Semantic Segmentation of High-Resolution Airborne Images With Dual-Stream DeepLabV3+

In geospatial applications such as urban planning and land use management, automatic detection and classification of earth objects are essential and primary subjects. When the significant semantic segmentation algorithms are considered, DeepLabV3+ stands out as a state-of-the-art CNN. Although the DeepLabV3+ model is capable of extracting multi-scale contextual information, there is still a need for multi-stream architectural approaches and different training approaches of the model that can leverage multi-modal geographic datasets. In this study, a new end-to-end dual-stream architecture that considers geospatial imagery was developed based on the DeepLabV3+ architecture. As a result, the spectral datasets other than RGB provided increments in semantic segmentation accuracies when they were used as additional channels to height information. Furthermore, both the given data augmentation and Tversky loss function which is sensitive to imbalanced data accomplished better overall accuracies. Also, it has been shown that the new dual-stream architecture using Potsdam and Vaihingen datasets produced 88.87% and 87.39% overall semantic segmentation accuracies, respectively. Eventually, it was seen that enhancement of the traditional significant semantic segmentation networks has a great potential to provide higher model performances, whereas the contribution of geospatial data as the second stream to RGB to segmentation was explicitly shown.

Cascade events in geographical space

In this paper, we present a simple discrete model of cascade behavior in an actual geographical space with built environments. By simultaneously triggering and relaxing random locations in a network of Voronoi cells interacting via the gravity model, we observe nontrivial statistics with heavy-tailed distributions of cells and actual area extents involved in the cascade. The distributions of these affected areas follow unimodal statistics, unlike the other externally-driven models operating over uniform neighborhoods that exhibit power-laws. Majority of the cascades are limited within the immediate neighborhoods of adjacent Voronoi cells, even for sufficiently large triggering magnitudes. The results are viewed from the perspective of inhomogeneous driving in sandpile-based models, and benchmarked with distributions obtained in other geographic datasets. The method offers a complexity perspective into the generation of large-scale events in physical and intangible flows, and explains their origins from cascaded accumulations of slow, random, and intermittent processes.

A Method for Constructing Geographical Knowledge Graph from Multisource Data

Global problems all occur at a particular location on or near the Earth’s surface. Sitting at the junction of artificial intelligence (AI) and big data, knowledge graphs (KGs) organize, interlink, and create semantic knowledge, thus attracting much attention worldwide. Although the existing KGs are constructed from internet encyclopedias and contain abundant knowledge, they lack exact coordinates and geographical relationships. In light of this, a geographical knowledge graph (GeoKG) construction method based on multisource data is proposed, consisting of a modeling schema layer and a filling data layer. This method has two advantages: (1) the knowledge can be extracted from geographic datasets; (2) the knowledge on multisource data can be represented and integrated. Firstly, the schema layer is designed to represent geographical knowledge. Then, the methods of extraction and integration from multisource data are designed to fill the data layer, and a storage method is developed to associate semantics with geospatial knowledge. Finally, the GeoKG is verified through linkage rate, semantic relationship rate, and application cases. The experiments indicate that the method could automatically extract and integrate knowledge from multisource data. Additionally, our GeoKG has a higher success rate of linking web pages with geographic datasets, and its exact coordinates have increased to 100%. This paper could bridge the distance between a Geographic Information System and a KG, thus facilitating more geospatial applications.

Spatially Disaggregated Car Ownership Prediction Using Deep Neural Networks

Predicting car ownership patterns at high spatial resolution is key to understanding pathways for decarbonisation—via electrification and demand reduction—of the private vehicle fleet. As the factors widely understood to influence car ownership are highly interdependent, linearised regression models, which dominate previous work on spatially explicit car ownership modelling in the UK, have shortcomings in accurately predicting the relationship. This paper presents predictions of spatially disaggregated car ownership—and change in car ownership over time—in Great Britain (GB) using deep neural networks (NNs) with hyperparameter tuning. The inputs to the models are demographic, socio-economic and geographic datasets compiled at the level of Census Lower Super Output Areas (LSOAs)—areas covering between 300 and 600 households. It was found that when optimal hyperparameters are selected, these neural networks can predict car ownership with a mean absolute error of up to 29% lower than when formulating the same problem as a linear regression; the results from NN regression are also shown to outperform three other artificial intelligence (AI)-based methods: random forest, stochastic gradient descent and support vector regression. The methods presented in this paper could enhance the capability of transport/energy modelling frameworks in predicting the spatial distribution of vehicle fleets, particularly as demographics, socio-economics and the built environment—such as public transport availability and the provision of local amenities—evolve over time. A particularly relevant contribution of this method is that by coupling it with a technology dissipation model, it could be used to explore the possible effects of changing policy, behaviour and socio-economics on uptake pathways for electric vehicles —cited as a vital technology for meeting Net Zero greenhouse gas emissions by 2050.

