Local Climate Zones (LCZs) and Urban Morphological Parameters Using GIS: an Application to Italian Cities

: LCZs refer to a classification system that exists out of 17 classes, 10 of which can be described as urban, proposed as new standard for characterizing and comparing urban landscapes. This work evaluates the reliability of the LCZ level 0 map obtained from WUDAPT by comparing it with a more detailed LCZ map inferred from morphological information of several Italian cities. Morphological data from Digital Elevation Models (DEMs) are used to obtain a detailed morphological characterization of each city. Preliminary results for the city of Lecce shows that the WUDAPT L0 method misclassified some LCZs especially at the core urban cells, whereas wider matching is observed at the boundary between urban and rural areas.


Introduction
Local Climate Zones (LCZs) were introduced in 2012 as a new standard for characterising urban landscapes that considers the micro-scale of land cover and associated physical properties. LCZs refer to a classification system based on 17 classes, 10 of which can be described as urban (Figure 1). The LCZ classes are formally defined as regions of uniform surface cover, structure, material, and human activity that span hundreds of meters to several kilometres in horizontal scale [1].
The concept of LCZs contributed to a progressive step in the thermal analysis of urban areas. Thermal differences, for example in Urban Heat Island/surface Urban Heat Island (UHI/SUHI) intensities, are no longer confined to urban/rural temperature differences, but could also focus more closely on the differences between LCZs. Thus, the LCZ system provides an approach to the comparison of the thermal features of various neighbourhoods/areas within a city (intra-urban analysis), and/or comparison of similar types of neighbourhoods/areas between cities (inter-urban analysis). The concept of LCZ classification has found wide acceptance and application in a range of urban climate investigations, as recently reviewed by Lehnert et al. [2].   LCZ maps of the city are referred to as the Level 0 product as they represent the first level of information about urban areas. Levels 1 and 2 represent more detailed and higher resolution information [4]. The World Urban Database and Access Portal Tools (WUDAPT, www.wudapt.org) community provides a procedure for generating the LCZ level 0 product, which uses freely available Landsat imagery, Google Earth for creating the training areas (TAs) [5].
There are other approaches for generating LCZ level 0 products, which can also lead to the more detailed Level 1 and 2 data. Where the data are available, administrative data (on building footprints, heights, green spaces, etc.) and Geographic Information System (GIS) software can be used. Raster-based methods superimpose a standard grid over the urban landscape and information about selected variables (e.g., building height, sky view factor) can be acquired at the scale of the individual cell. Each variable represents a layer and the gridded layers are then combined using rules to generate LCZ types. This method has been used to generate LCZ maps for Hong Kong [6,7], Nagpur in India [8], three medium-sized Central European cities, Brno, Hradec Kralove, and Olomouc in the Czech Republic [9] and for Bilbao (ES) [10]. Vector-based methods capture the boundary of the LCZ neighbourhood and represent a more precise delineation of contiguous neighbourhood types as individual objects, as employed by Unger et al. [11] for Szeged (HU), Perera et al. [12] for Colombo (Sri Lanka), Muhammad et al. [13] for Berlin (DE). This work compares the LCZ level 0 map obtained from WUDAPT (LCZ WUDAPT hereinafter) with a more detailed LCZ map retrieved from morphological information (LCZ GIS) of several Italian cities and show main differences in terms of fraction of area that has been classified correctly (same in WUDAPT and GIS), over the total area. For this purpose, morphological data from Digital Elevation Models (DEMs) are used to obtain different morphological parameters used for the detailed classification of LCZ classes.

Methodology
Both the LCZ WUDAPT and the LCZ GIS maps are currently being produced for several Italian cities (Lecce, Bari, Naples, Rome and Milan).

