DOT Hazardous Liquid Pipeline data from 2002 to 2021 was analyzed to understand US pipeline incident experiences with CO2 pipelines. Since CO2 pipelines make up only a small percentage of overall hazardous liquid (HL) pipeline mileage and incidents, the experiences from all HL pipeline miles is often included to improve upon the statistical uncertainty.
The DOT database continues to evolve and, especially for earlier years’ data, the information suffers from ambiguity and inconsistencies in inputs.
The following analyses use extracts from datasets available from PHMSA public domain resources: Hazardous Liquid Pipeline Accident Reports for 2010 - present (PHMSA Form 7000-1 Rev. 07-2014) - Data Dictionary for Flagged Data file.
These incidents represent ‘reportable’ events per PHMSA definitions of triggers requiring filing of incident reports.
In some cases, data back to 2002 was added in order to better include the most rare characteristics such as extreme volume losses, fatalities, etc. Fatalities are assigned ‘cost’ values of $10M, and injuries ~$300K, consistent with US government guidance on ‘valuations for statistical life’.
For the 4329 incidents in the database, general data breakdowns by several categories are shown in the following tables and graphs.
IYEAR | Mileage |
2,010 | 4,521 |
2,011 | 4,735 |
2,012 | 4,840 |
2,013 | 5,190 |
2,014 | 5,276 |
2,015 | 5,240 |
2,016 | 5,195 |
2,017 | 5,207 |
2,018 | 5,206 |
2,019 | 5,076 |
2,020 | 5,150 |
All Hazardous Liquid Pipelines, 2010–2021
commodity | Count | Percent_Incidents |
CRUDE OIL | 2,189 | 50.6 |
NON-HVL | 1,384 | 32.0 |
HVL | 689 | 15.9 |
CO2 | 62 | 1.4 |
BIOFUEL | 5 | 0.1 |
Note in the following graphs, the difference between database fields of ‘release type’ versus ‘leak type’. Ambiguities such as these make certain interpretations challenging.
Slightly more than 2% of all hazardous liquid incidents involve a rupture, nearly 80% are leaks.
commodity | RELEASE_TYPE | Count |
CRUDE OIL | LEAK | 1,694 |
NON-HVL | LEAK | 1,105 |
HVL | LEAK | 532 |
CRUDE OIL | OVERFILL OR OVERFLOW | 192 |
CRUDE OIL | OTHER | 186 |
NON-HVL | OTHER | 113 |
NON-HVL | OVERFILL OR OVERFLOW | 102 |
HVL | OTHER | 98 |
CRUDE OIL | MECHANICAL PUNCTURE | 71 |
CO2 | LEAK | 52 |
CRUDE OIL | RUPTURE | 46 |
NON-HVL | MECHANICAL PUNCTURE | 40 |
HVL | MECHANICAL PUNCTURE | 29 |
HVL | RUPTURE | 29 |
NON-HVL | RUPTURE | 24 |
CO2 | OTHER | 8 |
BIOFUEL | LEAK | 3 |
CO2 | RUPTURE | 2 |
BIOFUEL | OTHER | 1 |
BIOFUEL | OVERFILL OR OVERFLOW | 1 |
HVL | OVERFILL OR OVERFLOW | 1 |
Consistent with all hazardous liquid pipelines, CO2 incidents rarely involve a rupture.
commodity | RELEASE_TYPE | Count | Percent of Incidents |
CO2 | LEAK | 52 | 83.9% |
OTHER | 8 | 12.9% | |
RUPTURE | 2 | 3.2% |
The table below shows an overall count and percentage of CO2 incidents by system part involved, as listed in the database.
| Characteristic | N = 621 |
|---|---|
| SYSTEM_PART_INVOLVED | |
| ONSHORE PIPELINE, INCLUDING VALVE SITES | 38 (61%) |
| ONSHORE PUMP/METER STATION EQUIPMENT AND PIPING | 23 (37%) |
| ONSHORE TERMINAL/TANK FARM EQUIPMENT AND PIPING | 1 (1.6%) |
|
1
n (%)
|
|
This section narrows the focus to the ‘probability’ side of the Risk = Probability x Consequences equation.
This is the average frequency of failure per 1,000 miles by year. It was derived by taking the number of incidents per year divided by the total mileage reported on the annual reports submitted to PHMSA.
CO2 pipeline history is shown by the blue line and overall HL’s are in red.
Frequency of Failure
For the years 2002 and later the count of onshore pipeline incidents by cause was compiled as shown below.
For the years 2002 and later the count of CO2 pipeline incidents by cause was compiled as shown below. Note differences from overall HL chart above.
