Helsinki Region Travel Time Matrix 2018-2023

Introduction This travel time matrix records travel times and travel distances for routes between all centroids (N = 13132) of a 250 × 250 m grid over the populated areas in the Helsinki metropolitan area by walking, cycling, public transportation, and private car. If applicable, the routes have been calculated for different times of the day (rush hour, midday, off-peak), and assuming different physical abilities (such as walking and cycling speeds), see details below. The grid follows the geometric properties and enumeration of the versatile Yhdyskuntarakenteen seurantajärjestelmä (YKR) grid used in applications across many domains in Finland, and covers the municipalities of Helsinki, Espoo, Kauniainen, and Vantaa in the Finnish capital region. Data formats The data is available in multiple different formats that cater to different requirements, such as different software environments. All data formats share a common set of columns (see below), and can be used interchangeably. Helsinki_Travel_Time_Matrix_2023.csv.zst: comma-separated values (CSV) of all data columns, without geometries. This data set contains all routes in one file, and can be filtered by origin or destination according to the analysis at hand. The data records can also be joined to the geometries as available below. The file is compressed using the Zstandard algorithm, that many data science libraries, for instance, pandas, support transparently, directly, and automatically. Helsinki_Travel_Time_Matrix_2023_travel_times.gpkg.zip: an OGC GeoPackage standard file containing all data columns and the geometries that relate to the destination grid cell. The data set is delivered as a ZIP archive, which many GIS systems and libraries, e.g., GDAL/OGR, QGIS, or geopandas, support natively. Helsinki_Travel_Matrix_2023_travel_times.csv.zip: a set of 13132 comma-separated value files containing the routes to one destination grid cell each. The files contain all data columns, no geometry, and can be joined to the geometries as available below. Filenames of the individual files within the ZIP archive follow the pattern Helsinki_Travel_Time_Matrix_2023_travel_times_to_5787545.csv where 5787545 is replaced by the to_id by which the rows in the file are grouped. Use the from_id column to join with the geometries from one of the files below. Geometry, only: Helsinki_Travel_Time_Matrix_2023_grid.gpkg.zip: an OGC GeoPackage standard file containing the geometries and IDs of the grid used in the analysis. This file can be joined both to the from_id and to_id columns of the data files. The data set is delivered as a ZIP archive, which many GIS systems and libraries, e.g., GDAL/OGR, QGIS, or geopandas, support natively. Helsinki_Travel_Time_Matrix_2023_grid.shp.zip: an ESRI Shapefile archive containing the geometries and IDs of the grid used in the analysis. This file can be joined both to the from_id and to_id columns of the data files. Table structure from_id: ID number of the origin grid cell to_id: ID number of the destination grid cell walk_avg: Travel time in minutes from origin to destination by walking at an average speed walk_slo: Travel time in minutes from origin to destination by walking slowly bike_avg: Travel time in minutes from origin to destination by cycling at an average speed bike_fst: Travel time in minutes from origin to destination by cycling fast bike_slo: Travel time in minutes from origin to destination by cycling slowly pt_r_avg: Travel time in minutes from origin to destination by public transportation in rush hour traffic, walking at an average speed pt_r_slo: Travel time in minutes from origin to destination by public transportation in rush hour traffic, walking at a slower speed pt_m_avg: Travel time in minutes from origin to destination by public transportation in midday traffic, walking