ARTICLE IN PRESS

Energy Policy 32 (2004) 811822

Daily oil losses in shipping crude oil: measuring crude oil loss rates in daily North Sea shipping operations
Rohit Bhatia, John Dinwoodie*
Shipping, Logistics and Law Group, Institute of Marine Studies, University of Plymouth, Drake Circus, Plymouth, Devon PL4 8AA, UK

Abstract Catastrophic crude oil spills excite media interest but operators usually manage greater losses from daily shipping operations single-handed. This paper invites energy policy makers to re-embrace daily oil losses. It categorises them, discusses physical and accounting measures and reports empirically crude oil loss patterns and factors inuencing them on some recent North Sea voyages. Losses varied with crude-type, measured by its vapour pressure, API gravity and viscosity, but evaporative loss in transit was low on these short voyages. Where sediment and water content inuenced losses or error rates are exceptional, measurement techniques should be revised. Varying losses across load and discharge ports infer revisions to tank and terminal designs. Empirical evidence revealed reduced losses in double hull vessels, supporting clingage theories, although insufcient pre-1985 or segregated ballast tank vessels were in employment to permit meaningful comparisons with newer ships. Calculating and applying vessel experience factors assists comparison and loss prediction. To minimise losses requires closer involvement of policy makers with operators. r 2003 Elsevier Science Ltd. All rights reserved.
Keywords: Crude oil transport

1. Introduction In 2001, oil transportation and shipping input an estimated 354 thousand tonnes (kt) of oil pollution into the global marine environment (Table 1). Propitiously, total global inputs declined from 5.5 million tonnes (Mt) in 1973 to 2.1 Mt in 1990 typically comprising runoff (3050%), natural seepage (10%) and atmospheric deposition (10%) (Patin, 2002; Alexanders, 2002). More routine daily oil losses accounting for 80% of discharges in marine crude oil transportation systems (Baird and Hayhoe, 1993) often eluded energy policy makers. Although technological advances such as double-hulled vessels required by the 1990 US Oil Pollution Act, followed very public and catastrophic failure of the Exxon Valdez (Skea, 1992), professional operators and investors often acted alone to implement them. Atmospheric emissions from ships transporting crude oil also largely evaded the Kyoto protocol and carbon tax debates (Godal and Holtsmark, 2001).

1.1. Oil inputs into marine environments Data in Table 1 reects the timing of periodic estimates of oil inputs into the marine environment rather than turning points in economic and shipping cycles (Stopford, 1997; Fearnresearch). However, inputs from shipping crude oil reect interactions between both environmental and economic inuences. Global sea borne crude oil movements peaked in 1976 at 16.8 trillion tonne kilometres (Ttkm), 45% of world shipping, substantiating longer hauls and output at treble 1964 levels, falling to 6.5 Ttkm in 1985, but rising to 12.8 Ttkm in 2001. Crude oil transport costs, typically around $4 per tonne (Stopford, 1997), have reached $7 (1979, 1990, late 1990s) and $11 (1970, 1973) inating spot prices by 50% in 1973, but only 2% in 1980 and 4% in 1990. Crude oil shipping pollution losses per tonne shipped cost 3, 24 and 2 cents in 1973, 1980, and 2001, respectively. Given falling real costs in an intensely competitive transport market (Gilje et al., 2002), operators still perceive competitive cost advantages through further reducing oil losses already down from 3% of transport costs in 1980 to 0.5% in the 1990s. Technical advances that reduced global physical oil input rates to the marine environment vefold over

*Corresponding author. Tel.: +44-1752-232446; fax: +44-1752232406. E-mail address: [email protected] (J. Dinwoodie).

0301-4215/04/$ - see front matter r 2003 Elsevier Science Ltd. All rights reserved. doi:10.1016/S0301-4215(03)00003-X

ARTICLE IN PRESS
812 R. Bhatia, J. Dinwoodie / Energy Policy 32 (2004) 811822 Table 1 Inputs of oil pollution into the marine environment and marine oil transportation 1973 Estimated global annual inputs arising from: Tanker accidents (kt) Operational discharges from tankers (kt) Bilge and fuel oil (non-tanker shipping, kt) Shipping inputs arising from tanker accidents (%) Sea borne movements of crude oil Annual total B(billion)t Crude oil as a % of world shipping movements Mean crude oil haul (kkm) Oil hauled lost in tanker accidents and operations (%) Estimated monetary values of crude oil shipments and losses Dubai spot crude price ($/t) Value of sea borne crude oil shipped. $B Estimated value of tanker pollution losses ($M) Value of crude oil shipped lost as tanker pollution (%) Value of crude oil lost per tonne shipped ($)
a b