Examining data since 2010, this is the annual percentage of all hazardous liquid incidents reported grouped by cause. All HL’s are used here to benefit from the larger amount of records available.
Or, viewed with an alternative graphic:
Frequency of Failure
Incident causes and whether the locations were previously inspected by an ILI, are shown here. This provides some evidence for effectiveness of ILI on preventing incidents related to various failure mechanisms.
In this summary the count for that category is at the top and then each row under that column is the fraction of that column total. ie, column percentages total to ~100%.
| Corrosion Incidents and Whether Inspected by ILI | |||
|---|---|---|---|
| Characteristic | Overall, N = 5551 | NO, N = 3581 | YES, N = 1971 |
| CAUSE_DETAILS | |||
| EXTERNAL CORROSION | 238 (43%) | 99 (28%) | 139 (71%) |
| INTERNAL CORROSION | 317 (57%) | 259 (72%) | 58 (29%) |
|
1
n (%)
|
|||
This summary is for incidents noted to be related to Excavation and Natural Force Damage. The Count is the number of incidents since 2010 and the Total CoF is the total consequences in millions of dollars related to those incidents.
CAUSE | CAUSE_DETAILS | Count | Median_CoF | Total_CoF |
NATURAL FORCE DAMAGE | HEAVY RAINS/FLOODS | 50 | 0.15 | 224.7 |
EXCAVATION DAMAGE | EXCAVATION DAMAGE BY THIRD PARTY | 92 | 0.43 | 210.9 |
EXCAVATION DAMAGE BY OPERATOR'S CONTRACTOR (SECOND PARTY) | 48 | 0.07 | 93.4 | |
PREVIOUS DAMAGE DUE TO EXCAVATION ACTIVITY | 17 | 0.69 | 48.4 | |
NATURAL FORCE DAMAGE | LIGHTNING | 26 | 0.04 | 26.4 |
EARTH MOVEMENT, NOT DUE TO HEAVY RAINS/FLOODS | 9 | 0.82 | 23.0 | |
OTHER NATURAL FORCE DAMAGE | 8 | 0.06 | 11.5 | |
TEMPERATURE | 88 | 0.04 | 10.9 | |
HIGH WINDS | 4 | 0.01 | 0.1 | |
EXCAVATION DAMAGE | EXCAVATION DAMAGE BY OPERATOR (FIRST PARTY) | 2 | 0.01 | 0.0 |
In the time-dependent incident count all incident causes that are considered time-independent are filtered out and a count by year is created.
The incident causes interpreted to indicate time-dependent failure mechanisms are CAUSE = “CORROSION” and CAUSE = “MATERIAL FAILURE OF PIPE OR WELD”
The trailing end of the distribution on the far left is likely because pipelines of that old of a vintage represent a very small percentage of the population. Note that this is NOT mileage-normalized.
In the time dependent incident by decade the install year was grouped within its decade (i.e. 1976 = 1970) and the incident count was derived from that the same way as the by-year plot of the same.
This section narrows the focus to the ‘consequence’ side of the Risk = Probability x Consequences equation.
Overall, the incident costs breakdown as follows. The P50 and P90 values can be used to help calibrate estimates emerging from QRA’s when relevant pipeline populations are being assessed.
This histogram will show the range and central tendency for Consequences of Failure (CoF). The CoF was calculated as the sum of the total inflation-adjusted cost, the value of a statistical life (VSL) of $10 million times the number of fatalities and the number of injuries multiplied by 1/3 of a VSL.
This plot is the median total reported annual costs for Significant Incidents.
Significant incidents are incidents that would of exceeded $50,000 in 1984 dollars, a release of 50 Bbls or more of Non-HVL or 5 Bbls of HVL commodity or an injury or fatality.
The median was used as a central tendency because costs for an incident can jump magnitudes of order from one to the next and the average can be skewed by one large incident. Only significant incidents are used to provide a constant reference year over year.
Differences in cost per incident based on commodity are seen here. The cost was normalized to the incident count since some commodities are much more common in the database than others. The Crude Oil average is heavily influenced by Marshall Michigan incident in 2010.
To answer ‘Is the cost of an incident by Commodity Changing over time?’, the average cost by commodity shows the costs over the last 11 years of data. In this plot the cost per incident is plotted versus year to determine if there is time-based trend.
The net barrels lost represents the total volume spilled minus what was recovered. The plot is broken out by the type of commodity being transported with a best fit line for each plot.
A possible hypothesis to be further examined is that the commodities that do not stay liquid at atmospheric pressure have lower cost/bbl relative to heavier products such as crude oil.
