Developing climate services aimed providing actionable information for end-users, requires the generation, analysis, and integration of vast amounts of new data. To tackle the challenges related to dealing with fast growing multi-variable data, the Digital Earth project is currently being undertaken by eight research institutes of the Helmholtz Association in Germany. The project aims to advance the value chain of observation and scientific analyses through prognosis to societal actions.

Common data science techniques, including machine learning algorithms and novel visualisation techniques, are employed to extract information from large volumes of high temporal and spatial resolution data in both modelling and observations. At this symposium we will present first experiences in developing climate data products from using such data science techniques, to make better use of the vast amounts of data coming from state-of-the-art climate models. Regional climate models are increasingly used to develop regionally specific characterisations of the future climate based on ensemble simulations, and these produce vast amounts of data. Novel techniques are required to extract relevant information from these data, that can then be used by end-users, for instance with the aim of informing local decision makers on risks and opportunities from projected regional climate change.

Here we use different datasets generated by the regional climate model REMO, operated by the Climate Service Center Germany (GERICS) in Hamburg. We use daily and hourly precipitation input. These examples include the identification and tracking of heavy precipitation events in REMO daily and hourly climate data. For this purpose, we analyse extreme rainfall amounts at different durations and frequencies across Germany and the European domain. In a second example, we use a large ensemble (up to 1,000 years), for current and future climate under different levels of global warming from the HAPPI-DE project. The large ensemble allows us to reduce uncertainties associated with the estimate changes of rare rainfall events for the European domain. These results are useful for engineers and urban planners, for estimating plausible future changes of urban flood risks and for designing resilient drainage systems.

GERICS provides politicians, public administrations, and businesses with prototype products, climate advisory services and tailor-made decision-relevant information, based on scientifically sound practices, to support their efforts to adapt to climate change. Based on state-of-the-art ensemble climate change simulations, we provide customized products, such as ‘Climate Fact Sheets’, which provide concise information on the magnitude of projected climate change and the range of these changes for individual regions, as well as site-specific locations.

Area A - Board 123

If there is a construction site, there are also waste produced and new materials transported to the site. These mostly inert, therefore heavy volumes require the relocation via road infrastructure. Local and national road network is exposed to this amortization effect, which is resulted from construction activities. This effect can be significant if a region is observed, where suburbanization is of high power: construction and demolition has high likeliness, while road network is less built out – or less maintained, but still important in daily traffic.

In our case study from the Budapest agglomeration we present an approach, where the construction activities are detected and estimated in their volume with the utilization of Copernicus Sentinel 1 SAR data. This detection is built up in a quarterly reporting scheme (2014-2018). Then using the detected volumes, network analysis of road network is carried out with multiple approaches. The diffuse model, assumes, that the in- and output volumes are arriving/leaving to any possible directions through the border of the region where the network is present. The focus model assumes main access points (from road hierarchy) and a main destination (closest waste disposal site) of the change volumes. The mixed model will assume that input is diffuse, while output is focused. During the model calculation road hierarchy has a weighing force on the degree of amortization: by same volume of transportation 3rd level roads face bigger amortization effect then 1st level roads, since they are built out (and maintained) for lower capacities. Model outputs are in situ verified with assessment of road quality sampling. The model outputs and their differences are presented to the local governments for further utilization.

On long term, the presented case study intends to set up a regional planning tool, which helps the road network maintaining authorities to set the priority of their activities. The study has been supported by the EUREKA labelled REMEDI project (Hungarian National Research, Development and Innovation Office - EUREKA_15-1-2016-0023) in order to detect and monitor waste volumes at the points of their production – before they are becoming the part of waste management. According to the mission of the project, this transaction between the construction and waste management subsystems of our built environment is one of the most important questions, which shall be answered by the ideas of resilient cities. The study presents, that Sentienl-1 data may have truly relevant information in the decision support of sustainable development of our urbanisation.

Area A - Board 124

In 90% of all infrastructure projects globally that suffer from cost overruns, the concept-and-design phase is where the most project value can be gained (or lost). Geodetic data acquisition in urban areas during big infrastructure projects are major cost drivers. Modern remote sensing methods help us to manage large data acquisitions with high effectiveness.

Here, we show how we apply space-borne radar interferometry to monitor settlements and subsidence in the context of big road and railroads projects. We showcase three examples: a) the new E18-highway development project in Bærum, Norway, where radar interferometry is applied prior to (and eventually during) construction; b) the Follo Line railway project between Ski and Oslo, where radar interferometry is used to monitor ground stability during the underground advancement of two TBMs; and c) the 'Oslo central railway station monitoring project' where radar interferometry was used after the occurrence of damages at the Oslo central railway station (Oslo S in short), in order to track down when and where damaging ground subsidence had started and to which construction project in the vicinity of Oslo S the occurrence of these ground movements could be attributed.

Monitoring is done by exploiting very high resolution TerraSAR-X data in the E18 case and the Oslo S case, and high resolution Radarsat-2 fine mode data in the Follo Line case. While the interferometry results are communicated to the client as maps and reports in the Oslo S case, results are directly accessible to the clients through a web-based map portal in the other two cases.

In addition to the results of the TerraSAR-X and Radarsat-2 analysies, we show in which context Sentinel-1 data is sufficient for urban subsidence monitoring, and in which cases the use of higher resolution data is crucial to better understand the complex interaction of engineered structures and groundworks.

Area A - Board 125

Madrid and surrounding cities make one of the largest urban areas in Europe. Monitoring of large urban areas is a priority challenge in this century. Remote sensing and, more precisely DInSAR, is a promising technique for the continuous monitoring of smart cities. Although the potential and capabilities of the PSI technique have been shown in numerous applications in the last decade, the technique is far from being standardized for monitoring underground works. These uncertainties have limited the application of the SAR-based monitoring techniques to operational subsidence control systems. This work presents the potential of SAR data for the monitoring of underground works in the Madrid metropolitan area. The continuously growing urban area of Madrid over the last two decades led to traffic and mobility problems which had the maximum indicator in the M-30 beltway, which at the beginning of 2003 was on the brink of collapsing due to the heavy traffic. In order to ease this problem, the Madrid Council projected the remodeling plan of the M-30 implying the underground of up to 50 kilometers of roads and streets. One of the most important and challenging works of the framework was the southern By-Pass tunnel. This is the test site used in this work as validation scenario. A set of 26 Envisat images acquired during M30 Tunneling Project (2003-2008) were processed based on PSInSAR methodology. Conventional topographic measures (leveling, extensometers and inclinometers) were used for validation. Results showed a RMS of 2.6 mm with a 65% of coverage of the area of interest with persistent scatterers (PS). The temporal resolution of available data for 2003-2008 period (ERS & Envisat) showed limitations for a more continuous monitoring. Nevertheless, DInSAR technique applied to actual high temporal and spatial resolution data (Sentinel 1A&B) has a great potential to monitor large urban areas at lower costs.

Area A - Board 126

Coastal urbanization is rapidly growing worldwide and the management of the increasing environmental and security risks stressed by the climate changes will be a significant challenge in the years to come. Under the umbrella of the IGCP663 Project “Impact, Mechanism, Monitoring of Land Subsidence in Coastal cities” funded by UNESCO-IUGS, efforts are made to enhance the scientific understanding of land subsidence in coastal cities. In the framework of this project, this work aims to use Advanced Differential Interferometric Synthetic-Aperture Radar (A-DInSAR) displacement time-series analysis for assessing urban land subsidence and investigating the geotechnical properties of coastal soils. The approach has been applied along the coastland of the Venice Lagoon, in Italy, using A-DInSAR data acquired by ERS, ENVISAT, TerraSAR-X and COSMO-SkyMed. The results give insight about the capability of this methodology to understand the local effects of consolidation settlements due to the construction of new structures and infrastructures and to monitor the ancient historical center movements overlapped on the natural processes. The integration of A-DInSAR displacement time-series and geotechnical information allows to characterize the in situ heterogeneous proprieties of coastal soils over wide areas. The outcomes contribute to an improved understanding of the ground motion in response to construction and provide important information for urban planners about the expected time-dependent urbanization-induced deformations.

Area A - Board 127

1 Introduction

High and very-high resolution (HR and VHR) elevation models based on (multi-) stereo satellite data are an invaluable data source for disaster risk evaluation and urban planning activities especially for megacities in developing countries.

With the fast growing number of VHR satellites that are very agile and therefore capable of acquiring (multi-) stereo images (e.g. WorldView-1/-2/-3/4 and Pleiades 1A/B) and additionally HR satellites that are constantly acquiring stereo images (Cartosat-1, ZY-3 or ALOS Prism), there is a very reliable data source available for the generation of high quality digital elevation models over rapidly changing urban areas and their surroundings.


2 Workflow and product description

For the generation of VHR elevation models from 30 cm to 2 m resolution, two data sources are usable:

• Tri-Stereo archive data or new collections if no suitable data are available
• multiple mono archive scenes with acquisition geometries suitable for digital surface model (DSM) generation

By using a tri-stereo constellation, occluded / shadowed areas in cities are minimized and the resulting DSM is highly reliable and accurate.

For larger areas (metropolitan areas) HR true-stereo data can be used to generate elevation models with 5 m – 10 m resolution. By using HR data for the surrounding areas certain phenomena like flood risks can be examined on a larger scale and this is very important because cities cannot be seen without their urban fringe.

In the last 10 years, GAF has developed together and in close co-operation with the Remote Sensing Technology Institute (IMF) of the German Aerospace Center (DLR), a highly automated multi-source production line for generating digital surface models out of optical stereo and multi-stereo satellite data that is based upon a pixel-wise matching approach called Semi-Global Matching (SGM).

