Implementing a GIS-based Safe Routes To School Route-Finding Methodology: Part 2

This is Part 2 of a 2-Part Series. In Part 1, we introduced the project and went over the steps we took to build the GIS-based Safe Routes To School route-finding methodology. Read Part 1 here.

Cooking Up a Usable Network

Prep the Data

To start, we create a new file geodatabase and add multiple datasets from a variety of sources. The datasets are used in preparing the centerline as a network dataset source. For our centerline, we use Oregon Metro‘s BikeNetwork Centerline as the source, which was derived from the City of Portland’s street centerline and includes the City’s trails.

This being a Safe Routes To School project for Portland Public Schools, the first thing we do is to clip to the City of Portland Boundary. Once that is done, we clean up the data by deleting and adding fields in the new feature class so we’ve got attribute fields without unneeded bulk. Some fields are added to capture specific criteria that are used in Portland Public Schools’ Supplemental Plan field evaluation of student walk areas. Among those fields are shoulder widths and conditions, motorist visibility, and other conditions that may exist that creates a risk (data not available in existing data sources). After this information is collected in the field, they will be added to the network dataset, and can be used in route analysis.

The next step is to edit the centerline feature class, to add routing specific attributes. There is considerable amount of manual editing to prepare the centerline for use in a network dataset. The information comes from various sources, and the data do not always align well with the centerline. The focus of this project is student walk areas 1-mile from elementary schools (K-6, K-8) and 1.5 miles from middle schools (6-8, MS), therefore the editing can be prioritized for those areas. The routing analysis will be performed within the walk area boundaries.

Following are some of the major editing steps

A street hierarchy rating, from 1 to 3, based on the street TYPE values is calculated. We use two sources of sidewalk data (one from the City and the other from Metro) to add sidewalk information. We need to use both sources of data as there is no complete sidewalk dataset for the entire City. The data from the City consists of a polygon layer offset from the street centerline. We need to split the centerline where there are sidewalk polygons, and add field values for right side, left side or both sides sidewalks. The Metro data has left side/right side attributes, is aligned with the centerline, and has a unique ID that can be used to join with the centerline data to calculated field values for sidewalks.

After the sidewalk attributes are in place, we affix the speed limit attributes.The number of lanes is an important consideration in determining safe routes. Again, we need to do a considerable amount of manual editing to prepare the data.

The next step is to convert the crosswalk line feature class to a point feature class, and then merge it with the traffic signals point layer. We then strip many of the fields, except for the ones we need, and save the two merged layer as one intersection layer in our new Pedestrian feature dataset. Sinuosity – curviness – is calculated on the centerline, and values added to a field in the centerline feature class.

The final step in preparing the data is a manual check to confirm that we have retained necessary fields and deleted extraneous ones.

Create the Network Dataset

To start, we create a new feature dataset in the file geodatabase that we created when preparing the data. The projected coordinate system that we use is the Oregon State Plane North, NAD 83 HARN. We then copy the prepared feature class to the feature dataset, and use a network dataset tool to create a new network dataset. Using the tool, we specify the attributes for shoulder, speed, road hierarchy, lanes, sinuosity, traffic signals and crosswalks. As you can imagine, we are looking for roads with wide shoulders, low speeds, low traffic and minimal lanes.

Generate walking routes

In this phase, we start to build the route maps; both the specific routes to individual students’ homes as well as more generalized, “catchment” routes that groups of students can use. Using a network analyst tool, the closest facility analysis will be used to map individual student walk routes from their homes to their school. Students will be used as the “incidents” in the analysis, and the specific school site will be the “facility.” This phase consists of a good deal of trial and error as we tweak the attributes and their impact on the route. Additionally, we add analysis of barrier layers. Barrier layers are not part of the network dataset; they are comprised of individual points, lines, or polygon feature classes that can be added to restrict or alter impedances of the network routing. We use land use, slope, crash and crime data for our barrier layers. They can easily be added or removed from the analysis, as can the network dataset attributes. Field tests will aid in refining the use of attributes and barriers, as well as highlight data gaps that exists.

Lessons Learned

Despite all of the great data available in the City of Portland, we still had a good deal of manual work to do. While the initial data cleanup and prep is time consuming, we anticipate that as data is maintained and properly curated, the process will become far more automated.