The database shows lengths of segments involved in incidents. This may relate to valve spacing, but inconsistencies in data input must be considered.
To determine possible correlations between incident cost and segment length, the following plot is generated.
This plot also differentiates HCA-impacting incidents (taken from database field = could_be_HCA)
These plots examine volumes lost relative to isolation length and commodity type. CO2, for instance, shows little correlation between segment length and volume released.
These plots show costs distributions relative to two aspects of HCA: did it reach an HCA? and was it previously identified as a segment that could affect and HCA (CAS)?.
If the segment was not identified as a CAS yet the spill still reached an HCA, this would indicate that there are errors in the spill modeling assumptions or process.
Additionally, the differences in cost distributions among the four cases plotted, can be seen.
This table summarizes the spills that reached an HCA based on the type of HCA affected.
name | Count | Pct._HCA_Incident | Percent_Incidents |
HIGH_POP_IND | 1,043 | 57.6% | 24.1% |
OTHER_POP_IND | 615 | 34% | 14.2% |
USA_DRINKING_IND | 556 | 30.7% | 12.8% |
USA_ECOLOGICAL_IND | 387 | 21.4% | 8.9% |
COMMERCIALLY_NAV_IND | 157 | 8.7% | 3.6% |
This table summarizes the spills that reached an HCA based on the type of HCA affected.
name | Count | Pct._HCA_Incident | Percent_Incidents |
OTHER_POP_IND | 3 | 60% | 4.84% |
USA_ECOLOGICAL_IND | 2 | 40% | 3.23% |
There is an an intuitive relationship between the time to shut-in a pipeline in an accident and the consequences of a failure.
This histogram examines the variability of times to shut in a line after identification of the accident. For pre 2010 data the time to shut-in was reported directly on the incident report form. From 2010 onward it was calculated by taking the elapsed time between the date and time for identification and the date/time to pipeline was shut-in.
This plot shows the elapsed time in hours between identification of the failure and when the line was shut in. This is on a log scale since consequence aspects are often exponentially related (ie, the values can change by orders of magnitudes between different quantiles).
Note that all commodity types are included here.
This table is the percentile of spills for pipelines in the right of way where the leak or rupture occurred in the Body or Seam of the pipe. The volumes are in barrels. Note there is over a 100 times difference between an average spill (50%) and one in the 90th percentile.
Percentile | Volume |
0% | 0.1 |
10% | 0.4 |
20% | 1.6 |
30% | 3.7 |
40% | 5.7 |
50% | 17.8 |
60% | 22.4 |
70% | 45.6 |
80% | 78.6 |
90% | 96.7 |
100% | 106.0 |
Release location type with count and percentage. Incidents that did not report a location type were filtered out. For all HL pipelines, 72% of all spills are contained on the operator property while only about 6% leave the operator’s property.
LOCATION_TYPE | Count | Percent of Incidents |
ORIGINATED ON OPERATOR-CONTROLLED PROPERTY, BUT THEN FLOWED OR MIGRATED OFF THE PROPERTY | 5 | 0.1% |
PIPELINE RIGHT-OF-WAY | 18 | 0.4% |
TOTALLY CONTAINED ON OPERATOR-CONTROLLED PROPERTY | 39 | 0.9% |
This plot is the Median volume in barrels released per ‘significant’ incident by year. Only significant incidents are used because that provides a constant reference year over year. Note that ‘significant’ has a specific definition.
Significant incidents are incidents that exceeded $50,000 in 1984 dollars, a release of 50 Bbls or more of Non-HVL or 5 Bbls of HVL commodity or an injury or fatality.
Volumes originate from the field named: UNINTENTIONAL_RELEASE_BBLS
The median was used as a central tendency indicator because costs for an incident can jump magnitudes of order from one to the next and the average can be skewed by one large incident.
In this graphic, both variability and central tendency are apparent.
The box portion (known as the inner quartile range, IQR) is the range from the second and third quartile with the median line though it. The lines extend out up to the last data point that is less than or equal to 1.5 times the IQR away and then data points that are further than that are outliers and are plotted as single points.
Note that the y-axis is on a log scale so each division is ten times the previous.
Volume values come from field = UNINTENTIONAL_RELEASE_BBLS and cost values from field = TOTAL_COST_CURRENT
This same data can also be viewed in light of HCA designations. The following two plots differ by whether the spill reached an HCA.
Similar to the previous graphs, here is the yearly costs of CO2 releases.
Examining how the incident was identified yields the table below (with blanks removed). This provides insights into how leak detection has been occuring in the past, for all HL’s.
The table also shows the range (p50 to p90) of the volumes associated with each identifier. This gives suggestions, albeit incomplete, into how large a release must be to trigger the identifier.