at an average speed pt_m_slo: Travel time in minutes from origin to destination by public transportation in midday traffic, walking at a slower speed pt_n_avg: Travel time in minutes from origin to destination by public transportation in nighttime traffic, walking at an average speed pt_n_slo: Travel time in minutes from origin to destination by public transportation in nighttime traffic, walking at a lower speed car_r: Travel time in minutes from origin to destination by private car in rush hour traffic car_m: Travel time in minutes from origin to destination by private car in midday traffic car_n: Travel time in minutes from origin to destination by private car in nighttime traffic walk_d: Distance from origin to destination, in meters, on foot Data for 2013, 2015, and 2018 At the Digital Geography Lab, we started computing travel time matrices in 2013. Our methodology has changed in between the iterations, and naturally, there are systematic differences between the iterations' results. Not all input data sets are available to recompute the historical matrices with new methods, however, we were able to repeat the 2018 calculation using the same methods as the 2023 data set, please find the results below, in the same format. For the travel time matrices for 2013 and 2015, as well as for 2018 using an older methodology, please refer to DOI:10.5281/zenodo.3247563. Methodology Computations were carried out for Wednesday, 15 February, 2023, and Monday, 29 January, 2018, respectively. 'Rush hour' refers to an 1-hour window between 8 and 9 am, 'midday' to 12 noon to 1 pm, and 'nighttime' to 2-3 am. All routes have been calculated using r5py, a Python library making use of the R5 engine by Conveyal, with modifications to consider local characteristics of the Helsinki use case and to inform the computation models from local real-world data sets. In particular, we made the following modifications: Walking Walking speeds, and in turn walking times, are based on the findings of Willberg et al., 2023, in which we measured walking speeds of people of different age groups in varying road surface conditions in Helsinki. Specifically, we chose to use the average measured walking speed in summer conditions for walk_avg (as well as the respective pt_*_walk_avg), and the slowest quintile of all measured walker across all conditions for walk_slo (and the respective pt_*_walk_slo). Cycling Cycling speeds are derived from two input data sets. First, we averaged cycling speeds per network segment from Strava data, and computed a ratio between the speed ridden in each segment and the overall average speed. We then use these ratios to compute fast, slow, and average cycling speeds for each segment, based on the mean overall Strava speed, the mean speeds cycled in the Helsinki City Bike bike-share system, and the mean between the two. Further, in line with the values observed by Jäppinen (2012), we add a flat 30 seconds each for unlocking and locking the bicycle at the origin and destination. Public Transport We used public transport schedules in General Transit Feed Specification (GTFS) format published by the Helsinki Regional Transport Authority, and adjusted the walking speeds (for connections between vehicles, as well as for access and egress to and from public transport stops) using the same methods as described above for walking. Private motorcar To represent road speeds actually driven in the Helsinki metropolitan region, we used floating car data of a representative sample of the roads in the region to derive the differences between the speed limit and the driven speed on different road classes, and by speed limit, see Perola (2023) for a detailed description of the methodology. Because these per-segment speeds factor in potential waiting times at road crossings, we eliminated turn penalties from R5. Our modifications were