| ArcGIS REST Services Directory |
| Home > services > SunCloud_Segments_Safety (FeatureServer) | | API Reference |
This description consists of the methods used to create the safety data layer for SunCloud. The Highway Safety Manual (HSM) was utilized in this screening process. A brief overview can be found here: https://www.highwaysafetymanual.org/Documents/HSMP-1.pdf
Crash Frequency is the 5-year sum of crashes within 150 feet from a segment if it has not been assigned to a junction.
Measures:
ATTRIBUTE NAME | DESCRIPTION |
CRASHES | All crashes within 150 feet from a segment if it has not been assigned to a junction. |
fatal_injury_crashes | All fatal and injury crashes within 150 feet from a segment if it has not been assigned to a junction. |
injurY_crashes | All injury crashes within 150 feet from a segment if it has not been assigned to a junction. |
non_motorized_FATAL_INJURY | All non-motorized injury and fatal crashes within 150 feet from a segment if it has not been assigned to a junction. |
impaired_driving_crashes | All impaired driving crashes within 150 feet from a segment if it has not been assigned to a junction. Determined on if there was a flag in the crash of “Alcohol Involvement” or “Drug Involvement” |
Crash rates describe the number of crashes in each period as compared to the traffic volume (or exposure) to crashes. Crash rates are calculated by dividing the total number of crashes at a given roadway section or intersection over a specified time-period (typically three to five years) by a measure of exposure. While traffic volume is the most typically used measure of exposure, others such as population, lane or roadway miles, and licensed drivers within a community also can be used. In this effort we used volume as Vehicle Miles Traveled (VMT). Crash rate screening can identify low volume, high crash risk locations that do not necessarily experience a high total number of crashes and are often more valuable than crash frequency.
To derive volume, the Average Annual Daily Traffic (AADT) was averaged from 2019 historical AADT and the Travel Demand Model.
Segments:
Vehicle Miles Traveled (VMT) = AADT * Segment Length
Million Vehicle Miles (MVM) = (VMT * 365 * number of years) / 1,000,000
Segment Crash Rate = Crash Frequency (5-year) / MVM
Measures:
ATTRIBUTE NAME | DESCRIPTION |
fatal_injury_rate | Fatality and injuries crash rate. |
injury_rate | Injury crash rate. |
Safety analysis is progressing at a promising rate and can be used to attain significant reductions in fatal crashes and crash severity. To achieve this, the SunCloud effort is making a significant commitment to developing and maintaining a comprehensive database of roadway characteristics combined with crash data and average annual daily traffic volume data that are all linked through a common linear referencing system.
This 2016 study was to develop a process to evaluate the Safety Performance Functions (SPFs) contained in the HSM for road segments and intersections on the Arizona State Highway System and to determine if those SPFs should be calibrated or if Arizona-specific SPFs should be developed. The recommendations are that ADOT move forward with SPF calibration for all HSM safety performance functions as for project-level safety analysis in Arizona. The predictive analysis used in this layer utilized those SPFs, along with PAG, and HSM Pre-defined SPFs. No SPFs were calibrated for the MAG or SunCloud region at the time of this layer creation.
The predictive method in the HSM includes a safety performance function (SPF). Two alternatives exist for applying the HSM prediction methodology to local conditions. They are either:
Cal1. Calibration of the SPFs found in the HSM
2. Development of jurisdiction specific SPF
MAG is currently in the process of calibrating their own SPFs. For SunCloud, we used a combination of AZDOT calibrated SPFs, PAG calibrated SPFs, and HSM pre-defined SPFs. Expected crashes are based on applying an HSM method to calculate observed crashes, estimate expected crashes based on Safety Performance Functions provided by ADOT, PAG, and from the HSM, and then use the Empirical Bayes method to calculate “excess expected crashes” as the difference between observed crashes and predicted crashes (adjusting for reversion to the mean patterns).
SPF KABC Calibration
A calibration effort was conducted that moved the focus of the predictive attributes to KABC crashes while adjusting for the data issue in PAG region. Predicted, Expected, LOSS, and Excess attributes are only based on KABC crashes because of the following:
· The City of Tucson does not capture property damage crashes and when viewing the full Sun Corridor region, it’s important to have an apples-to-apples comparison.
· SPFs developed in the PAG area utilize the crashes in the Tucson area, which develops more of a “KABC-only” SPF.
Due to these two issues, the SunCloud team re-calibrated ADOT and HSM Pre-defined SPFs based on similar PAG SPFs to calculate predicted KABC crashes. Re-calibration did not occur on PAG SPFs as they are already calibrated to the (KABC only) crashes in their region. LOSS and Expected crashes both utilized these predictive crashes and only used KABC historical crashes. The Excess Expected Crashes attribute now is KABC only due to the changes in the predicted and expected calculations.
Overall, while this method did address the data-issue in the PAG region, it also makes the predictive values more valuable. Without a KABC-only predicted calculation, there may be oversight from regional engineers when a corridor or junction is chosen that has excess property damage crashes than one that has excess fatalities.