ACCIDENT_IDENTIFIER | Count | p50_vol | p90_vol | Percent_Incidents |
3rd Party That Caused It | 63 | 72 | 1,144 | 1.5% |
By Emergency Responder | 58 | 34 | 818 | 1.3% |
CONTROLLER | 104 | 30 | 1,042 | 2.4% |
Local Operating Personnel/Contractors | 2,286 | 2 | 100 | 52.8% |
OTHER | 194 | 5 | 263 | 4.5% |
PATROL | 231 | 5 | 203 | 5.3% |
Pressure/Leak Test | 14 | 7 | 654 | 0.3% |
Public | 329 | 10 | 540 | 7.6% |
SCADA | 312 | 14 | 2,654 | 7.2% |
Expanding on the previous table, this plot displays the full distribution of spill volume by accident identifier.
Possible conclusions include, for example, an observation that SCADA is more likely to identify larger spills while local personnel are more likely to identify smaller spills. The red vertical line is the median volume for the entire data set for comparison.
This displays the range of inflation adjusted Total Cost by incident cause. In this style graphic, both variability and central tendency are apparent.
The box portion (known as the inner quartile range, IQR) is the range from the second and third quartile with the median line though it. The lines extend out up to the last data point that is less than or equal to 1.5 times the IQR away and then data points that are further than that are outliers and are plotted as single points.
The x-axis is on a log scale so each division is ten times the previous, ie, note the wide range of costs associated with the incidents.
The following is a breakdown of the accident identifiers for PHMSA Hazardous Liquid incidents from 2010 to early 2021.
ACCIDENT_IDENTIFIER | Count |
Local Operating Personnel/Contractors | 2,286 |
738 | |
Public | 329 |
SCADA | 312 |
PATROL | 231 |
OTHER | 194 |
CONTROLLER | 104 |
3rd Party That Caused It | 63 |
By Emergency Responder | 58 |
Pressure/Leak Test | 14 |
Excluding blanks and limiting to incidents noted as “Pipeline Right-of-Way” and “Underground”, the largest identifier of incidents is still the Local Operating Personnel at about 64% of all incidents. Interestingly, Air and Ground Patrols accounted for about 6% of incident identifications and Controllers about half of that. Note Air and Ground Patrols are reported as separate categories in the PHMSA incident data but are combined into one category of “Patrol” for the sake of this summarization.
ACCIDENT_IDENTIFIER | Count |
Local Operating Personnel/Contractors | 223 |
Public | 180 |
PATROL | 95 |
SCADA | 74 |
OTHER | 46 |
3rd Party That Caused It | 42 |
By Emergency Responder | 37 |
CONTROLLER | 24 |
Pressure/Leak Test | 10 |
This is a summary of the accident identification method related to whether CPM was in place at the time of the incident.
SCADA and Controllers accounted for 16% of accidents identified when CPM was in place and about 13% overall. This is only for incidents reported as occurring on the pipeline right of way and excluding ‘aboveground’. These filters would therefore exclude, for example, tankage and piping incidents inside a pump station.
| Characteristic | Overall, N = 8521 | CPM | |
|---|---|---|---|
| NO, N = 3181 | YES, N = 5341 | ||
| ACCIDENT_IDENTIFIER | |||
| 3rd Party That Caused It | 48 (5.6%) | 17 (5.3%) | 31 (5.8%) |
| By Emergency Responder | 44 (5.2%) | 19 (6.0%) | 25 (4.7%) |
| CONTROLLER | 27 (3.2%) | 11 (3.5%) | 16 (3.0%) |
| Local Operating Personnel/Contractors | 275 (32%) | 94 (30%) | 181 (34%) |
| OTHER | 60 (7.0%) | 24 (7.5%) | 36 (6.7%) |
| PATROL | 102 (12%) | 46 (14%) | 56 (10%) |
| Pressure/Leak Test | 10 (1.2%) | 4 (1.3%) | 6 (1.1%) |
| Public | 204 (24%) | 88 (28%) | 116 (22%) |
| SCADA | 82 (9.6%) | 15 (4.7%) | 67 (13%) |
|
1
n (%)
|
|||
For the 4329 incidents in the database, the following tabulates the incident counts that involved segments monitored/controlled by SCADA and/or CPM.
