FREEWAY SEGMENT SAFETY PERFORMANCE
FUNCTION (SPF) DEVELOPMENT
INTRODUCTIONS
o Scott Himes, VHB o Ian Hamilton, VHB
o Kendra Schenk, B&N
AGENDA
o Introduction and Objectives – Scott Himes
o Data Collection and Integration – Ian Hamilton o SPF Development Approach – Scott Himes
o Project-Level Validation– Kendra Schenk
o Questions
INTRODUCTION AND OBJECTIVES
Scott Himes
INTRODUCTION AND OBJECTIVES
o SPFs play a critical role in reliable safety
management
o SPFs serve two primary roles
o Network screening
o Project-level safety prediction
INTRODUCTION AND OBJECTIVES
o There are two methods for obtaining SPFs
o Calibrating existing SPFs
o Developing SPFs from jurisdiction-specific data
INTRODUCTION AND OBJECTIVES
o Project motivations
o Ohio completed calibration of HSM project design-level models o Prioritized site types predicting
poorly
o Desire for directional prediction o Counterintuitive relationships
when adding lanes in some
circumstances after calibration
INTRODUCTION AND OBJECTIVES
o Project Objectives
o Develop planning-level SPFs for ODOT freeways
o Develop bi-directional and directional project design-level SPFs consistent with those found in the HSM
o Evaluate project design-level SPFs along with HSM CMFs for performance
o Suggest updates to ODOT’s ECAT to incorporate recommended project design-level SPFs
DATA COLLECTION &
INTEGRATION
Ian Hamilton, AICP
DATA COLLECTION – STUDY AREA o Collected data for
several freeways.
o Interstates 70, 71, 75, 77, 80, and 90.
o State Routes 2 and 11.
o Combination of manual location and linear
referencing techniques.
DATA COLLECTION – EXISTING INVENTORIES
o Data already available through ODOT’s comprehensive set of road and traffic data.
o Through lanes
o Median type and width
o Shoulder widths (inside/outside) o Annual Average Daily Traffic (AADT) o Ramp locations
o Supplementary datasets support nuanced differences in different segments.
o Barriers (inside/outside) o Lighting
o Curvature
o Additional data collected to support directional and bidirectional model development.
DATA COLLECTION – RAMP LOCATIONS o Located ramp gores
and any associated taper starts/ends.
o Determined key characteristics:
o Speed change lane or lane add/drop.
o Left- or right-side ramp
o Flagged recent construction or atypical situations (e.g., toll plazas).
DATA COLLECTION – FINAL DATASETS
o Created 3 datasets for model development:
o Speed change lanes.
o Directional segments.
o Bidirectional segments.
o Used ODOT’s Linear Referencing System
(LRS) to help integrate data and calculate
relative positions of data along freeways.
DATA COLLECTION – LINEAR REFERENCING
o Linear referencing served as the basis for all segment data collection.
o Buffer and remove construction locations.
o Locate supplementary datasets on segments (e.g., barriers, curvature, and lighting).
o Locate relevant ramps (and ramp AADT).
o Locate upstream and downstream ramp gores.
o Routes dynamically segmented based on ODOT
base data to create uniform segments.
DATA COLLECTION – LINEAR REFERENCING
o Challenge working with LRS involved county-
based routes.
o Difficult to track
upstream/downstream features across county lines.
o Reconstructed a temporary LRS to
calculate upstream/
downstream ramps and
AADT
DATA COLLECTION – DIRECTIONAL SEGMENTS
o Directional segments developed before bidirectional segments.
o Construction during the study period and other atypical situations buffered and removed.
o Cross-section data reflect single direction of travel.
o If a segment contained an A-, B-, or C-type weave, the lower
“letter” was assigned to the segment (i.e., C over B and B over A).
DATA COLLECTION – BIDIRECTIONAL SEGMENTS
o Bidirectional segments derived from overlapping directional segments.
o Construction, speed change lanes, or atypical situations may
remove one travel direction; these were removed from bidirectional pool.
o Data aggregated between both directions (e.g., travel lanes) or assumed based on common values.
o Upstream and downstream elements determined based on directional data.
o If a directional segment was truncated or distances changed, this value was recalculated.
o Same rules applied for determining weave sections.
