which classification has been done using lidar data (Table 1). The resulting surface-cover data are then used for simulating surface energy fluxes using the Surface Urban Energy and Water Balance Scheme (SUEWS version 2014b, Järvi et al., 2011, Järvi et al., 2014). The simulation is done for an eddy-covariance flux-measurement site, located in the lidar scanning domain (Nordbo et al., 2013). Recommendations for an adequate surface-cover database resolution are finally given.
Section snippets
Study area
The data included in this study cover an area of 6 km by 9 km (54 km2), consisting of six tiles from the national lidar database and covering central Helsinki (Fig. 1). Helsinki is the capital of Finland located on the coast of the Baltic Sea with a population within Greater Helsinki of about one million. The city is characterized by a very heterogeneous surface with a dense city centre (although without skyscrapers) and a large central park which is mostly a mixed forest. The vicinity of the sea
Eddy-covariance site and measurements
The eddy-covariance measurements, used to evaluate the energy balance model performance, were done at the Hotel Torni site located in the centre of Helsinki (Fig. 3). The site is one of the measurement sites of the SMEAR III (Station for Measuring Ecosystem-Atmosphere Interactions) and the site is also part of the intensive Urban Boundary-layer Atmosphere Network in Helsinki (Wood et al., 2013). The site is characterized by dense urban land-use with a high fraction of built-up area with
Evaluation of surface-cover classification
The large amount of data (over 40 million data points) including four flights were separately classified using the classification-tree machine-learning technique. The minimal leaf area ranged from 14 to 66, and the number of nodes (i.e. splits in classification tree) from 15 to 65 (Table 4). Height above ground was always the most important classifier, followed by RPP and intensity. The 3rd flight is an exception since the order of importance of RPP and I was reversed and the
Summary and conclusions
The study showed one of the largest published studies of an urban-to-suburban area (54 km2) classified based on airborne lidar data. Individual lidar returns were classified instead of pixels or segments using solely classification trees (overall accuracy 91%). The accuracy was worst when distinguishing impervious and grass, because the separation relies almost solely on return intensity. The number of returns-per-pulse distinguished buildings from trees but buildings with special roofs, such as
Acknowledgements
For funding we are grateful to the Academy of Finland (Projects 1118615, 138328 and 263149), the EU (Projects 211574 and 244122) and ERC (Project 227915). The work has also been part of the ICOS-EU, ICOS-SA (263149), InGOS, GHG Europe, and DEFROST -projects. Leena Matikainen was working in the project ‘Centre of Excellence in Laser Scanning Research’ (Academy of Finland, project 272195). For scientific and technical advice we thank Juha Hyyppä and Eero Ahokas from the Finnish Geospatial
References (45)
Object based image analysis for remote sensing
ISPRS J. Photogramm. Remote Sens.
(2010)- et al.
Urban greening to cool towns and cities: a systematic review of the empirical evidence
Landscape Urban Plan.
(2010) - et al.
Mitigating and adapting to climate change: multi-functional and multi-scale assessment of green urban infrastructure
J. Environ. Manage.
(2014) - et al.
The surface urban energy and water balance scheme (SUEWS): evaluation in Los Angeles and Vancouver
J. Hydrol.
(2011) - et al.
Tree species differentiation using intensity data derived from leaf-on and leaf-off airborne laser scanner data
Remote Sens. Environ.
(2009) Improved estimates of tree-shade effects on residential energy use
Energy Build.
(2002)- et al.
Integrated irradiance modelling in the urban environment based on remotely sensed data
Sol. Energy
(2012) - et al.
Mapping demand for residential building thermal energy services using airborne LiDAR
Appl. Energy
(2014) - Ahokas, E., Kaasalainen, S., Hyyppä, J. and Suomalainen, J., 2006. Calibration of the Optech ALTM 3100 laser scanner...
Two decades of urban climate research: a review of turbulence, exchanges of energy and water, and the urban heat island
Int. J. Climatol.
(2003)