An automatic outlier detection algorithm is applied, which utilizes all intermediate height models (in case of a triplet: six intermediate height models). After outlier detection and elimination, the process merges those intermediate models to a single height model. During this process, pixel-sharp traceable information layers are automatically generated. Those quality- and traceability layers serve as input for automated outlier detection and manual editing. Moreover, all quality and traceability layers are delivered to the customer in the same geometry as the DSM and allow for assessing the DEM quality down to each pixel (Carl et al., 2013; Bärisch et al., 2017). Depending on the degree of editing, DSMs can be made available starting from 48 h after product order and can therefore be also interesting for emergency management in case of disasters.

The GAF Elevation Suite is composed of different elevation models with spatial resolution between 30 cm and 10 m resolution and can be provided for all regions worldwide, including even remote areas. More than 150 cities worldwide are already generated in 30 cm and 50 cm resolution and more than 7.7 million km² are available in 5 m resolution.

Based on the resulting DSMs, further products like digital terrain models (DTMs), normalized DSMs (nDSMs), building block or building models can be generated and used for different kinds of applications as described below.

The most difficult task is to derive DTMs and building footprints in densely built-up and hilly areas with small houses or shacks. To overcome the problem of footprint detection for small, low and densely built buildings, we propose to summarize these areas to building blocks (including calculation of average heights for these blocks). To satisfy the requirements of complicated terrain situations and to enhance the results of the DTM generation, the combination of different filter methods is proposed within urban areas, depending on building density and relief.

In order to remove trees and higher shrubs from the DSM, we use the Normalized Difference Vegetation Index (NDVI) calculated from the ortho image with the same geometric resolution. By thresholding this NDVI result, we identify the areas, where the DSM has to be filtered.


3 Support for resilient cities planning

The elevation suite provides high-quality, up-to-date and standardized spatial information in 2D and 3D. Especially in times with a massive and unplanned growth of megacities, particularly in developing countries, these data can deliver reliable and reproducible results for many urban analysis. It also lays the basis for building standardized workflows for all kinds of institutions and local authorities in developing countries. Accurate geospatial data with a z-component can be also used to provide supporting information for disaster risk reduction (e.g. vulnerability mapping, flood risk modelling) and in the event of emergency cases. Furthermore, it can also be exploited for other advanced analysis like population estimation and population density calculation. The support of participatory, integrated and sustainable human settlement planning and management including environmental topics as well as safeguarding cultural and natural heritage is also a crucial point in developing countries. These products can also make a contribution towards accomplishing the Sustainable Development Goal 11 (Make cities and human settlements more inclusive, safe, resilient and sustainable).

4 Examples

The first example shows the usage of a VHR DSM for assessing the building structure within the heterogeneous and hilly area of Rwanda’s capital city, Kigali.

A 50 cm resolution DSM and ortho image, a DTM, a nDSM and subsequently 3D building models, based on Pleiades tri-stereo satellite data, were successfully generated and used for the city of Kigali, Rwanda as part of the ESA EO4SD project. These models were then intersected with cadastral maps and were then used for various purposes, such as urban planning and design, the assessment of property taxes and the calculation of floor-area-ratios, an element of legal zoning frameworks for the building/construction sector (Haeusler et al., 2017).

Another example for the usage of a digital elevation model was a dam break analysis of the Mosul dam in north Iraq based on a 5 m Euro-Maps 3D DSM. For the City of Mosul the potential flood extent and depth of the Tigris River was calculated. This example shows the significance of regional DEMs for analysing large-scale disasters with an impact on urban areas.

Within the Urban Atlas 2012 project update / extension (a common initiative between the European Commission (DG-ENTR und DG-REGIO) in cooperation with the European Environmental Agency (EEA) and ESA, a digital height model (DHM) with 10 m resolution (based on Cartosat-1) was generated for the EU-28 and EFTA capital cities, to assist in population estimation and density calculation. The resulting DHM is a building block model with an accuracy better than 3 m.

5 Conclusions

The project examples show the significance of high-quality elevation data for a sustainable urban planning and disaster risk reduction in developing countries.

With the launch of new satellites in the next three years (e.g. DigitalGlobe Legion, Pleiades Neo, and Resourcesat-3S und 3SA from ISRO) it is guaranteed that the availability of high-quality stereo (capable) satellites is assured. Furthermore this will offer new opportunities (e.g. higher resolution, faster availability of data, better coverage) and will as well enhance the overall quality of derived elevations models.


6 References

Haeusler, T.; Gomez, S., Enßle, F. (2017). Earth Observation For Urban Sustainable Development: Advancements For Supporting Land Use Planning In Urban Areas.

Bärisch, S., d’Angelo, P., Müller, A., Lang, F. (2017). Operationelle Photogrammetrie aus dem All – Digitale Oberflächenmodelle aus höchstauflösenden Satellitenbilddaten. In: Geoinformationssysteme 2017 Beiträge zur 4. Münchner GI-Runde. Kolbe, T.H.; Bill, R.; Donaubauer, Andreas (Hrsg.) ISBN 978-3-87907-626-0

Carl, S., Bärisch, S., Lang, F., d'Angelo, P., Arefi, H., Reinartz, P. (2013). Operational Generation of High-Resolution Digital Surface Models from Commercial Tri-Stereo Satellite Data. 261-269.

Area A - Board 128

This paper demonstrates a new open source method for generating building height data at a city-wide scale from satellite stereo imagery. The method we describe involves generating a Digital Surface Model (DSM) from stereo imagery, Normalized DSM (nDSM) generation through terrain subtraction using publicly available DEM data, calibration and accuracy assessment of the resulting nDSM, and vegetation removal. In this paper, we use this method to generate building height data for Bangalore using Cartosat-I stereo imagery.
We use a recently developed open source algorithm called Satellite Stereo Pipeline (S2P) for generating 5m resolution Digital Surface Model (DSM) from Cartosat-I stereo imagery. Terrain subtraction for nDSM generation is achieved using 30m resolution SRTM data resampled to 5m. For calibration and accuracy assessment we use a sample set of building height estimates generated by counting the number of floors in buildings across Bangalore. Vegetation removal is achieved using classified 5m resolution LISS – IV (Linear Imaging and Self-scanning Sensor) data from the Indian Space Research Organization. Results show that our method can generate building height maps with an RMS error of ±3.5 meters. Since this is an error of approximately one floor, it should be sufficient for most city-scale urban analysis purposes.
Although Cartosat-I stereo imagery has been available since 2005, it has not found widespread use in generating building height maps of Indian cities, presumably since existing building height extraction workflows require expensive proprietary software and Differential GPS surveys for collecting Ground Control Points (GCPs). The method we describe is not constrained by such requirements and makes building height mapping inexpensive and more accessible. Irrespective of the method used, our paper presents the first published instance of generating building height maps for an entire Indian city.
Access to inexpensive building height data at a city-wide scale is particularly significant for rapidly urbanizing countries like India since most urban planning regulations are focused on managing intensity of use of land, and building height data is crucial to monitoring and analyzing this. Using the case of Bangalore, we also demonstrate how the building height data generated using our method can be easily combined with other datasets to analyze the intensity of use of land at a city-wide scale.

Area A - Board 129

City compaction is argued to be the most sustainable strategy for the development of cities by the European Commission’s Green Paper and the UN-supported Millennium Ecosystem Assessment. Theoretically, a compact city can preserve natural and agricultural land, reduce air pollution and noise from traffic, mitigate greenhouse gas (GHG) emissions and enhance walkable and vibrant city life. However, the environmental validity of city compaction is dubious based on the reviewed scientific literature because many studies are inconsistent among each other, thus hardly comparable. Residential population density, for instance, has been often used as a measure of city compaction and more frequently than any other proxy for urban development, simply because of such data availability. At the same time, residential population in many circumstances does not necessarily correspond to other aspects of city compaction (e.g., building density) due to changing housing space and lifestyle but also due to existing land-use policies, such as zoning regulation. Here, we propose an innovative remote-sensing approach that combines optical USGS Landsat 8 OLI and Copernicus Sentinel-2 MSI imagery and very high-resolution satellite imagery available via Google EarthTM mapping service to address changing urban density from horizontal and vertical view. To overcome the spatial resolution limit of Landsat 8 OLI and Sentinel-2 MSI imagery for building structure detection, we applied conditional generative adversarial networks (cGANs) to create time-series very high-resolution imagery. We reconstructed building density change from 2015 to 2018 by bringing the examples of Copenhagen, Denmark and Taipei, Taiwan—the large cities in Europe and Asia, to compare the performance of the developed method in different geographical settings. We define neighbourhood-level horizontal density as the ratio of the built-up area to the total area of a neighbourhood, and vertical density is a category of the mean building height (<4m, 4-10m, 10-20m, >20m) of a neighbourhood. The method has a four-step procedure. First, we segmented the multi-year Landsat 8 OLI and Sentinel-2 MSI images into homogeneous spectral objects (object-based imagery analysis, OBIA) across the years. Second, we downscaled medium-resolution imagery objects into high-resolution objects by cGANs, which developed cross-domain object relations trained by 2018 data and we used the trained relations to enhance objects’ spatial resolution for year 2015 to 2018. Third, we employed the object-based convolutional neural network (CNN) and used the high-resolution objects to predict neighbourhood building density. Forth, we evaluated the accuracy performance. The training and validation neighbourhood objects were labelled with vertical and horizontal density attributes derived from the digital surface model (DSM). During the session we expect to present preliminary tests of our method on deriving the spatial-temporal information of building density. We will also present the accuracy assessment tests about the suitability of Landsat 8 OLI and Sentinel-2 MSI imagery, for building structure detection separately for OBIA with CNN and CNN alone. We expect this study will of high interest of ESA Living Planet audience particularly in relationship to the application of satellite imagery to monitor urban development globally.