Improvements in Evaluating Grids for Basic Living Infrastructure: The Case of Gwangjin District in Seoul, South Korea

The purpose of this study was to analyze the current status and needs of infrastructure for basic life in Gwangjin district in Seoul, South Korea. In this study, we examined whether the national minimum standard was satisfied in terms of the infrastructure for basic life in the district. Specifically, we employed and compared the empirical utilities of two types of geographic datasets, 100-square-meter grids and 500-square-meter grids. The study compares the prediction accuracy between two types of geographic datasets by employing multivariate linear estimation using influential factors. The evaluation methods for prediction accuracy were to compare the root mean of squared error (RMSE) and mean of absolute error (MAE) from each dataset. The results were as follows: (a) the dataset with 100-square-meter grids showed more significant associations among influential factors and the infrastructure than the dataset with 500-square-meter grids; (b) the 100-square-meter grids showed better prediction accuracy compared with the 500-square-meter grids; and (c) in terms of basic level local government, it was more powerful to use the datasets with 100-square-meter grids for finding blind sides of infrastructure than the datasets with 500-square-meter grids. The results imply that it is necessary to adjust urban policy by using appropriate datasets, such as 100-square-meter grids.

Medical first response models in rural villages and towns: A simulation study of response times

Introduction Medical first responders (MFR) shorten the response times and improve outcomes in, for example, out-of-hospital cardiac arrests. This study demonstrates the usability of open geographic data for analysing MFR service performance by comparing simulated response times of different MFR models in rural town and village settings in Finland. Methods Community first response (CFR) models with one to three responders obeying the speed limit were compared to a volunteer/retained fire department (FD) model where three responders first gather at a fire station and then drive to the scene with lights and siren. Five villages/towns, each with a volunteer/retained FD but no ambulance base within a 10 km radius, were selected to test the models. A total of 50,000 MFR responses with randomly selected buildings as potential responder and patient locations were simulated. Results In central areas, the simulated median response time for the one-responder model was 1.6 minutes, outperforming the FD model’s simulated response time median by 4.5 minutes. In surrounding rural areas, the median response times of one- and two-responder CFR models were still shorter (15.0 and 15.9 minutes, respectively) than in the FD model (16.4 minutes), but the FD model outperformed the three-responder CFR model (16.8 minutes). Conclusion Open geographic datasets were useful in performing logistic simulations of MFR. Based on the simulations, CFR without emergency vehicles may reach patients faster than FD-based MFR in central areas, whereas in surrounding rural areas the difference is less pronounced.

A harmonised, high-coverage, open dataset of solar photovoltaic installations in the UK

Solar photovoltaic (PV) is an increasingly significant fraction of electricity generation. Efficient management, and innovations such as short-term forecasting and machine vision, demand high-resolution geographic datasets of PV installations. However, official and public sources have notable deficiencies: spatial imprecision, gaps in coverage and lack of crucial meta data, especially for small-scale solar panel installations. We present the results of a major crowd-sourcing campaign to create open geographic data for over 260,000 solar PV installations across the UK, covering an estimated 86% of the capacity in the country. We focus in particular on capturing small-scale domestic solar PV, which accounts for a significant fraction of generation but was until now very poorly documented. Our dataset suggests nameplate capacities in the UK (as of September 2020) amount to a total of 10.66 GW explicitly mapped, or 13.93 GW when missing capacities are inferred. Our method is applied to the UK but applicable worldwide, and compatible with continual updating to track the rapid growth in PV deployment.

Comparative of Machine Learning Algorithms and Datasets to Classify Natural Coverage in the Cajas National Park (Ecuador) Based on GEOBIA Approach

GEOBIA is an alternative to create and update land cover maps. In this work we assessed the combination of geographic datasets of the Cajas National Park (Ecuador) to detect which is the appropriate dataset-algorithm combination for the classification tasks in the Ecuadorian Andean region. The datasets included high resolution data as photogrammetric orthomosaic, DEM and derivated slope. These data were compared with free Sentinel imagery to classify natural land covers. We evaluated two aspects of the classification problem: the appropriate algorithm and the dataset combination. We evaluated SMO, C4.5 and Random Forest algorithms for the selection of attributes and classification of objects. The best results of kappa in the comparison of algorithms of classification were obtained with SMO (0.8182) and Random Forest (0.8117). In the evaluation of datasets the kappa values of the photogrammetry orthomosaic and the combination of Sentinel 1 and 2 have similar values using the C4.5 algorithm.

Interactive & Multiscale Thematic Maps: A Preliminary Study

<p><strong>Abstract.</strong> <i>Problem:</i> Interactive or “slippy” web maps have revolutionized cartography. Slippy maps present a single, coherently-designed reference map that can be panned to numerous geographic locations and zoomed across multiple scales. Further, they apply scale-dependent style rules to detailed geographic datasets, with the resulting designs rendered as a large set of interlocking tiles. To account for constraints in data bandwidth, processing, and storage, only those tiles relevant to the user’s location and past interactions are served into the web browser or other application, resulting in a seamless, real-time user experience of “a map of everywhere”. These slippy tilesets often are used as basemaps for advanced cartographic web and mobile applications, overlaying thematic information and other linework. Arguably, such slippy map mashups are the most common map seen and used today (and perhaps of all time). Yet, most of the cartographic design canon was developed long before slippy maps were possible. Do any of our time-tested design traditions in thematic cartography apply in today’s interactive and multiscale mapping context? In this presentation, we discuss preliminary insights from an online map study about the design of interactive and multiscale thematic maps.</p>