-
Paper 103 - Session title: Methods of Urban Mapping
10:00 A Multi-scale Remote Sensing Approach to Derive a London-wide Estimate of AGB
Wilkes, Phil (1,2); Disney, Mathias (1,2); Boni Vicari, Matheus (1); Calders, Kim (3); Burt, Andrew (1); Baines, Oliver (1) 1: UCL, United Kingdom; 2: National Centre for Earth Observation, NERC, UK; 3: Computational & Applied Vegetation Ecology, Ghent University, Belgium
Show abstract
Global populations are becoming increasingly urbanised resulting in unprecedented urban expansion. Green infrastructure plays a vital role in improving the lives of these new urbanites, for example, improving air quality, reducing heat stress as well as other health and well-being benefits. Urban vegetation also plays an (increasingly) important role in reducing the carbon burden of Earth’s expanding population, yet until recently, this has been mostly overlooked in carbon accounting terms. Reasons for this could include the difficulty in measuring urban tree demographics owing to high species diversity, atypical crown shapes and, from a remote sensing perspective, highly heterogeneous and dynamic land cover. In an attempt to overcome these challenges we present a new multi-scale remote sensing approach to quantify above ground biomass (AGB) sequestered in the street trees and parklands of Greater London (1,569 km2). Terrestrial laser scanning (TLS) was used to estimate the volume and AGB of 385 trees across the London Borough of Camden. Using these tree models, new allometry was derived to estimate stem volume from projected crown area and canopy height (r2 = 0.94). This was subsequently applied to tree crowns extracted from the UK Environment Agencies open-access airborne LiDAR dataset (covering ~15% of the Greater London area). Tree density and AGB estimates were then scaled to the whole Greater London area by training a random forest model with Sentinel 1 and 2 bands as predictor variables; computation was done using Google Earth Engine. Early results indicate that there are ~9M trees in the Greater London area and that small pockets of urban forest have a carbon density similar to temperate and tropical forests.
-
Paper 145 - Session title: Methods of Urban Mapping
10:20 Evaluating Sentinel-2 imagery for mapping human settlements
Schug, Franz; van der Linden, Sebastian; Okujeni, Akpona; Hostert, Patrick Humboldt-Universität zu Berlin, Germany
Show abstract
Mapping urban areas with remote sensing is particularly challenging due to the spatial spectral heterogeneity of human settlements and their complex dynamics related to land cover change. Sentinel-2 offers improved spatial, temporal and spectral resolution compared to similar globally operating optical systems and is, therefore, a highly interesting data source for urban monitoring. In this regard, efficient workflows are desired to make best use of Sentinel’s rapidly growing archive. Moreover, the contribution of Sentinel-2, as a globally available source, to the existing state-of-the-art in urban mapping must be better understood. Especially its complementarity to existing urban mapping products is of interest.
In this study, we used machine learning regression to produce an impervious surface fraction layer for Berlin, Germany, and surrounding areas from Sentinel-2 data. For model training, we used synthetic mixtures generated from an image spectral library representing pure surface cover types. We apply support vector regression (SVR) in order to efficiently and accurately produce surface cover fractions. Based on these fractions we performed a systematic quality assessment of our product in comparison to existing continental and global scale settlement maps, such as
- the decameter resolution Global Urban Footprint (GUF) based on TerraSAR-X data,
- the Global Human Settlement Layer (GHS) based on Landsat imagery and other open source data,
- the European Settlement Map (ESM) with a focus on built-up vs non-built-up areas based on SPOT imagery and the EEA Urban Atlas,
- and the Copernicus Imperviousness (CopImp) product.
Our library-based approach is independent from the tedious spatial definition of training areas and appears promising for the generalized mapping of built-up areas across multiple cities. For validation, we defined 13 regions of interest in neighbourhoods with different characteristics and performed a pixel-wise comparison with cadastral reference data. The accuracy of the SVR product is largely similar to that of the other evaluated datasets with regard to above-ground human infrastructure. As also reported in Klotz et al. (Rem. Sens. Of Env., 178, 2016), we find that GUF and GHS are particularly useful to map dense settlements, but systematically overestimate urban coverage in low-density built-up areas. Whereas accuracies for dense urban areas are similar among the datasets, GUF and GHS feature lower user’s and producer’s accuracies than SVR and ESM in some neighbourhoods. SVR is comparable to CopImp in terms of overall and area-specific RMSE values.