Area A - Board 130

The Contemporary Remotely Sensed Data, With High Spectral and Spatial Resolution, Enable the Monitoring of a Complex Urban Tissue, Giving the Opportunity to Scientists to Process Credential, High-Quality Spatial Data. The use of This Kind of Data Though, Demands the Adoption of New Approaches Regarding Their Processing. In This Research, Taking Into Account the New Demands That Arise From the Sustainable Urban Planning, a Knowledge-Based Framework Is Introduced for High-Resolution Satellite Image Processing, Regarding the Extraction of Bioclimatic Parameters of an Urban Tissue. Combined With Spatial Analysis, These Methods Are to Be Used in Order to Develop Strategies for Improving the Sustainable Footprint of an Urban Tissue.
In Specific, The 8-Band Multispectral Image of Worldview-2, Depicting a Densely Populated Urban Area in Northern Greece, Is Used for Extracting the Necessary Information of the Examined Urban Area. The Calculation of Morphological Parameters Such as Building Density, Urban Green Density as Well as Building Information (Roof Types and Orientation) Takes Place, Using the Derived Information From the Satellite Imagery.
The Second Part of This Study Discusses the Analysis of Urban Parameters That Define the Sustainability Footprint of an Urban Area. The Proposed Methodologies Use the Extracted Features From Satellite Processing for Estimating the Surface Roughness Parameters. "Surface Roughness" Is Defined As the Set Of Obstacles That Wind Can Encounter in a Structured Environment, Which in This Case Is the Building Stock. Two Basic Parameters Widely Used to Determine the Surface Roughness of an Urban Tissue Are the Zero-Plane Displacement Height (Zd) and the Roughness Length (Zo). The First Parameter Is Defined By the Average Height Of Roughness Elements Alongside With Building Density; While the Second Parameter Is Defined by the Average Height of Roughness Elements Alongside With Density of Buildings’ Facades. Having Calculated the Zd and Zo, the Air Flow Type and the Porosity Inside the Examined Area Is Extracted, Leading to Classifying the Tested Area Based on the Wind Flow. These Results Can Be Vital Input Information for Analysis and Assessment of Several Urban-Related Applications and Problems. Activities Such as Identifying and Better Controlling Sources of Airborne Irritants, Allergens, and Air Quality, the Design and Planning of a City's Growth and Development or Even Where to Plan Additional Shade Trees Plantings, Can Readily Benefit From the Implementation of This High-Resolution Approach.

Area A - Board 131

Valuation of land in developing countries can contribute to the achievement of various important global, regional and national development goals such as: sustainable urban development, promoting responsible investments and addressing conflicts often associated with large-scale land investments. Taxes on land and property can serve as an important tool to finance these development goals. The revenue is critical for municipalities and other local authorities to upgrade informal settlements, support the resettlement of displaced people and for infrastructure investment and development projects. In addition, it is vital for improving transparency in opaque land markets.

Rwanda has historically had relative low land taxes but these were recently increased. A new property tax law will come into force on 1st January 2019. Along with Rwanda’s good land governance, the new tax law may represent an opportunity to generate municipal revenue. Land value can be determined by a list of factors such as public investment in infrastructure, land use regulations, demographic change and economic development, private investment in land, and locational properties. Earth observation (EO) and geospatial analysis provide data, tools and methods that help to explain the relationship between land values and such explanatory variables.

This study trials a property valuation methodology for Rwanda's capital, Kigali by applying machine learning techniques to parcel transactions data from 2013 to 2015 and remote sensing data from 2009 and 2015 on building footprints. The building information (building footprint and building type) is derived from aerial images (2009) and a Pleiades scene (2015). Additionally, variables describing the spatial interrelations of properties within the city are derived from a multi-source dataset and include distance-based values (distance to central business district, schools, public transport, etc.), accessibility, location parameters (topographic position) and neighbourhood statistics (green area, urban structure type, density values, etc.). This approach can help to understand the key determinants of land and building values and be used to create a database with accurate estimates of property values for each parcel in Kigali. The study is conducted with the local authorities, led by the Ministry of Finance and Economic Planning (MINECOFIN). This database will be used to provide the national tax administration (Rwanda Revenue Authority) with an important tool for collecting property tax revenues. At present, the main method of property valuation in Rwanda is through taxpayer self-assessments. In order to support the tax administration in determining the validity of these self-assessments, the estimates from this database can be used to trigger certified counter-valuations if necessary with the aim of raising domestic tax revenues. Therefore, this study highlights how remote sensing methodologies can be effectively used to complement traditional valuation techniques and demonstrated through this application in Rwanda.

Area A - Board 132

The last few decades witnessed rapid urbanization all over the world. The World Urbanization Prospects 2014 reported over 54% of the world’s population living in urban areas, which will increase to 66% in 2050. The expansion of building and impervious surfaces especially in the urban core can change the thermal environment. Urban heat island (UHI) shows thermal intensification in urban areas compared to surrounding areas. Land surface temperature (LST) has been widely applied in assessing UHI. However, due to the limitation of LST retrieval method, few researchers attempted to track the long-term dynamics of UHI and its responses to progressive urbanization process based on Landsat series data. In this study, we improved the retrieval method to extract the average UHI intensity of Hangzhou city core in 1990–2010, and explored the dynamics of UHI composition and configuration and their responses to urbanization. Google Earth Engine (GEE) was applied to calculate the average UHI intensity in the study area. Hangzhou city core was heating up in study period with extensive conversions from neutral and cool area to heated area. The expansion of heated area initially radiated from the original urban core in linear configuration, and subsequently expanded to infill areas away from the core by replacing neutral and cool area. The significant differences in average UHI intensity between different land use and land cover (LULC) types can be observed: urban land had the highest value whereas forest and water body registered the lowest. The expansion of urban land at expense of natural land had exceeded a threshold to induce spillover of heat energy from the former to the latter. The findings can inform nature-based solutions to optimize the thermal environment and liveability of fast-growing subtropical cities.
Keywords: Urban heat island; Thermal comfort; Google Earth Engine; Spatiotemporal thermal dynamics; Spillover thermal impact; Hangzhou

Area A - Board 133

The SMURBS/ERA-PLANET project (“SMartURBan Solutions for air quality, disasters and city growth”) is one of the projects of ERA-NET‘s transnational call and falls under Strand 1 “Smart Cities and Resilient Societies”. SMURBS project combines Earth Observation expertise and smart city methods to address several urban environmental pressures. One of these is the urban agglomerations’ growth (SMURBS UG theme). This trend is very important, since the growth processes of cities around the world and in Europe have intensified recently and it is crucial to assess the current situation and ensure a high quality of life in such an environment through the use of "smart" technologies and offering support to informed decision making.
The most advanced intelligent methods of information processing such as neural networks and deep-learning are used to solve the UG problems [1, 2]. In general, services for UG estimation can be divided into two main categories. The first one is pixel-based classification services at the city scale. These classification maps are provided by two leading partners in the UG theme, Space Research Institute (SRI, Ukraine) and Italian Institute for Environmental Protection and Research (ISPRA, Italy). The maps include major land cover classes (artificial, water, vegetation),as well as a land cover classification based on EIONET EAGLE concept (buildings and flat sealed surfaces), non-vegetated bareland, snow & ice with a spatial resolution of 10 m and 5 m. Land cover classification maps are built with machine learning techniques [3]. By comparing land cover/land use maps for different time periods, assessment of land use change and provision of the recommendations for further urban development is possible.
On the base of pixel-based classification maps more high level services have been built for UG indicators estimation. To assess the dynamics of urban agglomeration growth and provide modern methods of managing a city, specialized and tailored indicators will be used during the project that should be integrated with socio-economic information. The main indicators will be SDG 11.3.1 (UN Sustainable Development Goals, Targets and Indicators), i.e. the variation in land consumption with respect to population variation (dimensionless) and per capita land consumption, — the rate of land consumption per inhabitant in a year (square meters per inhabitant per year). Other activities are devoted to developing landscape metrics and several indicators at the urban scale. The main indicators will be LCPI (Largest Class Patch Index),RMPS (Residual Mean Patch Size),ED (Edge Density), ID (Dispersion index),land consumption, Annual percentage increase in land consumption and Loss of agricultural, natural and semi-natural areas. All these indicators will derive from the ISPRA/SNPA National built-up binary classification map with 10 m spatial resolution.
Based on SRI’s classification maps, SMURBS will also plan to develop a Copernicus-compliant Urban Atlas and GHSL (Global Human Settlement Layer) product validation procedure at the city level. Copernicus Urban Atlas [4] is a vector map developed for numerous Functional Urban Areas in the EU with more than 100.000 inhabitants. However, this product is available non-annually (at the moment only for 2006 and 2012). SMURBS will target the development of a Copernicus-compliant Urban Atlas for the Kyiv city on an annual basis by utilizing other auxiliary information on the top of Copernicus satellite data, such as vector data from Open Street map (OSM) [5] for land cover statistics calculation over the urban area. . With regards to the GHSL at 20 m spatial resolution [6] that is developed by JRC, as for any global product validation with local in-situ dataset is required. At the moment, preliminary investigation of GHSL products and comparison with Kyiv city already produced maps has been performed.
All the solutions and services within the UG/SMURBS theme use Copernicus products and services as input data [7–8] (Sentinel1, Sentinel 2, HRL, CORINE Land Cover) and are based on high resolution datasets that is beyond the usual modus operandi in current city authorities' approach. New SMURBS services should provide migration in terms of up-to-date satellite data processing, enhancement time for service availability and frequency of products’ updates.
The UG smart solutions will provide an integrated set of tools, which can be applied by other cities-followers within SMURBS project or any other cities outside the project. In particular, it’s planned that the solutions discussed concerning UG indicators and land classification will be used by five Italian pilot cities (Monopoli – BA, Naples, Bologna, Venice and Trieste) [9]. All the services will be provided in the form of web portal based on open-source and free software, free satellite data and will be available for information visualization and can be embedded into a smart city portal. All the products and serviceswill be useful for the stakeholders and city administration at different levels, NGOs (like Kyiv Smart City Hub) and any other initiatives and organizations involved in urban development monitoring and planning.