LCZ Map Generated in WUDAPT
Following the standard procedure formalized by Bechtel et al. [14] and adopted by the WUDAPT community project, an "off-line" workflow that integrates training areas (TAs, a set of LCZ labelled polygons) and Landsat 8 (L8) imagery within the SAGA software package [15] over a limited spatial domain of different cities is followed. More specifically, each TA is identified using Google Earth images aided by the visual and numerical information provided in Stewart and Oke [1]. The TAs dataset is used to extract spectral information from L8 images, which in turn is used in a supervised random forest classifier to categorize the entire region of interest into LCZ types. The LCZ Generator (https://lcz-generator.rub.de) is employed to map the cities into LCZs, solely expecting a valid training area file and some metadata as input [16].

LCZ Map Based on Morhological Parameters Using GIS
For the detailed study of the parameters that constitute the LCZ classes, DEMs of the cities are used in combination with shape files representing the buildings. Using QGIS, the following parameters are calculated: mean building height, fractional area of buildings (λp), fraction of urban land use (λu), pervious surface fraction, aspect ratio, sky view factor. A detailed description of the parameters can be found in Grimmond and Oke [17] and Burian et al. [18]. These parameters are calculated as mean values for each 100 m × 100 m cell over the grid covering the city. The detailed LCZ map is based on the parameters dataset to fit the LCZ look-up table as defined by Stewart and Oke [1].

Results and Discussion
Here as an example of application, preliminary results are shown for the city of Lecce.   Figure 3 shows the LCZ WUDAPT and LCZ GIS maps, as well as the percentage areas and the percentage occurrence for the different LCZ. For each LCZ class, the occurrence represents the amount of a class area of the LCZ WUDAPT that overlaps to that of the LCZ GIS map.

LCZ maps
The LCZ WUDAPT map shows that LCZ 2 is the most abundant (34%), characterizing the city centre, with a dense mix of midrise buildings 3-9 stories, few or no trees, land cover mostly paved and stone, brick, tile, and concrete construction materials. In the north-eastern part of the city there is the presence of the LCZ 4, while around the city there is the presence of LCZ 14 (also called LCZ D) and LCZ 9, both characterizing the rural and suburban parts. Also, in the north-western part of the city there is the presence of LCZ 8, corresponding to the industrial/commercial zone of Lecce.
When compared with the LCZ GIS map, it can be noted that a smaller area of the city centre belongs to the LCZ 2 class (6%), while moving towards the periphery open arrangements occur. It is noteworthy that GIS-LCZ method provides a more detailed representation of the city centre, reclassifying LCZ 2 into four LCZ classes. However, limitations in the description of LCZ 8 can be observed, as it appears scattered and widespread around the city. Further limitations in the description of LCZ 8 are represented by the presence of typically residential LCZ classes in the commercial/industrial area. The low percentage values of occurrence represent a broad reclassification of most LCZ GIS classes, except for LCZ9, which shows a high overlap (about 60%).

Conclusions
This work employs different Italian cities and compare maps generated in WUDAPT with those based on the direct calculation of morphological parameters using GIS over a grid of 100 m × 100 m cells. For the city of Lecce presented here, it is found that, while the general spatial distribution patterns of different LCZ classes in the two LCZ maps is qualitatively similar, the GIS-LCZ method can improve the accuracy as the WUDAPT L0 method misclassified some LCZs especially at the core urban cells, whereas wider matching is observed at the boundary between urban and rural areas. This is probably due to an inadequate description of urban canopy parameters and urban land use at the border between urban and rural areas, as reported in Pappaccogli et al. [19]. On the other hand, the WUDAPT method generates LCZ maps with a more homogeneous pattern than the GIS-based method. High-resolution remote sensing images or conducting remote sensing image fusion is expected to improve the quality of output of WUDAPT-LCZ classification and mapping.
Nevertheless, the GIS-LCZ method requires detailed environmental databases (i.e., digital elevation and surface models), which are not typically available for a larger number of cities in Europe, representing the main challenge for an operational use of this approach. Further work will be made to improve the GIS-LCZ method in classifying the zone properties using algorithms employed in the literature (e.g., the fuzzy logic algorithm adopted by Muhammad et al. [13] for Berlin). Findings from this work can provide a useful reference for researchers who are interested in LCZ classification and mapping work for their cities.