SPF DEVELOPMENT APPROACH
Scott Himes, PhD, PE
NETWORK SCREENING SPF DEVELOPMENT
o Network screening SPFs
o Function of segment length and AADT o Impact of other factors ignored
o Model forms tested
o Power model:
o Hoerl model:
o Predict crash frequency for bi-directional
segments
NETWORK SCREENING SPF DEVELOPMENT
o SPF development includes all segments o Freeway facility types evaluated
Area Type Lanes Segment
Type Number of
Segments Total Mileage Total
Crashes FI Crashes
Rural 4 Base 293 592.44 16,403 3,359
Interchange 331 103.05 4,112 829
6 Base 74 210.25 11,435 2,284
Interchange 92 21.84 2,056 422
Urban
4 Base 521 410.15 17,537 3,968
Interchange 1,355 359.83 33,718 8,012
6 Base 259 215.41 17,982 4,230
Interchange 930 273.90 59,836 14,517
8+ Base 81 46.96 8,001 2,073
Interchange 437 128.78 39,403 10,148
NETWORK SCREENING SPF DEVELOPMENT
o Example SPF representation
NETWORK SCREENING SPF DEVELOPMENT
o Cumulative residuals used for validation
PROJECT DESIGN-LEVEL SPF DEVELOPMENT
o Project design-level SPFs
o Include additional features to improve predictions
o Can be used to evaluate the effects of changing features o Subject to data availability
o Approaches tested
o Predict crash frequency for both directions combined o Focus crash prediction on individual direction
PROJECT DESIGN-LEVEL SPF DEVELOPMENT
o Crash types evaluated
o Multiple vehicle o Single vehicle
o Crash severity evaluated
o Fatal and injury
o Property damage only
o Individual severity evaluated using SDFs
PROJECT DESIGN-LEVEL SPF DEVELOPMENT
o Crash prediction model
o Basic freeway segments:
o Speed change lane segments:
o Attempted to identify AFs consistent with HSM o Some AFs excluded
o Data unavailable
o Values the same for all sites (LW generally 12 ft) o No significant relationship found
PROJECT DESIGN-LEVEL SPF DEVELOPMENT
o SPF applicability
Site Type Crash Type Crash Severity Area type Lanes
Freeway segments (fs)
Multiple vehicle
Fatal and injury Rural and urban All lanes Property damage only Rural and urban 2 lanes
Rural and urban 3+ lanes
Single vehicle
Fatal and injury Rural and urban 2 lanes Rural and urban 3+ lanes Property damage only Rural and urban 2 lanes
Rural and urban 3+lanes Entrance speed-change
lanes (en)
Combined Fatal and injury Combined 2 lanes
3+ lanes Combined Property damage only Rural and urban Combined Exit speed-change
lanes (ex)
Combined Fatal and injury Combined Combined
Combined Property damage only Combined Combined
PROJECT DESIGN-LEVEL SPF DEVELOPMENT
o AFs and applicability
Adjustment Factor Freeway Segments Entrance Speed-
Change Lanes
Exit Speed- Change Lanes MV FI MV PDO SV FI SV PDO FI PDO FI PDO
Inside shoulder width
Outside shoulder width
Depressed median width
Degree of curvature
Median barrier
Outside barrier
Downstream exit lane change
Upstream entrance lane change
Lane addition by ramp
Lane drop by ramp
Type A/B weaving section
Posted speed limit
Left-side ramp
Proportion curve
PROJECT DESIGN-LEVEL SPF DEVELOPMENT
o SDF Factors