A Framework for GIS and Safe Routes to School

Last year, a group of experts convened in Austin, TX to discuss existing datasets and what is not being collected, how the general public can create and access data, existing tools and technology and what is needed to improve data connectivity and mapping, how GIS could be better utilized in Safe Routes to School and other active transportation initiatives. The result of the meeting was a paper titled, A Framework for GIS and Safe Routes to School which we used as a starting point in determining our methodology. In the paper, the authors came up with the Top Ten GIS Datasets for Safe Routes to School. Those ten datasets are listed below, followed by an explanation of how (or if) we used them.

Standard level of comfort - A fairly subjective measurement, but one that is clearly important. This dataset is not something that is immediately available, but will become available as field data-collection efforts get under way. Our initial field testing involved a good deal of discussion about the best ways to collect and quantify this type of data.

• Presence of a Sidewalk - This is an important attribute. We also took into consideration, and feel that it’s an extremely important consideration; the presence and width of a shoulder.

• Intersection - This is an important attribute. We weighed our routes taking into consideration attributes such as marked crosswalks and signals.

• Bicycle Facilities - We did not take this into consideration as walkability was the primary focus of this Proof-of-Concept project. We look forward to working on SRTS bike maps!

• School Location and Student Catchment Areas - Of course this was taken into consideration. How to determine catchment areas is a whole other discussion!

• Speed - We added in an additional factor of sinuosity after going out in the field and seeing first-hand how curvy roads (without sidewalks) can be a safety issue.

• Collision Data - This was a primary barrier attribute.

• Health Indicators - We did not use this data.

• Existing patterns - We did not use this data as there is no current existing walking pattern data.

• Crime data - This was a barrier, but not ranked as high as crash or slope data.

Don’t Miss Part 1

Did you miss the first post in this series? Fear not, you can read it here.

Implementing a GIS-based Safe Routes To School Route-Finding Methodology: Introduction

The Need For Safe Routes To Schools

The rapid decrease in the number of children walking to school, coupled with the equally rapid increase in childhood obesity, has made encouraging more children to walk to school a top priority for schools around the country. In 1970, half of all school children walked or rode their bikes to school. By 2009, that number was down to 13%.

So glad I don't have to drive this anymore.

Many parents cite “traffic-related danger” as a reason they won’t allow their children to walk or ride their bike to school, which leads to more traffic, and thus, fulfills their fears of a more dangerous street.

It is estimated that a quarter of morning rush hour traffic is comprised of parents bringing their children to school.

With these sort of statistics in mind, school systems around the nation have been implementing “Safe Route To School” (SRTS) programs to encourage kids and parents to walk or ride bikes to school.  One tactic of the SRTS program is to provide citizens with maps of safe and recommended routes to guide pedestrian and bike traffic to-and-from schools. Until recently, creating a walkability map consisted of getting people together who know the area and have them draw out the “best” routes to take, but as more and more schools are taking up the SRTS initiative, that methodology becomes harder to scale.

A GIS-based approach needs to be developed. One that can scale out and apply to many schools, including those with territories that combine many complex variables and conditions related to the walking environment.

Our Involvement with Portland Public Schools in their Efforts

Along those lines, we’ve been working with Portland Public Schools over the past couple of months to investigate ways to implement a GIS-based Safe Routes To School route-finding methodology. Creating such a methodology has been a much-discussed topic within the world of active transportation planning. While there is broad consensus about this need, how to satisfy it is a different story.

The first step in the project was to design an initial approach to doing the route finding analysis. To that end, we spent a good deal of time on research. Our team reviewed topical literature and interviewed local transportation experts, analysts, activists, wonks, and data stewards. Our discussions brought us to a variety of public agencies and departments in Portland and the State of Oregon, from the Portland Bureau of Transportation, to the Police, Metro, TriMet, and the Oregon Department of Transportation.

We engaged with people who work with Portland-area transportation data on a daily basis to improve our understanding of all the data available for analysis and their relative condition of completeness and accuracy.