1981 363 635 290 28 1.2 34 9.4 0.083 252b 302 301 0.097 0.244

1990 100 145 227 21 1.2 30 8.5 0.020 150 180 37 0.021 0.032

2001 104 250 29 1.6 30 8.0 0.016a 167 267 28 0.010 0.017

272 980 680 14 1.4 44 11.0 0.089 21b 29 36 0.124 0.026

Estimated per 1990 ratios. Arabian light.

three decades to 1 tonne in every 6000 transported, were probably stimulated by a tenfold rise in their economic value in the 1970s. When combined with shorter hauls the 2001 value of crude lost per tonne shipped was 30% below 1973 levels. If successful, controlled releases preventing up to 80% of global oil cargo vapour losses, currently approaching 1 Mt (LL, 2002), could save operators $ 167 M annually, at 2001 crude oil prices. These potential rewards induced Intertanko, representing tanker owners, alongside other tanker operators, to investigate measures to raise the accuracy of vapour loss measurements. Marine pollution legislation (MARPOL Annex VI) to reduce carbon dioxide, nitrogen oxide and sulphur dioxide emissions, although not yet fully ratied even in Europe, could promulgate differing grades of low-sulphur fuels for use in EU waters, ports, and open seas. Alternative proposals include a trading system for emissions and scrubbers to collect emissions. In the North Sea, recent land based crude oil discharges (2476 kt annually, UKOOA, 1999) probably exceed illegal (1560 kt) and legal (12 kt) discharges and spills from ships, mirroring proportions elsewhere. Other signicant sources include sewage sludge, industrial waste and dredging (422 kt), atmospheric deposition (715 kt) and offshore oil and gas (12 kt). Atmospheric emissions from offshore operations account for a relatively small proportion of total annual UK atmospheric emissions (UKOOA, 1999) including 10% (19 kt) of volatile organic compounds, 4% of carbon dioxide (1065 kt), 4% of nitrogen oxide (3 kt) and 3% of methane (3 kt). 1.2. Operator responses Even in a relatively low UK taxation environment, economic pressures ensured that operators adopted

innovations quickly on North Sea operations as oil majors cut costs to improve their structural and operational efciency (Mokia and Dinwoodie, 2002). Whilst rising crude oil prices raise the values of oil losses sustained and stimulate operators to investigate daily oil losses during daily marine transportation (TO, 2002), policy makers must also become involved. Historically low crude oil prices generated little interest in losses under 0.5%, but today oil majors host loss control departments investigating how the tonnage employed, type of crude, varying load and discharge ports and evaporation during cargo operations affect losses. If loss histories indict equipment failure, then documented claims against vessel owners ensue, and unacceptable losses at load ports, prompt reviews of custody transfer equipment and procedures. In articulating these concerns, this paper reports empirical oil losses in some recent North Sea operations and investigates technical factors inuencing them. Typically, operators measure crude oil at load and discharge ports in land tanks or using ow meters, and aboard ship, in tanks after loading net of any already on board quantity (OBQ), and before discharge net of oil remaining on board (ROB). Physical loss reects leakage, evaporation, or theft, and accounting losses include inaccurate land and ship tank calibration or measuring equipment and non-standard methods of calculation. Acquisition loss discrepancies between Bill of Lading (BOL) quantities and measurements aboard a loaded ship could reect poor measurement, understated water, old measurement tables, or vapour generation. Shipboard measurement variations between loading and discharge may differ on ballast and loaded voyages or reect OBQ differences. Discharge loss discrepancies between ship discharge and outturn certicate estimates are inuenced by crude oil washing

ARTICLE IN PRESS
R. Bhatia, J. Dinwoodie / Energy Policy 32 (2004) 811822 813

(IP, 1999a) vapour, pipelines, ROB-OBQ, poor measurement or overstated water (Intertanko, 1996). Differences between Net Outturn and Net BOL quantity averaged over several voyages have generally reduced (Smith, 1988). Should operators attribute crude oil loss on a particular voyage to normal errors in measuring temperature, the distance between a liquid surface and tank-top (ullage) or tank-bottom (innage), meter reading, an unlled pipeline, valve leakage, theft, or understated water (IP, 1999b; Miles and Jelffs, 1988)? Earlier work (Harrison, 1999) correlated measurement losses with crude type, load port and type of tonnage employed but few policies other than improved techniques can rectify measurement errors (Smith, 1988). This paper builds on earlier approaches by analysing loss patterns and factors affecting them in recent North Sea operations.