Severity Factor Freeway Segments Speed-Change Lanes
KA Severity B Severity KA Severity B Severity
Posted speed Increase Increase Increase Increase
Outside shoulder width Increase Increase Decrease N/A
Inside shoulder width Decrease N/A N/A N/A
Median width Increase Increase N/A Decrease
Proportion outside barrier Decrease N/A N/A N/A
Average degree of curve N/A Increase Increase N/A
Urban area type Decrease Decrease Decrease Decrease
Proportion median barrier Increase Increase N/A Decrease
Presence of lighting Decrease Decrease N/A N/A
Ramp AADT N/A N/A Decrease Decrease
PROJECT-LEVEL
VALIDATION OF MODELS
Kendra Schenk, PE, PTOE, RSP2I
SAMPLE PROJECTS
26 Project Sites
Rural Urban
2 Lanes 5 sites 4 sites
3 Lanes 4 sites 5 sites
4 Lanes 0 sites 4 sites
5 Lanes 0 sites 4 sites
PROCESS
Uncalibrated HSM
Ohio- Calibrated
HSM
ODOT SPFs with HSM Adjustment
Factors
ODOT SPFs with ODOT Adjustment
Factors
2015 – 2019 Historical Crash Data
ADJUSTMENT FACTORS
Adjustment Factor ODOT HSMInside shoulder width
Outside shoulder width
Depressed median width
Degree of curvature
Median barrier
Outside barrier
Downstream exit lane change
Upstream entrance lane change
Lane addition by ramp
Lane drop by ramp
Type A/B weaving section
Posted speed limit
Left-side ramp
Proportion curve
Lane width
High volume
Shoulder rumble strips
Outside clearance
ADJUSTMENT FACTORS – MEDIAN WIDTH
Applies to all mediansHSM Data needs:
• Proportion of segment with median barrier
• Inside shoulder width
• Distance from inside shoulder to barrier face
Applies to only depressed mediansODOT Data needs:
• Proportion of segment with median barrier
• Median width
ADJUSTMENT FACTORS – DEPRESSED MEDIAN WIDTH
𝐴𝐴𝐴𝐴
3= 𝑒𝑒
𝑎𝑎(𝑀𝑀𝑀𝑀−60)×𝐷𝐷𝑀𝑀𝑎𝑎 = regression coefficient MW = median width in feet
DM = indicator for depressed median (1=yes; 0 otherwise)
> 60 feet AF < 1.0
< 60 feet AF > 1.0
ADJUSTMENT FACTORS – LANE CHANGES
Combines both directions together, combines HSM downstream exit and upstream entrance, and
includes Type B weaving section presence, proportion, and length in one Adjustment
Factor (AF7)
AF7 – Downstream exit lane changeODOT AF8 – Upstream entrance lane change
AF9 – Lane add by ramp AF10 – Lane drop by ramp
AF11 – Weaving section
ADJUSTMENT FACTORS – LANE CHANGES
AF9 – Lane add by ramp AF8 – Upstream entrance lane change
ADJUSTMENT FACTORS – EXIT LANE CHANGE
2.0-Mile Study Segment
7,500 AADT 0.1-mile S-C Lane
𝐴𝐴𝐴𝐴
7= 𝑒𝑒
𝑎𝑎(𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃 𝐷𝐷𝑃𝑃𝐷𝐷𝑃𝑃𝐷𝐷𝑃𝑃𝑃𝑃𝐷𝐷𝑎𝑎𝐷𝐷 𝐿𝐿𝐿𝐿 −0)𝑃𝑃𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟 𝐷𝐷𝑟𝑟𝐷𝐷𝑟𝑟𝐷𝐷𝑟𝑟𝑟𝑟𝑒𝑒𝑎𝑎𝐷𝐷 𝐿𝐿𝐿𝐿 = 𝑆𝑆𝐷𝐷𝑆𝑆𝐷𝐷𝐷𝐷𝑃𝑃𝑃𝑃 𝐿𝐿𝐷𝐷𝑃𝑃𝑆𝑆𝑃𝑃𝐿 𝐷𝐷𝑃𝑃𝑃𝑃𝐿𝑃𝑃𝑃𝑃 0.5 𝐷𝐷𝑃𝑃𝑚𝑚𝐷𝐷𝐷𝐷 𝑃𝑃𝑜𝑜 𝐷𝐷𝑃𝑃𝐷𝐷𝑃𝑃𝐷𝐷𝑃𝑃𝑃𝑃𝐷𝐷𝑎𝑎𝐷𝐷 𝐸𝐸𝐸𝐸𝑃𝑃𝑃𝑃
𝑇𝑇𝑃𝑃𝑃𝑃𝑎𝑎𝑚𝑚 𝑆𝑆𝐷𝐷𝑆𝑆𝐷𝐷𝐷𝐷𝑃𝑃𝑃𝑃 𝐿𝐿𝐷𝐷𝑃𝑃𝑆𝑆𝑃𝑃𝐿 𝑃𝑃𝑃𝑃 𝑀𝑀𝑃𝑃𝑚𝑚𝐷𝐷𝐷𝐷 × (
𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷 𝐸𝐸𝐸𝐸𝐸𝐸𝐷𝐷 𝐴𝐴𝐴𝐴𝐷𝐷𝐴𝐴 1,000
𝐷𝐷𝑃𝑃𝐷𝐷𝑃𝑃𝐷𝐷𝑃𝑃𝑃𝑃𝐷𝐷𝑎𝑎𝐷𝐷 𝐸𝐸𝐸𝐸𝑃𝑃𝑃𝑃 𝐿𝐿𝐷𝐷𝑃𝑃𝑆𝑆𝑃𝑃𝐿 𝑃𝑃𝑃𝑃 𝑀𝑀𝑃𝑃𝑚𝑚𝐷𝐷𝐷𝐷)
𝑃𝑃𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟 𝐷𝐷𝑟𝑟𝐷𝐷𝑟𝑟𝐷𝐷𝑟𝑟𝑟𝑟𝑒𝑒𝑎𝑎𝐷𝐷 𝐿𝐿𝐿𝐿 = 15.0
0.3-Mile Study Segment