The challenge of conceptually delineating urban space is important. GUF, GHS, ESM and CopImp offer accurate information with regard to their respective objective. Beyond the classic mapping of densely built-up urban areas, machine learning regression might be particularly beneficial in areas with low built-up density and thus complementary to existing products. Using free Sentinel-2 data, SVR is potentially able to globally map settlements based on dense time series that also allow monitoring dynamic urban developments. Additionally, its ability to provide gradual land cover information is useful to better understand urban patterns. We therefore conclude that there is great potential of mapping a diversity of human settlements with Sentinel-2 when employing an SVR-based approach.
-
Paper 150 - Session title: Methods of Urban Mapping
09:00 Urban Extraction Using Sentinel-1 and Sentinel-2 Dense Time Series with Google Earth Engine
Nascetti, Andrea (1); Kakooei, Mohammad (1,2); Ban, Yifang (1) 1: KTH Royal Institute of Technology, Sweden; 2: Babol Noshirvani University of Technology
Show abstract
Urban population has grown rapidly in recent years and it is expected that more than 1.4 billion people will move into cities by 2030. The United Nations (UN) 2030 Agenda for the Sustainable Development Goals (SDGs) gives a prominent role to monitoring and study the urbanization process. Satellite data has played a fundamental role to retrieve the urban footprint at regional and global scale. Several methodologies have been developed using multispectral and/or Synthetic Aperture Radar (SAR) imagery and relevant research projects have been carried out to generate global datasets as the Global Urban Footprint (GUF) by the German Aerospace Center (DLR) or the Global Human Settlement Layer (GHSL) by the Joint Research Center (JRC). In particular, these datasets provide a reliable and global map of the urban areas with a quite low temporal resolution (every five years). The aim of this work is to develop an entirely automatic method to continuously monitoring urban areas using the Sentinel-1 and Sentinel-2 time series and exploiting the Google Earth Engine (GEE) computation capability. An automatic method for urban extraction was developed and the GEE scripts will be available as Free and Open Source Software (FOSS) to promote the comparison with other existing methods. The innovative aspect of this method is the combination of the Sentinel-1 SAR and Sentinel-2 MSI dense time series using a totally unsupervised approach. The estimation of the urban extent is performed in several progressive steps. First, the area of interest is divided into mountainous and non-mountainous areas to take into account the layover and foreshortening SAR geometric distortions. Then, Sentinel-1 ascending and descending time series are processed in order to enhance the more stable urban areas and an automatic thresholding procedure is used to compute the Initial-Urban mask combing the ascending and descending information. The latest step is to refine the Initial-Urban area by combining the vegetation and water masks generated using cloud free Sentinel-2 imagery.
We applied the developed approach in several cities including Beijing, Milan, New York, Rio and Stockholm in order to test it in areas characterized by different urban density and morphologies. We computed the urban extent in different periods to evaluate the stability of the method with different image stacks highlighting the capability of continuous monitoring of the urban areas. We evaluated the overall accuracy using 10000 validations points manually collected in stable urban and non-urban areas respectively in each city. The results show that is possible to obtain high accuracy (higher than 91 percent, Kappa > 0.82) in all the studied areas and in different periods. The comparison with the GUF and JRC data highlights that in most of the tests the accuracy of the developed method is higher and a visual comparison shows the high detailed information of the proposed method compared to the other achievable datasets thanks to the integration of the Sentinel optical and SAR data.
-
Paper 169 - Session title: Methods of Urban Mapping
09:20 Rapid, AI-Augmented Generation of Town-Plan-Level Vectors to Support Short-Timescale Crisis Response
Petit, David (1); Williams, Gareth (2); Nicholas, Chris (3); Fletcher, Martyn (4); Hart, Sophie (3); Holland-Lloyd, Gilbert (2); McMillan, Anneley (1); Napiorkowska, Milena (1) 1: Deimos Space UK; 2: BMT; 3: Qumodo; 4: DSTL
Show abstract
There is a need for up-to-date maps to be generated quickly to enable better decision making in humanitarian emergencies and crisis situations. Map updating is typically a time consuming and error-prone process. In recent years, machine learning algorithms have demonstrated the capability to extract objects (buildings, roads, vehicles etc) from satellite imagery (e.g. DSTL Kaggle competition [1]) or to entirely transform satellite imagery into new representations (such as Open Street Map) using deep neural networks. Nowadays, neural networks can outperform humans in specific tasks like classifying in fine grain categories [2, 3].