2. Sources of crude oil loss To an oil accountant, simply equating oil loss to starting stock plus added receipts less closing stock and outgoing despatches, fails to ag why or where losses arise or where improvements are required. Re-categorisation into real or physical and apparent or accounting loss is needed (Jones, 1985). Real loss either directly or as hydrocarbon vapour includes spills and leakage, vapour loss by evaporation during cargo operations and transportation, entrapment in ballast water in transportation and efuent water from oilwater separators, ROB after discharge, oil unaccounted for in transfer pipelines as line ll differences, or theft. Some losses, even when not directly measurable, are estimable (Jones, 1985). Often undetected, loss through a continuous leak of 3 mm width exceeds 30 m3/month but through improved pipeline protection, gross spillage accounted for only 0.0003% of total oil transport in Western Europe in 1983, 40% of which was recovered. 2.1. Evaporative loss Evaporation from tankers occurs during marine transportation whenever a volatile liquid comes into contact with air (or inert gas) not saturated with hydrocarbon vapour regardless of whether the vapour is displaced from the compartment and emitted into the surrounding atmosphere. These losses increase during non-inerted loading and discharging operations (API, 1992), at atmospheric pressure. The total evaporative cargo loss for vessels operated in a non-inerted condition can be estimated by summing losses during loading and discharge. Whilst these loss estimates do not apply to inerted very large crude carriers (VLCCs), other estimates do (Uhlin, 1985; Jones, 1985).

Evaporative cargo loss, generated during loading operations as fresh oil pumped into the tank comes into contact with air not saturated with hydrocarbon vapours, adds to arrival vapour present in the tank prior to loading. Total loading loss equates cargo volume loaded multiplied by the generated part of the total loading emission factor (API, 1992) which varies with the true vapour pressure of loaded crude oil, its average molecular weight and vapour growth factor and inversely with the average vapour temperature. As cargo discharges from tanks, unsaturated air or inert gas is drawn in, causing evaporative loss. This hydrocarbon vapour may later be emitted into the atmosphere when tanks are ballasted with seawater or loaded with oil. Discharge loss equates ballasting or loading emission factors grossed up by total cargo discharged, although some vapour will have been in tanks prior to discharge and cargo loss varies with discharge rates and trim remaining after discharge. Typical discharge losses explaining 8090% of total crude oil losses are 0.03% of volume for fully loaded tanker crude oil and 0.05% for lightered or short loaded tankers, varying with vapour pressure or ullage prior to discharge (API, 1992). Volumetric loss factors depend on cargo vapour pressure and the vapour molecular weight, not applicable to inerted VLCCs, where physical and evaporative loss (Uhlin, 1985; Jones, 1985) occurs during discharge, but most vapours are lost by breathing during the ballast voyage and next loading. Based on shipboard measurements (Uhlin, 1985) evaporative loss from a 250,000 tonne dwt VLCC voyage Persian Gulf to Northern Europe of 0.13% of cargo volume includes loading (0.033%), loaded voyage (0.015%) and discharging (0.079%) with a further 83 metric tonnes emitted during the ballast voyage. Losses should be much lower on short North Sea voyages. 2.2. Vapour emission from shore tankage Evaporation loss in shore tanks depends on their design. Fixed roof tanks produce a variable vapour space above the liquid surface varying with quantity in the tank, and include open vented tanks for non-volatile products or others with pressure vacuum relief valves capable of withstanding pressure variations. Vapour losses from xed roof tanks occur either when being lled or emptied, or standing where breathing arises from diurnal variation of atmospheric temperature and pressure due to vapour-air expansion and contraction. Vapour loss varies with the oils vapour pressure and degree of vapour saturation (Jones, 1985). Given an average saturation factor of 0.9, the displacement emission loss is 0.14% volume of liquid volume transferred. During discharge operations, air drawn into the tank expands on becoming saturated with hydrocarbon vapour causing displacement into the