References
1. N. Kussul, M. Lavreniuk, S. Skakun, and A. Shelestov, “Deep Learning Classification of Land Cover and Crop Types Using Remote Sensing Data,” IEEE Geoscience and Remote Sensing Letters, vol. 14, no. 5, pp. 778-782, 2017b.
2. A. Shelestov, M. Lavreniuk, N. Kussul, A. Novikov, and S. Skakun, “Exploring Google Earth Engine Platform for Big Data Processing: Classification of Multi-Temporal Satellite Imagery for Crop Mapping,” Front. Earth Sci., vol. 5, no. 17, pp. 1-10, 2017. DOI: 10.3389/feart.2017.00017.
3. S. Skakun, N. Kussul, A. Shelestov, M. Lavreniuk, and O. Kussul, “Efficiency Assessment of Multitemporal C-Band Radarsat-2 Intensity and Landsat-8 Surface Reflectance Satellite Imagery for Crop Classification in Ukraine,” IEEE J. of Select. Topics in Applied Earth Obser. and Rem. Sens., vol. 9, no. 8, pp. 3712-3719, 2016.
4. https://land.copernicus.eu/local/urban-atlas
5. https://www.openstreetmap.org
6. https://ghsl.jrc.ec.europa.eu/
7. https://land.copernicus.eu/user-corner/technical-library/clcplus-draft-technical-specifications-v4
8. https://land.copernicus.eu/
9. ISPRA-SNPA, 2018. Consumo di suolo, dinamiche territoriali e servizi ecosistemici. Edizione 2018, Rapporti, 288/18, ISPRA, Roma.

Area A - Board 134

Latest satellite-borne sensors provide access for researchers to dense time-series data and high spatial and spectral resolution data. These data are vital for a wide range of applications including urban area mapping and monitoring, slum detection, and crop mapping and yield estimation. In addition to advances in the technology used for remote sensing data collection, there have also been important scientific and methodological developments to enable the vast quantities of remote sensing data generated to be converted into useful information. Most of these developments have been based on supervised machine learning methods. Supervised machine learning methods, have gained popularity with the emergence of “big data” in remote sensing and high-performance computing. These methods face two interrelated challenges. First, target classes are highly variable and are usually embedded in a heterogeneous natural or anthropogenic landscape. Second, available training samples are not enough to adequately represent this high level of variability characterizing the target classes. These challenges are the main reason why supervised methods work efficiently on a local scale, but provide less reliable classification results when reused across different regions.

One way of addressing the first of the challenges mentioned above is to decompose landscape complexity into homogeneous ecological and socioeconomic strata or objects through regionalization and to use these objects as spatially explicit constraints when collecting training samples and running the classification procedure. Supervised methods achieve better classification results when there are large quantities of labeled training data available. Despite important efforts to collect and label training data using either conventional methods or crowdsourcing initiatives, collecting large amount of training data remains a tedious and expensive task. Remote sensing therefore requires methods that are able to cope with the relatively small amount of training data available. One possibility is to improve the generalization capability of supervised methods for different geographic regions through advanced methods such as transfer learning.

Transfer learning with domain adaptation methods aims to adapt a classification model trained on a particular image to another image with the same set of information classes and where the class distributions are correlated. A variety of transfer learning methods have been proposed and tested resulting in significant progress in the generalization capabilities of supervised methods. However, the successful use of transfer learning in remote sensing remains a matter of ongoing research mainly because previous research on transfer learning has focused mainly on what, and how it should be transferred, with far less emphasis on where to transfer to. Knowing where to transfer to is very important for avoiding “negative transfer learning”, which is likely to reduce the accuracy of classification results. Innovative solutions are required to guide generalization of supervised methods in the spatial domain.

The objective of this study is to evaluate empirically the generalization capability of Random Forest (RF) and Support Vector Machine (SVM) across different geographic regions. These two supervised classification methods have been selected due to their already demonstrated efficiency in classifying remote sensing data for different applications (Belgiu and Dragut, 2014). The innovation of this study is based on the fact the RF and SVM-based classification models trained on the source geographic region are transferred to target geographic regions that are identified as being similar to the source regions according to specific socioeconomic and biophysical conditions. The empirical evaluation is carried out for slums detection from high resolution remote sensing imagery. The urban ecoregions identified by Schneider et al. (2010) are used as spatially explicit regions to guide the generalization of supervised methods across similar regions. RF and SVM classification models are transferred to target urban ecoregions which are similar to the source urban ecoregions where the classification models are originally trained. Preliminary results on slums detection from Africa and South America show that the classification accuracy obtained by transferring RF and SVM classification models is higher when the source and target urban regions are similar from a socioeconomic and biophysical point of view. The results of this study are expected to make important contribution to the development of operational solutions required to convert remote sensing data into useful information, thus reducing the time-lag involved in providing decision makers with accurate data on slums.

References:

Belgiu, M., & Drăguţ, L. (2016). Random forest in remote sensing: A review of applications and future directions. ISPRS Journal of Photogrammetry and Remote Sensing, 114, 24-31

Schneider, A., & Woodcock, C.E. (2008). Compact, dispersed, fragmented, extensive? A comparison of urban growth in twenty-five global cities using remotely sensed data, pattern metrics and census information. Urban Studies, 45, 659-692

Area A - Board 135

Currently, an estimated 1.6 billion people live in inadequate housing globally, of which 1 billion live in slums and informal settlements. The rapid urban growth of recent decades has led to an increase in slums and informal settlements, as well as air pollution and inadequate basic services and infrastructure. The lack of proper urban planning and management bring tenure insecurity, increase poverty, pollution, health risks, as well as a higher vulnerability to natural disasters. To make urban spaces more inclusive, safe, resilient and sustainable the development, monitoring and management of better urban plans are urgently needed.

Earth Observation (EO) based services have a huge potential to help in addressing these critical gaps. EO can help both understanding and monitoring slums, but can also link their morphology with socio-economic data, identify hazardous areas where these settlements are located or model various services deprivation. Recent years have brought tremendous advancements in the area of automated information extraction from Earth Observation (EO) imagery, but even the most advanced state-of-the-art algorithms often do not provide a satisfactory solution when based on imagery alone.

Informal/slum areas receive a multiplicity of names in many languages, dialects and slangs all over the world and their definition (UN HABITAT, 2015) includes far more than morphological and contextual aspects, easily observable using EO techniques. Due to this broad definition and local specificities of slum areas, EO based workflows need the support of local in-situ information for calibration, machine-learning, QC/QA and validation as inputs into the automated detection and characterisation processes. Unfortunately, in general, there is a lack of appropriate in-situ data collections for slum mapping services since this is demanding both organizationally and financially or simply impossible due to access to these areas. In this context, an emerging area of the crowdsourcing of human intelligence, often with the help of gaming approaches, can be utilized to achieve the required extraction quality and operationalization capability for in-situ information.

The GAME.EO project explores the potential of a synergy of integrated crowdsourced and EO data-based information extraction framework for services related to current ESA engagement in sustainable development support (EO4SD) to the International Financing Institutions (IFIs). In the frame of the project, mobile-based tools are developed and tested to mobilize volunteers to feed EO-based information workflows in a timely and accurate manner. The approach is exemplified on the dedicated use cases for slum/informal settlements monitoring in Dhaka and Lima, to demonstrate added-value for current machine-learning algorithms for the identification, delineation and further characterization in these areas in support of SDG11.1 – ‘Ensure access for all to adequate, safe and affordable housing and basic services and upgrade slums’.

The GAME.EO project is funded by the European Space Agency (ESA) under the EO Science for Society Call. The project is led by GISAT, teaming with the International Institute for Applied Systems Analysis (IIASA) and Spatial Focus. The developed framework and tools are evaluated in cooperation with the multidisciplinary team of the World Bank Group’ Water Practice (GWAGP) under the ‘Water Supply and Sanitation in Rapid Urbanization’ umbrella in collaboration with the researchers from the University of Massachusetts Boston and the WASH Poverty Diagnostic team in Bangladesh.

Area A - Board 136

The capabilities of amplitude and coherence change detection based on Copernicus Sentinel-1 SAR data have been evaluated for the assessment of urban damage due to natural disasters and/or conflicts. The relevance of the topic "urban change detection" is strongly related to the emergency management and humanitarian assistance domains, since a SAR based approach can complement analyses based on optical data (with intrinsic limitations). The goal is to assess if changes in the backscattering response of the incoming microwave signal on man-made features can be related to consistent structural damages and, in case, automatically detected. Several case studies covering different urban environment, event type and level of detail (from single buildings to building blocks) are presented and discussed. The analysis of the test outcomes shows that group of damaged buildings or larger damaged structures could be detected by means of multi-temporal SAR data, while single damaged buildings in dense urban areas are often omitted. Furthermore the incidence of false positives/negatives negatively impact on the adoption of this approach as a standard methodology in emergency mapping.