We also wanted to get the experts’ opinions on the best approach to working with the data. The literature review was extensive, going beyond the SRTS-specific papers and into the realm of theoretical GIS analysis. Two documents, in particular, were very helpful starting points: A Data Model and Internet GIS Framework for Safe Routes to School identified a detailed data model for using street centerlines and the network source, and, A Framework for GIS and Safe Routes to School, identified datasets that would be helpful.

After Extensive Research, Our Methodology Takes Shape

After a number of interesting interviews, extensive lit review and some practical trial-and-error analysis, we have developed and refined a GIS-based Safe Routes to School route-finding methodology that we expect will be able to be scaled out and replicated for multiple schools beyond our test school. We still have some important field verification and refinement to do, but the initial results have been promising. Our methodology is broken into multiple phases.

The first phase, Data Preparation, is the most time-intensive, as it requires a number of steps to prepare source data for inclusion in a pedestrian network. In the second phase, the network is actually created. The final phase is focused on generating individual student routes and then more more general recommended routes from specified “gatherings” or “collector points” using a “closest facility analysis” technique.

The first two phases are a one-time effort to establish the network data model and the network itself. After this initial setup effort is complete, there will be on-going work related to maintaining, updating, and enriching the data used in the analysis to support ongoing improvement of the model and the results it produces.

There are different ways that one may start when creating a pedestrian network for use in Safe Routes To School route-finding. In the case of Portland, OR, we were faced with a couple of leading options, the first being to use the existing street centerline dataset and build on that, or, alternatively, to use distinct sidewalks and crossings datasets as the primary sources for development of a walking-focused network. Ultimately, we elected to use the street centerline, mainly due to gaps and suitability issues discovered with sidewalk data for this area, however, with limited modification, the same approach we have developed to apply to centerline data could used with a different network derived from sidewalk and crossing data, for instance.

Up Next: Our Recipe for Cooking up a Usable Network

I don’t want to leave you hanging, but I think we’ve covered a good amount of ground today. Join us for our next installment, where I show you a bit of the recipe that we cooked up. If you’d like, you can read about our adventures in field testing.

Field-testing safe walking routes methodology

Yesterday, I had the opportunity to go out in the field with my colleagues Molly Earle and Bryce Gartrell to do a little pre-test of a routing methodology that figures prominently in one of our current projects. We are in the process of helping the Portland Public School system  update their approach for determining safe walking routes to each of their elementary and middle-schools. I will dive into details of the methodology at a later date, but since it is still something that we are developing, discussion on that topic would be a bit premature at this point.

Yesterday’s field-check was a valued opportunity to assess how well our conceptual modeling and manipulations of data are capturing real-world factors that impact walking safety. Armed with old-school plotted maps and iPads alike, we traced a number of sample walking routes, observing in peripatetic fashion some of the strengths of the current routing solution and some of the areas where tuning and expansion may be required.  Braving rain showers, we logged several miles while closely assessing the effects of giving different weighting levels to different variables used to calculate routes.

Our test school is Rieke Elementary School in Southwest Portland, which somewhat uniquely combines a broad variety of terrain, traffic, infrastructure, and neighborhood conditions. Narrow, curvy streets that spill down steep hillsides, broad, multi-lane boulevards bustling with traffic and commercial activity, and sylvan, relatively flat and gridded subdivisions — Rieke has a bit of everything that occurs within the Portland Public School system.   Many of the residential areas from which students access the school were developed in the post-WWII years, a time when automobile ownership was skyrocketing, and streetcars were no longer a ubiquitous sight in Portland. As a result, there are far more dead-end roads and even fewer with sidewalks. For these reasons, Rieke is a great test area and quite a challenge.  (Someone among us also observed that our routes had us strolling past several brew pubs and offered that that could be counted among the benefits of selecting Rieke is a testing spot).  If we can solve routing at Rieke, it seems that most other schools should be significantly easier to work out.

It was nice to get out and see something materialize out of what has been conceptual for the past several months. While testing is a common feature of our work, I really enjoyed the opportunity for “real world” testing that this excursion provided. We will soon be going back to Rieke and undertaking a more extensive and formalized approach to testing our routing solution following some refinements that are being made based on yesterday’s experience.  As I promised, once we have finalized the methodology, which is intended to be one that is repeatable and that should transfer well to other areas, we will share more information and details.