ARTICLE IN PRESS
814 R. Bhatia, J. Dinwoodie / Energy Policy 32 (2004) 811822

atmosphere, with up to 50% variation in discharge loss factors dependant on the rate and manner of discharge. Daily breathing loss depends on the range of pressure vacuum relief valve settings, cargo vapour pressure, the size and shape of the tank, paint condition and colour and diurnal temperature range (Jones, 1985; Purvey and Robinson, 1985). Cumulative working losses usually exceed breathing losses in vapour tight tanks with dip and sampling hatches effectively closed and tted with pressure vacuum relief valves. Where breathing and withdrawal emissions interact during normal tank operations, afternoon liquid withdrawals when the static tank is expected to breathe out can reduce the internal pressure and hence evaporative loss. Floating roofs on liquid surfaces either as open, external roofs or internal blankets inside the xed roof tank shell eliminate direct vapour spaces removing displacement type losses. Breathing or storage losses occur through peripheral seals and standing losses depend on wind speed and eddying airow in rim vapour spaces. Eddying airow with vapour mounted resilient seals may be 5.5 times those for mechanical shoe seals (Purvey and Robinson, 1985). As the oating roof descends when liquid is withdrawn, evaporation from the exposed wetted tank surfaces may produce wetting losses of 0.0034% of throughput on rusted surfaces. With well-maintained seals, losses from oating roof tanks are 0.0020.02% of throughput (Jones, 1985). Maintaining a low oil temperature minimises evaporative loss from storage tanks by cooling prior to storage or using reective paint. Avoiding turbulence while lling and emptying tanks, using bottom loading and avoiding low tank liquid levels by using fewer tanks also assist (Jones, 1985). Through optimising tank pressure vacuum relief valve settings, annual breathing emissions may vary from 1.21% of tank capacity with negligible pressure vacuum range to 0.17% at 56 mbar. 2.3. Accounting loss Accounting loss is apparent rather than a real physical loss, but accurate volume measurement is needed to dene and identify sources of physical oil losses. Repeat readings of standard loading and discharge oil measurements, in land tanks or aboard ship, will vary between static readings in tanks and dynamic ow meter readings. Such measurement errors may be spurious, systematic or random (SGS, 2000). A second surveyor should reduce the risks of spurious errors involving mistakes where instruments are misread. Systematic errors, in-built and not readily detectable, occur where tanks or metering systems may be incorrectly calibrated reecting bias in the system. A wrongly calibrated tank will systematically err when delivering and oppositely when receiving, and being small, unnoticed errors may cumulate or be masked and

discarded as repeatable differences. Random errors, being bi-directional, may stem from surface ripples in liquids, or parallax errors in thermometers held with stems not perpendicular. Equipment checks and maintenance should minimise them. Measurement errors may relate to the object being measured, including a liquids viscosity and volatility (SGS, 2000) and unsuitability and working order of gaugers instruments. Human errors may reect training that varies between terminals, fatigue, or measurement environments where the weather, terminal and design may vary.

3. Oil measurement loss 3.1. Categories To assist investigation and comparison, measurement losses or differences are calculated for each stage of the voyage (Harrison, 2001). The net standard volume (NSV) of all petroleum liquids excluding sediment, water and free-water is corrected for observed temperature, American Petroleum Institute (API) gravity measure relative to a standard temperature, applicable pressure correction and meter factors. Total calculated volume (TCV) includes corrected volumes of petroleum liquids, sediment and water and free-water at observed temperatures and pressures. Historical comparisons of vessel TCV measurements, adjusted for OBQ and ROB, against shore TCV measurements build up a vessel experience factor (VEF), available for adjusting other measures if desired. Variations in water, as water loss % (WL%), NSV loss % (NSL%), and Gross or TCV loss % (TCL%) show the percentage difference between Outturn and BOL valuations. NSL% and TCL% are attributable to port or terminal design variations including tank design and volume measurement techniques and procedures. Positive correlation between NSL% and WL% (Smith, 1988) indicate that crude type inuences net losses. Crude oil type (Smith, 1988) vessel size, age, hull type and ballast tank arrangement (Harrison, 1999) inuence WL%. After loading, the percentage difference of the loaded vessels TCV over the BOL TCV denes the load difference % (LD%). Where calibrated ship tanks often overstate meter readings from shore tanks or loading, or OBQ is understated prior to loading, shore readings rarely exceed ship gures (Van Horne, 1985). Crude oil type and associated load port inuence the LD% (Smith, 1988). Losses during a voyage, the ship loss % (SL%) show differences between a vessel discharge TCV and loading TCV, reecting voyage evaporation and measurement errors particularly at offshore load ports (Van Horne, 1985), affected by crude type, load port or vessel-type (Smith, 1988).