Area A - Board 137

Urban land cover classification has always been crucial in providing links between many elements of human and physical environments. Timely, accurate, and detailed knowledge of the urban land cover extents from remote sensing data is increasingly required for various forms of applications. This surge of interest has been predominately driven by the recent innovation in open data access, processing and visualization technologies in the remote sensing field, giving us more capability to analyze objects on the earth surface, and consequently distribute them into classes of interest. Also, machine learning models like random forest (RF), support vector machines (SVM), classification and regression trees (CART) provide immense capability to model remotely sensed data in spaces where supervised classification can be performed.
To this end, this study presents a methodology for extracting urban extent layers from Sentinel-2A and Landsat-8 data, with all processes implemented in the cloud processing environment of Google Earth Engine (GEE) platform. This classification procedure was actualized using the Normalized Difference Spectral Vector (NDSV) feature as input to the classifiers, and tested in across locations in Nigeria.
Working with a collection of all images captured by the Landsat-8 and Sentinel-2 acquiring sensors between 01/07/2015 and 31/12/2015 (6 months), and considering the difference in temporal resolution of both sensors (5 days for Sentinel-2 and 16days for Landsat-8), image collections of different sizes were obtained in input for both data, and in both sites. It should be noted that the filtering of all the available images to a few ones with less than 10% cloud cover also played a role in determining the size of the input data collection. We considered first the two main cities of the countries, Lagos and Abuja. In Lagos, the input collection for Landsat-8 combined 5 images, while Sentinel-2 stacked 12 images. In Abuja, while the Sentinel-2 input data collection combined 8 images, Landsat-8 combined 7 images.
In all cases, our results show that, working with the classifier with highest accuracy in mapping, Sentinel-2A edges above Landsat-8 in accuracy of derived maps across all the test sites. This can be attributed to contributions from the higher spatial and temporal resolutions of the data.
In order to explore the possibility of obtaining a more accurate urban map, we considered to combine the properties of both Landsat-8 and Sentinel-2A data. The classifier with the highest accuracy, random forest in both input data cases and both sites, was chosen for this purpose. To this aim, we found that Random Forest classifier outperforms SVMs and CART in mapping urban extents in all the select test sites in Nigeria. However, all the three classifiers displayed relatively good performances due to their non-assumption of a parametric behavior of the input data.
In applying feature level fusion method, the 45 band NDSV feature obtained from Sentinel-2A data was combined with the 15 band NDSV feature obtained from Landsat-8 to form a 60 band composite NDSV feature. This feature was then fed into the classifier and an urban map was obtained. Results did not improve, as expected because the combination includes only redundant features.
Finally, we extended the test area, considering the training sets defined in Lagos and Abuja (with urban and non-urban classes), and classifying the whole Nigeria with Sentinel-2A and Landsat-8 input data. The resulting maps were then clipped to extract results obtained in the prior selected sites. Also, AND/OR operators were applied to fuse the maps before morphological filtering. As expected, combining both training set for classification at the national scale impacts local results. Hence, results at local levels, where the training set were selected changed, because the urban materials are not consistently the same all over the country. Moreover, the massive increase in the input data size leads to a dramatic increase in the amount of content in the information space to be split at each node of the random forest decision trees. This, of course, leads in turn to an increase in the probability of error in the decision making process (much more decisions to make on the same data set). Comparing this result to the one obtained by classifying in Lagos and Abuja using only training sets selected locally, it is clear that classifying Nigeria (at national scale) would need more training points, especially for the environments in the northern part of the country, not currently represented in the training set selection.

Area A - Board 138

The urgent need to monitor urbanization at the global level is the main driver for the recent advances in urban extent mapping from Earth Observation (EO) at multiple dates. Examples are the multi-temporal Global Human Settlement Layer [1,2], as well as the recent additions to the Global Urban Footprint data set [3], both focusing on the extraction of urban extents by means of multispectral sensors for past years, and multispectral and/or radar sensor for the more recent ones. The future is bright, thanks to the availability for free of multispectral and radar data sets by the Sentinel constellation [4-6].
In this work, the stress is on ways to further exploit the Sentinel data sets in conjunction with nighttime lights [7]. Urban layers for urban extent extraction using SAR Sentinel-1 (S-1) [8] and/or multispectral Sentinel-2 (S-2) [9] data are already available. This work aims at understanding which changes would be possible to detect at the block or building levels.
Indeed, although in the last years there have been papers devoted to urban extent monitoring, only recently more attention has been devoted to detect the changes in the two- and three-dimensional structure occurring inside urban areas around the world. In [10, 11], for instance, patterns due to the change of the power of return by microwave scatterometers in urban (and suburban) areas are described, extracting hints about the spatial patterns and structural features of different urban environments. A similar approach has been applied in [12], adding to power of return changes the nighttime light intensity changes, to recognize temporal patterns in 100 cities around the world. However, since scatterometer data allow to analyze trends in volumetric changes, but do not have enough spatial resolution to map the expansion of human settlement, fusion of information from other sensors is still required to map urban areas ``up and out". For instance, this issue is solved in [13] by using multispectral data, in addition to scatterometer and nighttime light measures.
We peopose an integrated approach to map all changes in urban areas, exploiting high spatial resolution Sentinel-1 (S-1) radar data. S-1 data are used in this work to achieve a more temporally and spatially fine analysis of the 2D and 3D changes occurring in fast growing megacities located in East Asia than existing techniques. We show that the approach using S-1 data as opposed to the original one with scatterometer data does not degrade the ability to discriminate among temporal growth patterns. On the contrary, it provides finer spatial and temporal details, and allows to automatically focus on the built-up area, as opposed to manually preselecting a few points close to a predefined geographical position, or to averaging data using administrative boundaries. Therefore this research shows that, by the use of S-1 data, it is possible to further differentiate the spatial and temporal change patterns of urban areas that, as a whole, experience the same reflectivity and nighttime light change pattern.
Experimental results focus on fast growing megacities in Asia, allowing a better understanding about the causes driving their changes in 3D. Examples for Jakarta, Beijing and Tianjin are discussed to prove this statement.

[1] M. Melchiorri, A.J. Florczyk, S. Freire, M. Schiavina, M. Pesaresi, and T. Kemper , “Unveiling 25 Years of Planetary Urbanization with Remote Sensing: Perspectives from the Global Human Settlement Layer”, Remote Sens. vol. 10, no. 5, 768, 2018.
[2] C. Corbane, M. Pesaresi, P. Politis, V. Syrris, A.J. Florczyk, P. Soille, L. Maffenini, A. Burger, V. Vasilev, D. Rodriguez, F. Sabo, L. Dijkstra, T. Kemper, “Big Earth data analytics on Sentinel-1 and Landsat imagery in support to global human settlements mapping” Big Earth Data, vol. 1, pp. 118-144, 2017.
[3] T. Esch, W. Heldens, A. Hirner, M. Keil, M. Marconcini, A. Roth, J. Zeidler, S. Dech, E. Strano, “Breaking new ground in mapping human settlements from space - The Global Urban Footprint,” ISPRS J. Photogramm. Remote Sens., vol. 134, pp. 30-42, Dec. 2017.
[4] R. Goldblatt, K. Deininger, G Hanson, “Utilizing publicly available satellite data for urban research: Mapping built-up land cover and land use in Ho Chi Minh City, Vietnam,” Development Engineering, vol. 3, pp. 83-99, 2018.
[5] J. Haas and Y. Ban, “Sentinel-1 SAR and Sentinel-2 MSI data fusion fo urban ecosystem service mapping”, Remote Sens. Applications: Society and Env., vol. 8, pp. 41-43, 2017.
[6] Y. Ban, L. Webber, P. Gamba, and M. Paganini, “EO4Urban: Sentinel- 1A SAR and Sentinel-2A MSI data for global urban services,” Proc. of the 2017 Joint Urban Remote Sensing Event (JURSE), Dubai, 2017, unformatted CD-ROM.
[7] Q. Zhang, K. Seto, “Mapping urbanization dynamics at regional and global scales using multi-temporal DSMP/OLS nightime light data”, Remote Sens. Environ., vol. 115, pp. 2320-2329, 2011.
[8] G. Lisini, A. Salentinig, P. Du, P. Gamba, “SAR-based urban extents extraction: from ENVISAT to Sentinel-1,” IEEE J. of Selec. Topics in Appl. Earth Observ. Remote Sens., vol. 11, no. 8. pp. 2683-2691, Aug. 2018.
[9] G.C. Iannelli and P. Gamba, “Jointly Exploiting Sentinel-1 and Sentinel- 2 for Urban Mapping,” Proc. of IGARSS’18, Valencia (Spain), July 22- 27, 2018, pp. 8218-8221.
[10] S.V. Nghiem, D. Balk, E. Rodriguez, G. Neumann, A. Sorichetta, C. Small, C.D. Elvidge, “Observations of urban and suburban environments with global satellite scatterometer data,” ISPRS Journal of Photogrammetry and Remote Sensing, vol. 64, pp. 67-380, 2009.
[11] L.H. Nguyen, S.V. Nghiem, G.M. Henebry, “Expansion of major urban areas in the US Great Plains from 2000 to 2009 using satellite scatterometer data,” Remote Sens. Environ., vol. 204, pp. 524-533, 2018.
[12] S. Frolking, T. Milliman, K.C. Seto, and M.A. Friedl, “A global fingerprint of macro-scale changes in urban structure from 1999 to 2009,” Environ. Res. Lett., vol. 8, 024004, 2013.
[13] D.L. Balk, S.V. Nghiem, B.R. Jones, Z. Liu, G. Dunn, “Up and out: a multifaceted approach to characterizing urbanization in Greater Saigon, 2000-2009”, Landscape and Urban Planning.