SPF_ID | Agency | Type | Location | SPF Equation | Overdispersion Parameter | Calibration |
SC01 | ADOT | Segments | Rural 2-lane undivided | n = math.exp(-5.140+0.638 * math.log(aadt) + 0.719 * math.log(length) -0.045 * 4.48) | 0.789 | 1.33 |
SC02 | ADOT | Rural multi-lane undivided | n = math.exp(-5.970+0.702 * math.log(aadt) + 0.838 * math.log(length)) | 1.362 | No data / No need | |
SC03 | ADOT | Rural multi-lane divided | n = math.exp(-6.958+0.841 * math.log(aadt) + 0.828 * length) | 0.497 | 0.22 | |
SC04 | HSM | Rural 2-lane divided | n = math.exp(-0.312) * 365*10^-6 * aadt * length | 0.49 | 0.68 | |
SC05 | PAG | Urban 2-lane undivided | n = math.exp( -2.94)*[aadt]^0.48*[length] | 0.39 | PAG SPFs already calibrated to KABC only | |
SC06 | PAG | Urban multi-lane undivided | n = math.exp( -8.47)*[aadt]^1.12*[length] | 0.23 | ||
SC07 | PAG | Urban multi-lane divided | n = math.exp( -7.75)*[aadt]^1.00*[length] | 0.16 | ||
SC08 | PAG | Urban 2-lane divided | n = math.exp( -7.75)*[aadt]^1.00*[length] | 0.16 |
EB Method:
Incorporating the Empirical Bayes (EB) method into this development compensates for the RTM bias. The EB method can be used to calculate a site’s expected crash frequency. The EB analysis requires AADT, crash data, and SPFs to determine the average crash frequency at similar sites. The EB method pulls the crash count towards the mean, accounting for RTM bias. The reliability of the data affects the “weight.” The more reliable the data is, the more weight will go to the data; conversely, the less reliable the data is, the more the weight will go to the average. The sites expected crash frequency can be calculated is as follows:
EB Weight = 1 / [1 + (Predicted Average Crash Frequency * Number of Years) / Overdispersion Parameter]
Expected Average Crash Frequency = EB Weight * Predicted Average Crash Frequency + (1 – EB Weight) * Yearly Crash Count
More information can be found at the link below:
https://safety.fhwa.dot.gov/hsip/resources/fhwasa09029/sec6.cfm
Measures:
ATTRIBUTE NAME | DESCRIPTION |
Level_of_safety_service_kabc | Level of Service Safety (LOSS) value. |
PREDICTED_kabc_CRASHES | Predicted average crash count utilizing SPF. |
EXPECTED_kabc_CRASHES | Expected average crash count utilizing EB Method. |
EXCESS_EXPECTED_kabc_CRASHES | Excess expected crash count. |
https://safety.fhwa.dot.gov/hsip/resources/fhwasa09029/sec2.cfm
Predicted KABC Crashes - The predicted crash frequency is calculated using an SPF as defined above. It is a predicted yearly average crash amount based on roadway characteristics.
Expected KABC Crashes - The expected crash frequency takes the predicted crashes and is weighted with the observed crash frequency using the EB method. The EB method accounts for regression to the mean bias.
Excess Expected KABC Crashes - The predicted crash frequency determined from a SPF is weighted with the expected crash frequency using the EB method. The resulting weighted crash frequency is then compared to the expected crash frequency using the SPF to determine the difference between the two values. The excess average crash frequency is equal to the average observed crash frequency minus the average estimated crash frequency from the SPF.
Level of Service Safety (LOSS) KABC - This method compares the observed crash frequency and/or severity to the mean expected crashes using a SPF that is EB Weighted. The difference between the two values yields a performance measure that ranges between LOSS I and LOSS IV, with LOSS I indicating a low potential for crash reduction and LOSS IV indicating a high potential for crash reduction. Please note that LOSS values can be due to the lack of regional calibration and should only be used as an overview. Only KABC crashes were used in this calculation to account for a similar statewide coverage. In 2015, this method was proposed for using LOSS in concert with the EB method (i.e., difference between expected and predicted) to correct for RTM bias. Read more about that here.
LOSS-I - Indicates low potential for crash reduction.
LOSS-II - Indicates low to moderate potential for crash reduction.
LOSS-III - Indicates moderate to high potential for crash reduction; and
LOSS-IV - Indicates high potential for crash reduction.
For more information on LOSS, see the links below:
https://safety.fhwa.dot.gov/rsdp/downloads/fhwasa16027.pdf
LOSS Value = Historical Crash Count - Expected Crash Count
Excess expected kabc crashes (1.5 standard deviation) | level of safety service kabc |
<= - 25.91 | LOSS I |
- 25.91 to 1.0 | LOSS II |
1.0 to 25.91 | LOSS III |
> 25.91 | LOSS IV |