SCADA_IN_PLACE_IND | CPM_IN_PLACE_IND | Count |
YES | NO | 1,405 |
YES | YES | 1,699 |
NO | NO | 463 |
NO | YES | 23 |
739 |
SCADA_IN_PLACE_IND | Count | P50_Volume | P90_Volume |
NO | 486 | 3 | 118 |
YES | 3,104 | 4 | 297 |
739 | 1 | 3 | |
(Bbls) | (Bbls) |
CPM_IN_PLACE_IND | Count | P50_Volume | P90_Volume |
NO | 1,868 | 4 | 201 |
YES | 1,722 | 3 | 328 |
739 | 1 | 3 |
This histogram is the consequence of failure differentiated by whether the spill was above or below the median spill volume. The size of the spill is not the sole indicator of the potential consequences. Several below median spills had consequences of 10 million dollars or more and a number of above median releases had consequences that were relatively low.
This summary table is a breakdown of the area types for incidents that there was SCADA or CPM in place. It
INCIDENT_AREA_TYPE | INCIDENT_AREA_SUBTYPE | Count | Percent |
ABOVEGROUND | IN OR SPANNING AN OPEN DITCH | 11 | 0.3% |
INSIDE OTHER ENCLOSED SPACE | 4 | 0% | |
OTHER | 13 | 0.2% | |
OVERHEAD CROSSING | 2 | 0% | |
TYPICAL ABOVEGROUND FACILITY PIPING OR APPURTENANCE | 87 | 1.8% | |
12 | 0% | ||
TANK, INCLUDING ATTACHED APPURTENANCES | 1 | 0% | |
TRANSITION AREA | OTHER | 2 | 1.7% |
SOIL/AIR INTERFACE | 1 | 0% | |
UNDERGROUND | EXPOSED DUE TO EXCAVATION | 93 | 4.5% |
IN UNDERGROUND ENCLOSED SPACE (E.G. VAULT) | 12 | 0.3% | |
OTHER | 26 | 3.7% | |
UNDER A BUILDING | 1 | 0% | |
UNDER PAVEMENT | 34 | 1.8% | |
UNDER SOIL | 568 | 85.6% | |
56 | 0% |
This is a summary of the median and total Consequences of failure by state and the total count since 2010 for CO2 pipelines.
Viewing all HL incident counts/costs by state paints a different picture. Higher counts are not a direct indication of the relative safety of those states. Texas, Louisiana and Oklahoma are the states with the largest energy production and consequently the highest incident counts. It can be seen that states that have incident counts that the median CoF is on par with most other states and the total CoF is in most cases is not inordinately higher than other states, indicating that even though the count is higher the consequences of those spills are less on a per incident basis than lower count states.
state | Median_CoF | Total_CoF | Count |
TX | 3.3 | 790.2 | 26 |
MS | 14.0 | 4,493.0 | 11 |
OK | 11.0 | 110.3 | 8 |
NM | 4.6 | 111.7 | 7 |
CO | 10.3 | 30.4 | 3 |
LA | 22.4 | 268.4 | 3 |
WY | 5.3 | 10.5 | 2 |
KS | 4.2 | 4.2 | 1 |
ND | 2.2 | 2.2 | 1 |
($M) | ($M) |
Company | Mileage |
AIR PRODUCTS & CHEMICALS INC | 13 |
AMPLIFY ENERGY OPERATING, LLC | 20 |
APACHE CORPORATION | 41 |
BRAVO PIPELINE COMPANY | 1,065 |
BREITBURN ENERGY CO2 | 177 |
CHEVRON PIPE LINE CO | 129 |
DAKOTA GASIFICATION COMPANY | 167 |
DAYLIGHT PETROLEUM LLC | 148 |
DENBURY GREEN PIPELINE-TEXAS, LLC | 132 |
DENBURY GULF COAST PIPELINES, LLC | 382 |
DENBURY ONSHORE, LLC | 419 |
DEVON ENERGY PRODUCTION CO. LP | 47 |
ELK OPERATING SERVICES LLC | 28 |
ENMARK ENERGY, INC | 10 |
EXXONMOBIL PRODUCTION COMPANY, A DIVISION OF EXXON MOBIL CORPORATION | 160 |
FDL OPERATING LLC | 155 |
GREENCORE PIPELINE COMPANY LLC | 235 |
GREENLEAF CO2 SOLUTIONS, LLC | 5 |
KINDER MORGAN CO2 CO. LLC | 1,310 |
LINDE GAS NORTH AMERICA, LLC | 7 |
OCCIDENTAL PERMIAN LTD | 6 |
PERDURE PETROLEUM, LLC | 272 |
PETRA NOVA CCS I LLC | 0 |
PETROSANTANDER (USA) INC. | 14 |
TREETOP MIDSTREAM SERVICES, LLC | 0 |
TRINITY PIPELINE GP LLC | 180 |
XTO ENERGY INC | 29 |