7,500 AADT 0.1-mile S-C Lane
𝑃𝑃𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟 𝐷𝐷𝑟𝑟𝐷𝐷𝑟𝑟𝐷𝐷𝑟𝑟𝑟𝑟𝑒𝑒𝑎𝑎𝐷𝐷 𝐿𝐿𝐿𝐿 = 75.0
ADJUSTMENT FACTORS – LANE ADDITION/DROP BY RAMP
𝐸𝐸𝑟𝑟𝐸𝐸𝐴𝐴𝐴𝐴𝐷𝐷𝑇𝑇 = Entrance ramp AADT
LA = Indicator for the presence of a lane addition by entrance ramp (1 if yes, 0 if otherwise) 𝑎𝑎 = Regression coefficient
𝐴𝐴𝐴𝐴
9= 𝑒𝑒
𝑎𝑎 𝐸𝐸𝑃𝑃𝐸𝐸1,000 −0 ×𝐿𝐿𝐴𝐴𝐴𝐴𝐴𝐴𝐷𝐷𝐴𝐴𝐸𝐸𝐸𝐸𝐸𝐸𝐴𝐴𝐴𝐴𝐷𝐷𝑇𝑇 = Exit ramp AADT
LD = Indicator for the presence of a lane drop by exit ramp (1 if yes, 0 if otherwise) 𝑎𝑎 = Regression coefficient
𝐴𝐴𝐴𝐴
10= 𝑒𝑒
𝑎𝑎 𝐸𝐸𝐸𝐸𝐸𝐸1,000 −0 ×𝐿𝐿𝐷𝐷𝐴𝐴𝐴𝐴𝐷𝐷𝐴𝐴Lane Addition by Ramp
Lane Drop by Ramp
ADJUSTMENT FACTORS – WEAVING SEGMENTS
𝐴𝐴𝐴𝐴
11= 𝑒𝑒
𝑎𝑎(𝑀𝑀𝐷𝐷𝑎𝑎𝑊𝑊𝑃𝑃𝑃𝑃𝑆𝑆 𝑇𝑇𝑇𝑇𝑃𝑃𝐷𝐷 𝑋𝑋)𝑎𝑎 = regression coefficient
Weaving Type X = Indicator for the presence of applicable weaving section (1=yes; 0 otherwise)
ADJUSTMENT FACTORS – POSTED SPEED LIMIT
𝐴𝐴𝐴𝐴
12= 𝑒𝑒
0.012 ×(𝑃𝑃𝑆𝑆𝐿𝐿 −𝐵𝐵𝐿𝐿)𝑃𝑃𝑆𝑆𝐿𝐿 = Posted speed limit in mph
BL = Baseline for rural (70 mph) or urban (65 mph) segment
ADJUSTMENT FACTORS
Lane Width
ODOT roadways all had 12-foot lanes, unable to determine effect
High Volume
ODOT data not readily available
Shoulder Rumble Strips ODOT data not readily available
Clear Zone Width
ODOT data not readily available
Offsets to Barrier
ODOT data not readily available
EVALUATION BETWEEN MODELS
Uncalibrated HSM
Ohio- Calibrated
HSM
ODOT SPFs with HSM Adjustment
Factors
ODOT SPFs with ODOT Adjustment
Factors
2015 – 2019 Historical Crash Data
Mean Absolute Deviation (MAD) Root-Mean-Square Error (RMSE) Mean Absolute Prediction Error (MAPE)
CURE Plots
EVALUATION BETWEEN MODELS
Uncalibrated HSM
Ohio- Calibrated
HSM
ODOT SPFs with HSM Adjustment
Factors
ODOT SPFs with ODOT Adjustment
Factors
2015 – 2019 Historical Crash Data
EVALUATION BETWEEN MODELS
Ohio- Calibrated
HSM
ODOT SPFs with ODOT Adjustment
Factors
• Tends to underpredict crashes
• Applies more variables; more sensitive to geometric factors
• Provides a better fit for the speed-change lanes
• Tends to overpredict crashes
• Applies fewer variables; easier to implement
• Provides better fits by severity level and for the basic freeway segments
DATA NEEDS
Data InputArea Type ODOT HSMLength of freeway segment or s-c lane
Number of directional through lanes
Lane width
Inside and outside shoulder widths
Median width
Clear zone width
Presence of rumble strips
Freeway Segment AADT
Proportion of AADT during high-volume hours Distance to upstream entrance/downstream exit
Side of freeway ramp enters/exits
Ramp AADT
Curve length and radius
Length of median and outside barrier
Offset to barrier from traveled way
Posted speed limit
Type of weaving segment (A or B)
Presence of Lighting*
*Needed for Severity Distribution
NEXT STEPS FOR OHIO
o Update the ODOT-specific HSM analysis tools to reflect new SPFs and new adjustment factors
o Improve the fit of the models through more extensive data collection
o Horizontal curves o Clear zone
o Barrier presence and offset o Volume implications
ODOT SPFs with ODOT Adjustment
Factors