carried out in two ways: some changes can be controlled by preparing input data sets in a certain way, or by setting model parameters outside of R5 or r5py. Other modifications required more profound changes to the source code of the R5 engine. You can find a fully patched fork of the R5 engine in the Digital Geography Lab's GitHub repositories at github.com/DigitalGeographyLab/r5. The code that handles input data mangling and model parameter estimations is kept together with the logic to read input parameters and to collate output data, in the repository at github.com/DigitalGeographyLab/Helsinki-Travel-Time-Matrices.## 引言 本出行时间矩阵记录了赫尔辛基大都市区人口聚居区范围内，250×250米网格的所有质心（共13132个）之间，通过步行、骑行、公共交通及私家车出行的耗时与出行距离。根据应用场景需求，本数据集针对不同时段（早高峰、午间、非高峰时段）以及不同出行者身体能力（如步行、骑行速度）规划了对应的出行路径，详见下文。该网格遵循芬兰多领域应用中通用的**Yhdyskuntarakenteen seurantajärjestelmä（YKR）网格**的几何属性与编号规则，覆盖芬兰首都圈的赫尔辛基、埃斯波、考尼艾宁和万塔四座城市。 ## 数据格式 本数据集提供多种适配不同软件环境需求的格式，所有格式共享统一的字段集（详见下文），可互换使用。 1. `Helsinki_Travel_Time_Matrix_2023.csv.zst`：全量数据字段的逗号分隔值（Comma-Separated Values, CSV）文件，不含几何信息。该数据集将所有路径整合至单个文件中，可根据分析需求按起点或终点进行筛选，数据记录可与下文提供的几何信息进行关联。该文件采用Zstandard压缩算法，可被pandas等多数数据科学库直接、透明地自动读取。 2. `Helsinki_Travel_Time_Matrix_2023_travel_times.gpkg.zip`：符合OGC GeoPackage标准的文件，包含全量数据字段与目标网格单元的几何信息。数据集以ZIP压缩包形式交付，多数地理信息系统（GIS）及库（如GDAL/OGR、QGIS、geopandas）均原生支持该格式。 3. `Helsinki_Travel_Matrix_2023_travel_times.csv.zip`：包含13132个CSV文件的压缩包，每个文件对应单个目标网格单元的出行路径数据，包含全量数据字段但不含几何信息，可与下文提供的几何信息关联。压缩包内单个文件的命名遵循`Helsinki_Travel_Time_Matrix_2023_travel_times_to_5787545.csv`格式，其中`5787545`替换为该文件分组依据的`to_id`字段值。可通过`from_id`字段与下述几何文件进行关联。 4. 仅几何数据： - `Helsinki_Travel_Time_Matrix_2023_grid.gpkg.zip`：符合OGC GeoPackage标准的文件，包含本次分析所用网格的几何信息与ID，可与数据文件的`from_id`及`to_id`字段关联。数据集以ZIP压缩包形式交付，多数GIS系统及库原生支持该格式。 - `Helsinki_Travel_Time_Matrix_2023_grid.shp.zip`：ESRI Shapefile格式压缩包，包含本次分析所用网格的几何信息与ID，可与数据文件的`from_id`及`to_id`字段关联。 ## 字段结构 - `from_id`：起点网格单元的编号 - `to_id`：终点网格单元的编号 - `walk_avg`：以平均步行速度从起点到终点的出行耗时（单位：分钟） - `walk_slo`：以慢速步行从起点到终点的出行耗时（单位：分钟） - `bike_avg`：以平均骑行速度从起点到终点的出行耗时（单位：分钟） - `bike_fst`：以快速骑行从起点到终点的出行耗时（单位：分钟） - `bike_slo`：以慢速骑行从起点到终点的出行耗时（单位：分钟） - `pt_r_avg`：早高峰时段乘坐公共交通，以平均步行速度完成接驳的出行耗时（单位：分钟） - `pt_r_slo`：早高峰时段乘坐公共交通，以慢速步行完成接驳的出行耗时（单位：分钟） - `pt_m_avg`：午间时段乘坐公共交通，以平均步行速度完成接驳的出行耗时（单位：分钟） - `pt_m_slo`：午间时段乘坐公共交通，以慢速步行完成接驳的出行耗时（单位：分钟） - `pt_n_avg`：夜间时段乘坐公共交通，以平均步行速度完成接驳的出行耗时（单位：分钟） - `pt_n_slo`：夜间时段乘坐公共交通，以慢速步行完成接驳的出行耗时（单位：分钟） - `car_r`：早高峰时段驾驶私家车从起点到终点的出行耗时（单位：分钟） - `car_m`：午间时段驾驶私家车从起点到终点的出行耗时（单位：分钟） - `car_n`：夜间时段驾驶私家车从起点到终点的出行耗时（单位：分钟） - `walk_d`：步行方式下的起点至终点出行距离（单位：米） ## 2013、2015及2018年数据集 数字地理实验室（Digital Geography Lab）自2013年起开始开展出行时间矩阵的计算工作，期间方法学经历迭代，不同年份的计算结果存在系统性差异。部分历史数据集无法通过新方法重新计算，但我们已采用与2023年数据集一致的方法复现了2018年的计算结果，详见下文（格式与2023年数据集一致）。关于2013、2015年及采用旧方法计算的2018年出行时间矩阵，请参考DOI:10.5281/zenodo.3247563。 ## 方法学 本次计算分别基于2023年2月15日（周三）与2018年1月29日（周一）的交通场景。其中，“早高峰”指当日8:00-9:00的1小时时段，“午间”指12:00-13:00时段，“夜间”指2:00-3:00时段。所有路径计算均通过r5py实现——这是一款基于Conveyal公司R5引擎的Python库，我们针对赫尔辛基本地场景特性对其进行了修改，并基于本地真实数据集优化了计算模型。具体修改内容如下： 1. **步行模块**：步行速度及对应的步行耗时基于Willberg等人2023年的研究成果，该研究实测了赫尔辛基不同年龄群体在不同路面条件下的步行速度。本数据集选用夏季实测平均步行速度作为`walk_avg`（及对应`pt_*_walk_avg`）的依据，选用所有实测样本中最慢的五分之一步行速度作为`walk_slo`（及对应`pt_*_walk_slo`）的依据。 2. **骑行模块**：骑行速度基于两类输入数据集：首先，我们基于Strava数据计算各道路路段的平均骑行速度，并得到该路段骑行速度与整体平均速度的比值；随后基于该比值，结合Strava整体平均速度、赫尔辛基城市共享单车系统的平均骑行速度及二者的均值，计算得到各路段的快速、慢速及平均骑行速度。此外，参考Jäppinen（2012）的观测值，我们在起点和终点分别添加了30秒的自行车解锁与上锁耗时。 3. **公共交通模块**：我们采用赫尔辛基区域交通局发布的通用交通馈送规范（General Transit Feed Specification, GTFS）格式的公共交通时刻表，并采用与步行模块一致的方法调整公共交通接驳步行速度（含车辆间换乘步行、往返公交站点的步行）。 4. **私家车模块**：为还原赫尔辛基大都市区实际驾车速度，我们基于区域内代表性道路样本的浮动车数据，推导了不同道路等级、限速条件下的实际行驶速度与限速的差值，具体方法学细节详见Perola（2023）。由于该路段速度已包含路口等待耗时，我们移除了R5引擎原有的转弯惩罚项。 上述修改通过两种方式实现：一部分通过预处理输入数据集或在R5/r5py外部设置模型参数完成，另一部分则需要对R5引擎源代码进行深度修改。您可在数字地理实验室的GitHub仓库（github.com/DigitalGeographyLab/r5）获取经过完整补丁的R5引擎分支；处理输入数据预处理、模型参数估计的代码，以及读取输入参数、整理输出数据的逻辑，均托管于仓库github.com/DigitalGeographyLab/Helsinki-Travel-Time-Matrices。