Area A - Board 139

Time series analysis of remote sensing can reveal urban growth patterns, helping to understand spatial-temporal trends of city development to, in turn, inform current and future planning and decision making on land use. The latter is important for the sustainable development of the Maritime Silk Road that connects tens of ports throughout Asia (Southeast Asia, South Asia, and West Asia), Africa, and Europe for opening a new doorway of tremendous opportunity for the local and regional economies of these port cities. Port cities are located along coastal zones of complex ecological and environmental settings and pursuing economic growth and development without compromising the nature, environment and biodiversity is a challenge. Drivers of city expansion, characteristics and rates of that expansion are valuable pieces of information that governments need to address the aforementioned challenge. Time series analysis of remote sensing feeding into landscape analysis models for quantifying urban expansion serve to that end, as demonstrated by this research.
By Selecting 35 port cities of Southeast Asia, South Asia, West Asia, Africa and Europe throughout the maritime silk route (Figure 1), we conducted Landsat-based time series image analysis (1990-2015) of land cover change, and using classification and regression trees models identified key characteristics of these cities, and conducted comparative analyses to establish similarities and differences of these urban development processes. We found that all the port cities expanded exponentially over the 25 years period examined, with the highest regional pattern of annual growth rate in Southeast Asia (6.77%), followed by South Asia, West Asia and Europe, and the minimum in Africa (3.85%). The total built-up land footprint of the 35 port cities increased at an average annual rate of 4.85% from 1990 to 2015; the average time for doubling the built-up footprint was 15 years, approximately.
Landscape pattern analysis and models of urban expansion showed three distinct growth patterns. Port cities of Southeast Asia did not follow a consistent pattern of development: capital port cities of this region exhibit a stable and compact spatial structure, while small and medium size port cities expanded by transitioning from diffusion to aggregation. On the other hand, port cities of West Asia and Africa showed compact and regular shapes, and their expansion overtime followed two types of spatial processes: edge-expansion towards sub-urban regions and infilling of voids within existing urban structures. The analysis also shows that port cities of Europe expanded synchronously following leapfrogging, edge-expansion and infilling growth models, developing complex shapes.
Our results point towards a development process of port cities consistent with diffusion-coalescence urbanization processes. The changes in various landscape indexes used in the research indicate that the port city tends to be compact. The fraction of leapfrogging, edge and infilling has changed form 4:4:2 to 3:4:3 over the past 25 years.
This study also shows the significance of remote sensing time series analysis in support of enhanced understanding of urban growth patterns of port cities overtime, for forward looking planning. It also builds evidence towards the need for consistent monitoring strategies to facilitate sustainable human settlement planning and management, for governments to align the vision of the Maritime Silk Road with the UN Sustainable development goal of making cities and human settlements inclusive, safe, resilient and sustainable.

Figure 1: Distribution of the 35 port cities along the Maritime Silk Road. Blue dots denote a low growth rate of urban expansion between 1990-2015, green dots denote a high growth rate, and red dots denote a very-high growth rate.

Area A - Board 140

The Landsat satellite mission has provided Earth observation imagery at medium resolution since 1972. Since 2008, data from these satellites has been made freely available, and currently accessibility is improved through the cloud-based Google Earth Engine, making the data and a library of functions accessible to the research community (Gorelick et al., 2017). The improved accessibility has led to a dramatic increase in usage for time series analysis to detect land cover change (Zhu, 2017). This paper explores the use of harmonic analysis for land cover change assessment for a study area in the south of England. The case is of particular interest because of the abundance of cloud cover and our interest in both urban expansion and intra-urban change. The method identifies land cover change to an overall accuracy of 79.6 %.

In order to detect urban growth, we have used, the change over time of the Normalized Difference Vegetation Index (NDVI). Abrupt decreases in NDVI can be attributed to urban replacement of vegetation (Marsek et al., 2000) whereas gradual changes may be related to maturing environments (Zhu, 2017). Land cover classes that include vegetation are expected to experience annual phenological variations in their NDVI. Harmonic functions can be used to estimate and generalise such variations. (Jung and Chang., 2015). Here, we used one-year period sinusoidal functions superimposed on a linear trend, similar to the application by Zhu and Woodcock (2014). Unlike this earlier paper, we divided the time period in two segments (before and after the land cover change) and fitted two different harmonic functions to both periods. We then estimate the time of land cover change by minimising the Root Mean Square Error (RMSE) over the combined time period. Thus, for each pixel we estimate the following model:

NDVI(t) = { t < t change: a₀ sin(2πt) + b₀ cos(2πt) + c₀t + d₀ + e₀ ,
t ≥ t change: a₁ sin(2πt) + b₁ cos(2πt) + c₁t + d₁ + e₁} ,

where, t is time in years, parameters a₀/₁, b₀/₁, c₀/₁ and d₀/₁ , are parameters to the function fitted with ordinary least squares linear regression and e₀/₁ is the error term. The parameter t change is found by iterating over the domain in increments of one year and retaining the parameter set with the lowest overall RMSE. If this shows an improvement relative to the RMSE of a single function for the whole time period, the pixel is deemed to have changed.

Firstly, the method was used to correctly identify known changes, confirmed by historical aerial photographs from Google Earth. The method was successful in identifying the location and time of change in pixels that were transformed from agricultural to residential land cover. It also succeeded in the more challenging case of urban regeneration where are pixel changed whilst remaining in the high density urban fabric class, and a third case where no transition took place and the pixel remained in a natural land cover. Whilst the availability of data is patchy, due to the abundant cloud cover the method appears to be robust under these circumstances, because the change detection is based on a change in trend that is apparent over a longer time period rather than the difference apparent in specific pairs of images.

We then applied this method to Swindon (a town in Wiltshire, England, which has undergone recent large scale development). We used all available Landsat data from 1985 – 2015, but only registered changes occurring between 2006 and 2015 (coinciding with the Centre for Ecology and & Hydrology (CEH) Land Cover Map 2007 (Morton et al., 2011) and 2015 (Rowland et al., 2017)) and produced a classification of change for the years 2006 – 2015.

The accuracy was assessed for 500 randomly selected pixels. Visual assessment was compared to the model layer and kappa, weighted kappa, producer’s, user’s, and overall accuracy were calculated for various thresholds. The final threshold was decided upon by balancing user’s and producer’s accuracy. The analysis was rerun using all available Landsat images from 2001 – 2018, however, whilst kappa and overall accuracy remained similar in this analysis, producer’s accuracy increased at the cost of user’s accuracy. As such, a longer time period is advised where a higher user’s accuracy is desirable, such as this study due to our interest in accurate determination of the location of change, therefore all data from 1985 – 2015 is used in subsequent analysis.

Fig. 1 shows classified change superimposed on a high resolution image of Swindon. The study area experienced several urban change events of agricultural land into suburban areas in the North West (Haydon Wick) and south (East Wichel), as well as road reconstruction of the A419. Reconstruction of Swindon Academy occurred during 2008 – 2009. These locations were assessed and were detected in the analysis showing robustness to different types of land conversion. The areas where this method underperformed was in agricultural land, where there may have been a notable change in phenology (e.g. due to a change in crop) but our visual assessment registered no change.

Comparison of change from post classification differencing of the CEH LCM 2007 and LCM 2015 reveals a significantly smaller amount of change. However, CEH documentation advises users to avoid using these maps for change assessment, however, for many applications, this remains the only recourse. CEH maps are known to overestimate the proportion of change despite being accomplished single date classifications. The post classification comparison of LCM 2007 and 2015 underwent accuracy assessment yielding an overall accuracy of 56.4 %. Despite the rudimentary choice of NDVI and indices, our map convincingly outperforms post classification change assessment of high accuracy land cover maps. This provides may provide an important improvement for post-classification change analysis for applications where temporal change is investigated. Currently, at least in urban change analysis, such applications are severally marred by data quality issues (Grinblat et al., 2016).



References
Gorelick, N., Hancher, M., Dixon, M., Ilyushchenko, S., Thau, D. and Moore, R., 2017. Google Earth Engine: Planetary-scale geospatial analysis for everyone. Remote Sensing of Environment, 202, pp.18-27.

Grinblat, Y., Gilichinsky, M. and Benenson, I., 2016. Cellular automata modeling of land-Use/Land-Cover Dynamics: questioning the reliability of data sources and classification methods. Annals of the American Association of Geographers, 106(6), pp.1299-1320.

Jung, M. and Chang, E., 2015. NDVI-based land-cover change detection using harmonic analysis. International Journal of Remote Sensing, 36(4), pp.1097-1113.

Masek, J.G., Lindsay, F.E. and Goward, S.N., 2000. Dynamics of urban growth in the Washington DC metropolitan area, 1973-1996, from Landsat observations. International Journal of Remote Sensing, 21(18), pp.3473-3486.

Morton, D., Rowland, C., Wood, C. Meek, L., Marston, C., Smith, G., Wadsworth, R., Simpson, I.C. 2011. Final Report for LCM2007 - the new UK land cover map. Countryside Survey Technical Report No 11/07 NERC/Centre for Ecology & Hydrology 112pp. (CEH Project Number: C03259).

Rowland, C.S.; Morton, R.D.; Carrasco, L.; McShane, G.; O’Neil, A.W.; Wood, C.M. (2017) Land Cover Map 2015 (25m raster, GB). NERC Environmental Information Data Centre.

Zhu, Z. and Woodcock, C.E., 2014. Continuous change detection and classification of land cover using all available Landsat data. Remote sensing of Environment, 144, pp.152-171.

Zhu, Z., 2017. Change detection using landsat time series: A review of frequencies, preprocessing, algorithms, and applications. ISPRS Journal of Photogrammetry and Remote Sensing, 130, pp.370-384.

Area A - Board 141

The refugee camp Zaatari was created as a result of more than seven years long civil war in Syria. The camp is rapidly evolving from the crowd of tents and caravans, into a real city.
Syrian families are abundant in closely related, many children were born in the camp and it is the only home they know. Therefore Zaatari is also called the City of the children.

There was a need for improved public infrastructure, which is not intended only for the basic needs of survival, but quality accommodation.
For a permanent stay in the better future of the inhabitants, a better quality architecture is needed, no longer a temporary crisis one. It is necessary to regulate public space, build a new school in and large family homes.

Our vision is to create a permanent, sustainable architecture that covers the needs of the inhabitants in the appropriate culture frame and serve the needs of the users, moreover to protect against the climatic conditions of the desert’s edge, in which the camp stands.

Spatial data and analyzes of camp rapid development and expansion were studied through time series of satellite images provided by UNISAT with UNITAR-United Nations Institute for Training and Research as their source from 2012 until nowadays.