ARTICLE IN PRESS
R. Bhatia, J. Dinwoodie / Energy Policy 32 (2004) 811822 815

Losses during discharge, the discharge difference % (DD%) show differences between vessel outturn and discharge TCV. Any bias in ships estimates at the load port is reversed on discharge and after correcting for load port bias, losses have been attributed to errors in gauging and temperature correction (Van Horne, 1985) or discharge port and crude type (Smith, 1988). A positive OBQROB difference % (OR%) between the OBQTCV and ROBTCV represents a loss attributable to cargo clinging to tank bulkheads, not measured as ROB, which settles to become OBQ at the next loading (Harrison, 1999), varying by vessel size, age and hull type. The load water % (LW%) represents the amount of water measured at loading as a percentage of the BOL NSV, dependant on the port via tank and pipeline design and water measurement techniques and type of crude oil affecting water content and difculty in measurement. 3.2. Effects of vessel type Earlier work (Harrison, 1999) related crude oil measurement losses including NSL%, WL%, SL%, OBQ%, ROB% and OR% to vessel type, load port and crude type. Vessels were categorised as pre or post-1985, long-range carrying o1.5 million barrels (Mbbls) or very large crude carriers (VLCCs) carrying >1.5 Mbbls, with double or single skin, double sides or double bottom hulls, with ballast tanks segregated (SBT), clean (CBT), mixed, or SBT non CBT. This is the vessel taxonomy adopted here. Pre-1985 VLCCs recorded high NSL% rates, gaining water between load and discharge ports with mixed SBT/CBT vessels recording high losses. On WL% measures, early VLCCs and non-SBT vessels performed badly, but given identical ballast tank arrangements, older VLCCs under-performed against long-range vessels. Although differences in SL% were statistically insignicant, VLCCs under-performed against longrange vessels as surprisingly did double hulls against single hulls. Differences in OR% correlated with average tank size on later SBT non-CBT vessels. Evidence that OBQ measurements correlated with tank capacity and ROB supported the view that increases in OR% reect cargo clinging to tank bulkheads at discharge, eluding measurement as ROB but later settling to be recorded as OBQ at the next loading. OR% increased in older VLCCs and newer long-range vessels but double hull vessels performed better on NSL%, WL% and OR% measures. 3.3. Load port effects Losses attributable to measurement errors may arise from land tank design, the accuracy of equipment

measuring temperature, oil-volume and free-water including automatic temperature gauges and ow meters. Some land tanks are relatively inaccessible for measuring, and tilted tanks, at tank farms adjacent to wharves or jetties, may have few access points for obtaining innage or ullage. Extraneous thermal effects may distort temperature readings taken through a single hatch close to the tank shell and horizontal temperature stratication errors occur where product temperatures differ from ambient temperatures (SGS, 2000). Measurement errors may arise from tank shell distortion as tanks ll. Initially, tank oors are calibrated in an empty tank, but depending on plate thickness, shape and the strength of foundations, the oor shape may distort on lling, usually within the rst two meters of liquid. Maintaining this minimum depth can avoid oor distortion errors (SGS, 2000). Head pressures act equally in all directions distorting lower plates. Radial distortion, highest within two meters of the tank oor, varies with the head and density of liquid, and is estimable if density is known. Movements of the dip plate and reference point occur during lling and emptying, with the dipping point on a oating roof tank dropping 5 mm with tanks 80% full at density 0.865. Tank size affects sample validity where standard 500 ml samples are unrepresentative in large tanks as are samples from stilling wells where safety reasons prevent personnel from standing on oating roofs (SGS, 2000). Coned-down V-shaped tank oors hamper free-water measurements, where it accumulates at the centre. Dip points on most tanks, often located one meter from the tank side, produce a below-datum quantity, implying that a tank bottom will hide water. Floating roof tanks may prevent dipping from the tank-centre and even if possible, further calibration tables incorporate corrections for conical oors. Such vessels register a high water content in outturn gures, actually present in the tank prior to lling, being discovered following a small increase in water level from the vessels discharge. Bulk liquid at oil terminals is measured either statically in tanks or dynamically in pipelines, for both volume and temperature. Measurement errors stem from human errors in using gauging equipment or marking tapes, mechanical errors in connecting lengths in brass weights on tape or damaged sensors on sonic ullage tape or mis-calibration of automatic level gauging equipment (SGS, 2000). Dynamic measurement using ow meters should enhance accuracy with fewer random errors that should eventually cancel out. If static and dynamic measurements fail to correlate, built-in systematic error should be suspected. Flow meter performance depends on meters having been calibrated at expected ow rates, appropriate to the viscosity, and not inuenced by liquid temperature, line pressure, wear and mechanical damage or ow conditions. Proper maintenance and calibration or meter proving at specied

