HiAccess

A parallel processing model for accelerating high-resolution geo-spatial accessibility analysis.

We present a parallel processing model, named HiAccess, to solve the high-resolution accessibility analysis problems in real time. One feature of HiAccess is a fast road network construction method, in which the road network topology is determined by traversing the original road nodes only once. The parallel strategies of HiAccess are fully optimized through theoretical analysis and experimental comparisons. Moreover, map generalization techniques are adopted to reduce computation load without accuracy loss. The flexibility of HiAccess enables it to work well when applied to different accessibility analysis models. HiAccess demonstrates the striking performance of measuring high-resolution (using 100m × 100m grids) accessibility of a city (in total over 250k grids, roads with 232k segments, 40 facilities) in 1 second without preprocessing, while ArcGIS takes nearly 1 hour to achieve a less satisfactory result. HiAccess is tested on much larger datasets with excellent performance.

The deployments of HiAccess using different accessibility analysis measurements:

CDM: Apply HiAccess to a distance-based measurement, namely, the closest-distance measurement.

OBM: Apply HiAccess to the opportunity-based measurement.

CDMMG: Use map generalization techniques to reduce computation of CDM without sacrificing accuracy.

Other measurements: To be continued...

Usage

Software dependencies:

MPICH (recommended version 3.0.4)

Boost C++ Libraries (recommended version 1.64)

Geospatial Data Abstraction Library (GDAL) >=2.0 (recommended version 2.1.2)

Compile:

Make sure that the header and lib files of Boost and GDAL are included in the Makefile, then run the following command to generate the executable program:

$ make clean && make

Dataset:

Two test datasets are provided in the data folder:

Tab1. dataset1

Name	Type	Size
Road	Road layer	16212 segments
Education	Facility layer	298 points

Tab2. dataset2

Name	Type	Size
Road	Road layer	265610 segments
Hospital	Facility layer	59 points

Run:

Parameters:

• --road Road_file: The road files are in Shapefile Format. Multiple input road files are supported. For multiple road files, use comma as separation (e.g. --road road1.shp,road2.shp.road3.shp ). The road files should be in Projection coordinate system with meter as measurement unit.
• --facility Facility_file: The facility file is in Shapefile Format. The facilities can be jobs, health care, educational resources and so on. The facility file should be in Projection coordinate system with meter as measurement unit.
• --resolution R(meters): The resolution of accessibility analysis. It is the size of the grid.
• --tolarence T(meters): Snapping tolerance of disconnected edges.
• --maxBound B(meters): Off-road travel is taken into account, and maxBound is the maximum distance for an off-road point to road segments.
• --rate Rate: The ratio of off-road and road travel cost.
• --output out_tiff: Output file is in GeoTIFF Format.

Examples:

$ mpirun -np 4 ./CDM/facilityAccess  --road ./data/dataset1/roads.shp --facility ./data/dataset1/education.shp --resolution 100 --tolarence 10 --maxBound 3000 --rate 2 --output ./results/dataset1_100m_CDM.tif

$ mpirun -np 4 ./CDMMG/facilityAccess  --road ./data/dataset1/roads.shp --facility ./data/dataset1/education.shp --resolution 100 --tolarence 10 --maxBound 3000 --rate 2 --output ./results/dataset1_100m_CDMMG.tif

$ mpirun -np 4 ./OBM/facilityAccesscount  --road ./data/dataset1/roads.shp --facility ./data/dataset1/education.shp --resolution 100 --tolarence 10 --maxBound 3000 --rate 2 --output ./results/dataset1_100m_OBM.tif

Performance:

We compare HiAccess with four representative accessibility analysis methods, namely, the GIS-based method (using Network Analyst extension in ArcGIS), the method based on spatial databases (using PostGIS with pgRouting), and two standalone accessibility analysis tools (UrbanAccess and AccessMod5). We choose the roads and hospitals data of Huayuan county as test dataset(seeTab2). In the contrastive experiments, we use the closest-distance model for accessibility measurement. The experimental environment is shown in Tab 3. The computation time and accessibility results of the methods are shown in Fig1. and Fig 2. HiAccess has the best performance with the best accessibility analysis result among the methods.

Tab2. Roads and hospitals of Huayuan County

Name	Type	Size
National Highway	Road layer	23646 segments
County Road	Road layer	632 segments
Rural Road	Road layer	205473 segments
Hospital	Facility layer	40 points

Tab3. Experimental environment

Item	Description
CPU	4core*2, Intel(R) Core(TM)[email protected]
Memory	16 GB
Software	ArcGIS 10.2, PostgreSQL 9.5, PostGIS 2.2, pgRouting 2.3.2, UrbanAccess 0.1.0, AccessMod_shiny 1.1.1
Operating System	Windows10, X86_64 for ArcGIS-based method.Ubuntu 16.04, X86_64 for PostGIS-based method, UrbanAccess and AccessMod5.

Fig1. Accessibility analysis results ( 100m × 100m grids) of hospitals in Huayuan county using different methods

Fig2. Computation time of the methods

HiAccess can be used to solve high-resolution accessibility problems on large datasets. We apply HiAccess to the medical facilities accessibility analysis in Beijing, China, of which the dataset scale is much larger than that of Huayuan County. Accordingly, we use a Symmetrical Multi-Processing(SMP) server for experiments. The accessibility analysis results of medical facilities in Beijing is shown in Fig 3. As Fig 4 shows, we conduct a series of experiments to evaluate the parallel performance of HiAccess. With the increase of process numbers, HiAccess achieves high performance of parallel acceleration, which is approximate to linearity when the process number is below 8. The final results are produced in only 10 seconds using HiAccess with 32 processes. Additionally, the performance of HiAccess not using map generalization is also tested as a contrast. The results show that map generalization greatly reduces the computation time without sacrificing measurement accuracy.

Tab4. Roads and medical facilities of Beijing from OpenStreetMap.

Name	Type	Size
Beijing Road	Road layer	1126123 segments
Medical facility	Facility layer	1481 points

Tab5. Experimental environment (SMP Server)

Item	Description
CPU	32core*2, Intel(R) Xeon(R)[email protected]
Memory	512 GB
Operating System	CentOS 6.5

Fig3. Accessibility analysis results (100m × 100m grids) of medical facilities in Beijing

Fig4. Parallel performance of HiAccess

Citation:

[1] Mengyu Ma, Ye Wu, et al. A parallel processing model for accelerating high-resolution geo-spatial accessibility analysis[J]. IEEE ACCESS, 2018 DOI: 10.1109/ACCESS.2018.2870168

[2] Mengyu Ma, Wei Xiong, et al. HiAccess: High-resolution Accessibility Analysis with High Performance Geographic Information System. In 7th ACM SIGSPATIAL International Workshop on Analytics for Big Geospatial Data (BigSpatial 2018), November 6, 2018, Seattle, WA, USA. ACM, New York, NY, USA, 4 pages. DOI： 10.1145/3282834.3282838

Contact:

Mengyu Ma@ National University of Defense Technology Email: [email protected]