This work focuses on bringing together the best of the machine and the human to provide a fast and reliable up-to-date map. It engages with three challenges:
The development of a Machine Learning Framework for rapid mapping, incorporating Open APIs, considering the latest findings in Convolutional Neural Networks and Conditional Adversarial Networks
The relevance and usability of neural networks used in areas of the world where generally little and poor-quality training data sets are available
The integration of the human in the loop as a decisive factor in the actual efficiency of the process
Despite remarkable progresses on algorithms, we recognise that the overall efficacy of the process is not only determined by the performance of the algorithms but by the role of the human in this new workflow. Feature extraction for map production is not yet fully automated, and a task consisting of correcting small errors of a very well performing algorithm can be less productive than a task based on creating the output with a less efficient algorithm. The tediousness of a task impacts significantly the human productivity after a certain time (e.g. staff turnover) and must be taken into account in the design of the solution. This potentially adds a new dimension (appeal) to the Miller’s triangle[4] which describes the trade-off in a human task when automation modifies the ratio of the three variables Workload-Competency-Unpredictability
We will discuss the preliminary findings of our research, by reviewing the progress made by neural networks in the process of automatically and quickly extracting accurate urban features. The solution integrates open data of various quality and the skills of the geospatial analyst to create a “human in the loop”design. The study analyses the benefits of pretrained neural networks to mitigate issues with training data in places where ground truths are scarce. Training of algorithms are performed in two test sites, using SpaceNet and Worldview imagery combined with open data such as Open Street Map.
References
[https://www.kaggle.com/c/dstl-satellite-imagery-feature-detection
Project Adam: Building an Efficient and Scalable Deep Learning Training System. Trishul Chilimbi, Yutaka Suzue, Johnson Apacible, and Karthik Kalyanaraman. Proceedings of the 11th USENIX Symposium on Operating Systems Design and Implementation, 2014
“PowerAI DDL”. Minsik Cho, Ulrich Finkler, Sameer Kumar, David Kung, Vaibhav Saxena, Dheeraj Sreedhar. arXiv:1708.02188
“Designing for Flexible Interaction Between Humans and Automation: Delegation Interfaces for Supervisory Control “A Miller, Christopher & Parasuraman, Raja. (2007). Human factors. 49. 57-75. 10.1518/001872007779598037.
-
Paper 170 - Session title: Methods of Urban Mapping
09:40 Urban Structure Mapping and Change Analysis in Highly Dynamic Urban Areas based on Spectral and Three-Dimensional Analysis of VHR Remote Sensing Imagery
Warth, Gebhard (1); Braun, Andreas (1); Bachofer, Felix (2); Hochschild, Volker (1) 1: University of Tuebingen, Germany; 2: German Aerospace Center (DLR)
Show abstract
Rapid growing cities face challenges in planning infrastructure, regarding water and electricity supply, as well as waste treatment. Missing information on spatial population distribution and population trends make urban planning inaccurate and error-prone. To support urban planning, we developed an approach which considers these problems by exploiting very high resolution (VHR) remote sensing data for mapping urban structure and monitoring structural development. In this study, we present an approach based on the analysis of tri-stereoscopic Pléiades imagery from 2015 and 2017 covering the urban area of DaNang (Vietnam). These acquisitions offer a spatial resolution up 50 centimeters, which allows building detection and the generation of VHR digital surface models (DSM).
In the first component of our approach, it is intended to derive an initial current state of the urban structure of DaNang from the 2015 acquisition. Pléiades imagery offers three bands covering the visible spectrum and one near infrared band, which allow to precisely detect build-up areas. Streets have been masked out by building block information prior to the classification process. Especially the near infrared band has proven to be very valuable the detect and exclude water bodies and small vegetated areas. The application of object-based image analysis (OBIA) techniques enables classification of building types by means of area, shape and neighboring relations of built-up patches. Based on the statistical analysis of information on building type, built-up area ratio and vegetation cover ratio i.a., urban structure types could be assigned to building blocks. Our results show very good consistency in comparison with in-situ data.
The second component of our approach includes the integration of three-dimensional information. We photogrammetrically processed two tri-stereoscopic acquisitions of Pléiades from 2015 and 2017 to derive surface heights of the area. Due to discrepancies in the acquisition geometry, both scenes had to be adjusted vertically based on a linear regression which was based on 950 in-situ reference points and approximately 80.000 control points representing actual ground heights which were distributed with a stratified random sampling outside built-up areas and surfaces covered by vegetation. Both surface models were subtracted to retrieve vertical changes between 2015 and 2017. Based on the near infrared band, a vegetation mask was generated to exclude bushes and trees from interpretation of changes. First findings indicate clear patterns of construction work, especially along the shoreline in the west, in the Hòa Hải district in the South and on the development areas in Hòa Xuân district. Demolishions of buildings are concentrated in the central part of the city and around the airport. Negative differences have to be interpreted with caution because many of the removed buildings are expected to be replaced by more modern ones within short intervals.
Methods of Urban Mapping
Back2018-10-31 09:00 - 2018-10-31 11:00
Chairs: Soukup, Tomas (GISAT s.r.o.) - Sannier, Christophe (SIRS SAS)