Area A - Board 142

Since 2008, more than half of the world population live in cities, and by 2015, nearly 4 billion people -54 per cent of the world’s population - lived in cities. That number is projected to reach 5 billion by 2030 (UN, 2018). Rapid urbanization poses significant social and environmental challenges, including sprawling informal settlements, increased pollution, urban heat island, loss of biodiversity and ecosystem services. Therefore, timely, accurate and consistent information on urban growth patterns is of critical importance to support sustainable development. With the launches of the ESA Sentinel-1 and Sentinel-2 satellites, Synthetic Aperture Radar (SAR) and MultiSpectral Instrument (MSI) data with global coverage and frequent revisits become freely available. The Sentinel big data present an excellent opportunity for developing methods to support urban planning at local level as well as operational monitoring of the indicators of the UN 2030 Urban Sustainable Development Goal. The main challenge is the lack of robust and automated processing methods to extract valuable urban information from the massive amount of EO data.

The objective of this research is to develop novel methodologies to exploit Sentinel-1 SAR and Sentinel-2 MSI time series for urban change detection aiming at globally applicable methods. Previous studies have developed various SAR-based change detection techniques and applied them in variety of applications. However, most of these methods and algorithms were developed in consideration of the availability of a few or just a couple of images (pre- and post-). Urban change detection using SAR dense time series is still an open topic. Furthermore, due to the complexity of the urban environment, change detection in urban areas remains a challenging task. Many of the common issues, such as commission and omission errors due to seasonal or temporary environmental conditions, however, could be resolved by exploiting Sentinel time series using the highly redundant information following the so-called big data paradigm. The idea is to model the natural fluctuations (i.e. seasonal effect, weather condition) of the remote sensed measurements to decrease the confusion between the natural and man-made changes in urban areas.

A model-based urban change detection method was developed using multitemporal Sentinel-1 SAR data. Then an integrated approach between Sentinel-1 SAR and Sentinel-2 MSI data is being developed in order not only to detect changes but also to label the different types of changes (e.g., agriculture or forest to urban, old low-rise urban to new high-rise urban, etc.) using near real-time processing of the Sentinel big data. The preliminary results show that urban changes in general, new builtup areas in particular, could be detected reliably in a continuous manner (Fig. 1). The urban change information could be used not only to support sustainable planning at municipal and regional levels, but also contribute to the monitoring objectives of UN Sustainable Developments Goal (SDG) 11: Making cities and human settlements inclusive, safe, resilient and sustainable.

Area A - Board 143

The term asbestos refers to a group of fibrous silicate minerals recognized as carcinogenic to humans by the International Agency for Research on Cancer. Asbestos was commonly used in the past in products where asbestos fibers were bounded with cement forming corrugated tiles, commonly used as a roofing material mainly for industrial and rural buildings. Since 1987 it has been banned in most industrialized countries, although it is still used in poorer developing countries. Erosion and weathering processes caused by meteorological agents can degrade asbestos roofs exposing the population to severe health risk. On the other hand, roof reparation or asbestos removal need to be performed following strict safety rules. Therefore, public administrations and different stakeholders require detailed information on the quantity and location of asbestos roofs along wide geographic areas (e.g., regions or countries), so as to adequately manage this issue and plan successful remediation measures. The objective of this study was to design a methodology to detect asbestos roofs based on supervised classifications using high spatial resolution information obtained from orthophotos.
With this aim a case study was carried out focused in Navarre, a Spanish province with 10,391 km2, where a cadastral database of ~400,000 building polygons was analyzed. The main information source were two orthophotos acquired in 2017 and 2014 with a spatial resolution of 0.25 m and 0.50 m, respectively. The airborne sensor used had four spectral bands in the red, green, blue and near infrared spectral regions, whose mean and standard deviation were computed for each polygon. In addition the Intensity, Hue and Saturation channels were computed as an alternative representation of the color space that might be useful for roof material identification. Apart from the spectral information, several textural features were computed for each band based on the Grey Co-ocurrence Level Matrix (GCLM), i.e. Homogeneity, Dissimilarity, Contrast, Entropy, Angular Second Moment (ASM) and Correlation. In total 70 spectral-textural features were obtained for each building polygon.
Ground truth was done by means of field work and photointerpretation, leading to a database of 3,126 building polygons classified according to their roof material as (1) asbestos, (2) red metal, (3) green metal, (4) white metal, (5) terracotta tiles and (6) others. The database was divided in training (3/4) and test (1/4) sets. A Random Forest (RF) classification algorithm was implemented following a 4-fold cross validation procedure to avoid over-fitting. The selection of the input features to use was optimized based on the variable importance ranking produced by RF, as well as feature separability and correlation analysis. The optimum set of input features was composed of: mean values of the four spectral bands, Hue, Saturation, ASM and Contrast for both the 2014 and 2017 acquisitions. RF was parameterized with 700 trees and random selection of 4 variables at each node.
RF was run considering six roof classes and the results were recoded into a two class problem (asbestos vs. other materials), showing high performance indicators. The True Positive Rate (TPR) was 97%, the Precision of 95%, the True Negative Rate (TNR) was 98% and the Geometric Mean between TPR and TNR was 97%. In total 35,679 buildings were identified as having asbestos roofs, with a total projected roof area 7.5 km2. Besides the high performance achieved, RF was particularly useful as it allowed assessing the uncertainty of the identification of each building through the vote counting of the individual trees, showing a negative relation between building size and classification uncertainty. To conclude, the applied methodology can be easily adopted in other regions as long as up to date orthophotos and a GIS building layer are available.

Area A - Board 144

Outdoor air pollution is considered as the most serious environmental problem for human health, associated with some million deaths worldwide per year. Cities have to cope with the challenges due to poor air quality impacting human health and citizen well-being. Green urban infrastructure can improve air quality by removing air pollutants such as tropospheric ozone (O₃), nitrogen dioxide (NO₂) and particulate matters (PM₁₀, PM₂.₅).

Algorithms and models linking the properties of plant functional types, species groups, or single species to their impact on specific ecosystem services have been developed. However, due to the complexity associated with urban areas (e.g. high diversity of plant species, spectral similarity of tree species, urban forests are small sized and trees are often isolated, distinction between deciduous and coniferous tree), a comprehensive assessment of the role and impact of urban trees on air quality at city-scale is limited. The removal capacity of air pollutants is species-specific; therefore the finest classification of tree species is needed. Today, by combining remote sensing, modeling and GIS tools, it is possible to answer this question by identifying and mapping trees through vegetation indexes.

This study aims to i) establish a satellite-based methodology to detect, categorize and map tree species; ii) quantify air pollution removal by urban trees and shrubs; iii) provide novel insights on the management of urban green spaces to maximize urban air quality.

We have selected the city of Florence, Italy (380,000 inhabitants) as study area (72 km²). Different processing steps of the images (Pleiades at 0.5m*0.5m of spatial resolution) were realized: correction, fusion and textural and spectral classification. A validation of the tree species classification was performed by comparing the species-specific inventory, determined by field observation, with our categorization. The Leaf Area Index (LAI), the Normalized Difference Vegetation Index (NDVI) as well as the coverage fraction of each urban forest category were estimated. Finally, the deposition capacity of the common air pollutants has been estimated by i-Tree dry deposition model (i.e. city averaged). For instance, the urban trees removed about 98 tons annually for O₃ in Florence in 2017. Urban vegetation, as a cost-effective and nature-based approach, aids in meeting clean air standards and should be taken into account by policy-makers.

Area A - Board 145

No two cities are identical because they are living entities that evolve and require care to thrive. Earth observation (EO) of the urban environment is therefore essential to keep a watchful eye on their condition, however, other in-situ measurements are also needed to provide details that cannot be easily derived from remotely sensed imagery. The majority of Europeans today live in cities and/or urban agglomerations and the health of the urban citizens is influenced significantly by the health of the city. Our understanding of a healthy city is derived from multiple environmental indicators directly linked to the wellbeing and comfort of those living there. Cities and their urban population also have a significant effect on the environment including traffic, waste production and noise pollution to just name a few examples. Furthermore, the complexity of cities makes them particularly vulnerable to climatic changes. In order to monitor and develop cities in the best possible manner, big data and information methods must be applied in order to understand each individual system and plan for the future. Both EO and Internet of Things (IoT) together with crowdsourced information are crucial to gathering the relevant data to produce information on which to base healthy city growth, maintenance and change.

The GreenCities project [1,2]is being realised through a partnership between Terranea [3] and EOXPLORE [4] to develop innovative big data methods and applications that can improve the planning, management and control of urban green infrastructure. This work is supported by the German Federal Ministry for Economic Affairs and Energy (BMWi). Big data applications require relevant input data and the GreenCities project is concentrating on using remotely sensed imagery, IoT sensor networks as well as the input from local citizens. Today, EO is one of the best sources of data to monitor urban areas however, not all cities have the data, information, or resources to capture other types of vital information. The GreenCities project therefore, also looks at providing preparatory services related to the acquisition and integration of a variety of wanted data sources such as the locations of trees and their physical characteristics.

Green infrastructure, nature based urban solutions as well as ecosystem services all contribute to making cities more livable and sustainable while reducing the potential impacts of climatic changes. A number of services based on big data applications are being developed for the GreenCities project in order to prepare and monitor urban nature solutions and ecosystem services. Not all cities have extensive planning documentation especially about the location and state of their vegetation. Therefore, the GreenCityMap service is being developed to quickly and automatically detect green urban areas. The goal of this service is to provide a level of detail applicable to urban planning and the creation and maintenance of cadasters to record green infrastructure. As shown in the figure, automatically delineating tree crown and location from imagery with high accuracy is an important requirement for such a cadaster. Once such a cadaster is available, cities need to monitor continuously urban vegetation and therefore the project is developing the GreenCityMonitor service. Furthermore, the ability to measure the potential ecosystem services of individual trees within an urban setting will be provided by the UrbanTree service.
The GreenCities project will also provide relevant information that can help cities adapt to potential climate changes. The services related to this topic include CityHeatMonitor to monitor temperatures at the neighbourhood scale and the GreenRoofMap to help city authorities understand the potential of converting rooftops to use nature-based solutions for the reduction of urban heat islands. Together, the different services will be used to produce urban sustainability indicator maps which allow public authorities, decision makers and the city’s inhabitants better understand the potential for improvement based on ecosystem services.
The team will present the initial outputs of the GreenCities project related to ecosystem services and platforms from the first year of work including maps and monitoring services.

[1] GreenCities (DE) – https://greencities.terranea.de/
[2] GreenCities (EN) – http://www.eoxplore.com/greencities/
[3] Terranea – https://www.terranea.de/
[4] EOXPLORE – http://www.eoxplore.com/

Area A - Board 146

In 2015, the United Nation highlighted in Goal 11 of the SDGs (Sustainable Development Goals) the importance to “Make cities inclusive, safe, resilient and sustainable”. In order to monitor progress regarding SDG 11, there is a need for suitable indicators, representing different aspects of city conditions, obviously taking into account both the Land Cover (LC) changes and the urban climate with its most distinct feature, namely the Urban Heat Island (UHI) [1]. In particular, the term UHI refers to the mesoscale phenomenon, due to the impact of the alterations of land cover caused by human activities, for which an urban area is significantly warmer than its rural surroundings.
One of the different aspects of the UHI is the Surface Urban Heat Island (SUHI), which has been investigated through airborne and satellite remote sensing over many years. The aim of this work is to show the potentialities of the Google Earth Engine (GEE) [2] cloud platform to process free Earth Observation (EO) Big Data for long-term and wide spatio-temporal monitoring of SUHI and its connection with LC changes [1].
A large-scale spatio-temporal analysis was thus performed through GEE, also benefiting from the already established Climate Engine (CE) [3] tool to extract the Land Surface Temperature (LST) from Landsat imagery; the simple indicator Detrended Rate Matrix was introduced to globally represent the net effect of the LC changes on the SUHI phenomenon [1].
The implemented procedure (see Figure 1) was successfully applied to six metropolitan areas in the U.S., subjected to a heavy urban expansion in the last decades: a general increase of SUHI due to urban growth was clearly highlighted [1].
As a matter of fact, GEE indeed made possible to process more than 6000 Landsat images acquired over the period 1992–2011, performing a long-term and wide spatio-temporal study on SUHI vs. LC change monitoring. The present feasibility of the proposed procedure and the encouraging obtained results, although preliminary and requiring further investigations (calibration problems related to LST determination from Landsat imagery were evidenced), pave the way for a possible global service on SUHI monitoring, able to supply valuable indications to address an increasingly sustainable urban planning of our cities [1].


REFERENCES
[1] R. Ravanelli, A. Nascetti, R. V. Cirigliano, C. Di Rico, G. Leuzzi, P. Monti, and M. Crespi (2018). Monitoring the impact of land cover change on surface urban heat island through google earth engine: proposal of a global methodology, first applications and problems, Remote Sensing, 10(9):1488, doi: 10.3390/rs10091488.

[2] N. Gorelick, M. Hancher, M. Dixon, S. Ilyushchenko, D. Thau, and R. Moore (2017). Google Earth Engine: Planetary-scale geospatial analysis for everyone. Remote Sensing of Environment. 202, 18–27, doi: 10.1016/j.rse.2017.06.031.

[3] Huntington, J. L., Hegewisch, K. C., Daudert, B., Morton, C. G., Abatzoglou, J. T., McEvoy, D. J., and Erickson, T. (2017). Climate Engine: cloud computing and visualization of climate and remote sensing data for advanced natural resource monitoring and process understanding. Bulletin of the American Meteorological Society, 98(11), 2397-2410, doi: 10.1175/BAMS-D-15-00324.1.

Area A - Board 147

Urban areas face critical impacts from global warming, due to extreme meteorological events, including water scarcity, high temperatures, floods. Excessive urbanization, expansion of build-up areas and consumption of natural soil, make cities less resilient.
The climate phenomenon of heat waves is increasingly affecting many countries in the world, described as a prolonged period of extreme temperatures, much higher than the historical mean value in the same location and period. The frequency, length and intensity of heat waves is grown in the last decades. In Italy, the summers of 2003, 2007, 2010, 2015 and 2017 were exceptionally hot, characterized by various and persistent heat waves.
As a result, an increased number of elderly people suffers from severe health problems in summer due to the hot temperatures, with an increment of deaths during heat waves.
The effect of heat waves in the urban structure is not homogeneous: it is amplified in dense built-up areas, with narrow roads, non-natural soil covered by concrete or asphalt, traffic and other human activities. High buildings “capture” the heath and reduce the cooling effect of wind and vegetation, especially at nighttime. In some very densely built zones the night temperatures may remain hotter than in the outskirts, with peaks of up to 10°C exceeding the mean temperatures in neighboring rural zones. This phenomenon is known as “Urban Heat Islands” (UHI).
This study, carried out at e-Geos, a company jointly owned by Telespazio and Italian Space Agency, with the support of Legambiente and Greencitynow (for the 2017 year), was focused on the hot 2015 summer in Milan and 2017 summer in Rome. Satellite thermal imagery, especially nighttime acquisitions, integrated with ground meteorological and cadastral data, were used to deliver an operational tool to support urban planning to identify the best positions where to install adaptation devices, such as green and blue infrastructures (green roofs, gardens, planting of trees, fountains), for the reduction of extreme temperatures in the hottest areas.
The adopted methodology integrated daily Sentinel-3 SLSTR (1 Km res., for Rome 2017), MODIS (1 Km res., for Milan 2015) and Landsat-8 TIRS data (100 m res, 30 m pixel, for both Milan and Rome), to allow a more precise analysis of the thermal distribution in the urban areas, to produce the Thermal maps of the towns, with the distribution of night temperatures during the heat waves.
Through a GIS, the integration of population and cadastral data (especially of elderly people location) allowed the delivery of a Vulnerability Map of Milan and Rome urban areas, focused on the effects of UHI over the population.
To analyze the UHI effects, the following data were used:
- Meteorological data from ground stations (temperatures, humidity, winds)
- Daily (day and nighttime) thermal data from Modis (2015) and Sentinel-3 SLSTR (2017) sensors (1 Km res.), to highlight heat islands (Tday > 34°C, Tnight > 24°C);
- Landsat-8 TIRS nighttime data, to gather more detailed information on the thermal distribution in the towns;
- Landsat-8 OLI visible/NIR images to analyze urban structure and to derive the normalized index vegetation index (NDVI);
- Cadastral and Census data for vulnerability assessment to be managed through a GIS.

A comparison of thermal data derived from satellite imagery and those from ground meteorological stations confirmed the tight relation between satellite land surface temperatures and air temperatures as measured by ground stations.
The study showed that three different heat waves hit Milan and Rome during 2015 and 2017 summers and depicted heath distribution patterns inside the towns, highlighting the areas where Urban Heat Islands occurred. The final thermal map of Milan and Rome were crossed with the population data to assess the number of people under health risk during the heat waves. A more detailed analysis was carried out focused in Rome V Municipality (East area), using GIS Spatial tools (Sky View Factor, building height and other parameters) to establish the best and more effective location to install green infrastructures for a cooling effect over the hot temperatures and for the improvement of the life quality of the population.

Area A - Board 148

Sustainable management of urban areas has become one of the most important challenges today and in the decades to come, due to the pressures of rapid urbanization and population growth. Projections show that cities will host around 66% of the world's population by 2050, increasing the pressure on remnants of natural and semi-natural habitats. When these habitats are converted into artificial surfaces, local climate is altered, and the urban heat island effect occurs. This effect impacts human health and well-being of urban dwellers. There is a consensus that the network of natural and semi-natural areas, known as urban green infrastructure, influences the urban microclimate by ameliorating local climate, and creating cold islands. These reduce the risks to human health associated with heat waves, added to other benefits, such as pollutant capture. If the urban green infrastructure is planned and implemented strategically it can increase cities' resilience to climate change, which is expected to exacerbate the urban heat island effects. To a proper planning, remote sensing information can be used to estimate the contribution of green areas to microclimate control. However, holistic approaches and innovative methodologies are needed to better planning and management of green urban infrastructure. In this context, integrating remotely sensed data with other ground-based data sources can provide essential tools to support decision-making for more resilient and sustainable urbanization, and contributing to sustainable development goals.
In this work, evaluated the effect of the urban green infrastructure characteristics on the local urban climate regulation at multiple spatial scales of analysis, from the municipal to the park level. This was tested in a Mediterranean city, a region where large effects of climate change are expected, and where urban green infrastructure can contribute the most to ameliorate its effects. We combine multispectral (Sentinel-2A) and thermal (Landsat 8) data with in-situ observations based on biodiversity metrics. Here we used lichens, which are known to respond to microclimate changes. We also used local measures of temperature and relative humidity, using automated sensors. Land surface temperature was retrieved from Landsat 8 thermal data, extracted during the warm and daytime period when the surface urban heat island phenomenon is best observed. The land surface temperature was estimated by the Radiation Transfer Equation model. These data, along with biophysical characteristics describing urban green infrastructure structural patterns and configurations extracted by Sentinel-2A, were correlated with lichens trait-based metrics. The results indicate that the size and location of the green area affect the microclimate regulation. These data were compared with a very high spatial resolution land use map and the sensor outputs placed in-situ. The results showed that lowest temperatures were recorded on natural surfaces, such as water bodies and vegetated areas, especially when trees were present, whereas non-evaporative surfaces such as extensive open areas of pavement (e.g. airport and major highway intersections) and high density residential and industrial-commercial areas, recorded highest temperatures. The cooling distance of urban green areas was detected up to 200m into the built-up areas, but not in all types of land use.
These results can be used to improve the planning of urban green infrastructure, both at the city and park levels, ensuring a higher contribution of those areas to reduce the impact of the urban heat island in a context of climate change. We show the importance of relating remote sensing information with ground-data, for validation, to overcome the limitation of satellite-based metrics and to go beyond the quantification of environmental variables, looking at the ecosystem level.

Area A - Board 149