ARTICLE IN PRESS
816 R. Bhatia, J. Dinwoodie / Energy Policy 32 (2004) 811822

intervals to obtain a reliable meter factor can minimise errors. 3.4. Variations by crude type Measurement errors and consequent losses depend on crude oil viscosity and volatility. Sampling is impeded where crude oils tend to stratify vertically into layers of differing density when standing undisturbed in land tanks, or in highly viscous oils exhibiting horizontal density stratication (SGS, 2000). Handling and storing volatile crude oils samples may be problematic and proper procedures must be adhered to prevent light crude oils evaporating. High viscosity oils complicate determination of the free-water in tanks, often recording higher losses. When a measuring tape smeared with water-nding paste enters a tank of viscous crude stored at pour point temperature, waxy components adhere to it making water detection impossible. With viscous crude, sludge forming on tank bottoms may complicate measuring free-water as the horizontal oil-water interface is unlikely to extend across the tank, generating spurious volumes. Sludge may accumulate on dip plates, around the stilling wells in oating roof tanks. Sometimes, large volumes of free-water measured in land tanks prior to ship discharge apparently disappear during discharge, being displaced along with sludge to locations inaccessible to measurement. Such pockets of free-water on tank oors also distort the oilwater interface seen in low viscosity oils. High viscosity crude generates more water in suspension or emulsions, taking longer to settle as free-water on the tank bottom, and complicates pumping, taking longer to travel to the suction bell-mouth in the after part of the tank. Volatile or gaseous crude reduces pumping efciency, resulting in higher ROB. Being temperature-sensitive, in lighter oils with high coefcients of thermal expansion, small temperature measurement errors create signicant discrepancies in volume measurements (Jones, 1985).

vessel age, size and hull type (Van Horne, 1985). SL% should vary with voyage evaporation and on- or offshore load ports (Van Horne, 1985) and be greater in VLCCs but reduce in single hull vessels (Harrison, 1999). DD% should vary with discharge port and crude type (Smith, 1988) and OR% should be high in pre-1985 VLCCs but reduced in double hull vessels (Harrison, 1999). Finally WL% should vary with crude type (Smith, 1988) being greatest in pre-1985 VLCCs but reduced in double hull vessels (Harrison, 1999). 4.2. The database STASCO provided voyage records. Vessel data included deadweight (dwt), cargo capacity (bbls), age, hull type, and ballast tank conguration obtained from Clarksons vessel database and STASCO and the Institute of Petroleum (IP) provided crude oil data. Unstructured interviews with oil loss control personnel claried operational issues before establishing a database of voyages by STASCO chartered vessels from March 1999 to March 2001. This showed vessel name, issue date of BOL, load and discharge port, VEF, Gross and Net BOL, Gross and Net Outturn, TCV after loading and arrival both corrected and uncorrected for VEF, OBQ, ROB, sediment and water, BOL and Outturn. Oil losses were calculated, and the crude Reid vapour pressure (RVP) and viscosity, the ships year of building, cargo capacity (bbls), hull type and ballast tank arrangement added. Missing values were recorded for null entries in the original voyage database. Voyages were excluded for OR% where oil was carried as ballast; for crude oil types involving only one voyage, and ports reporting under 6 voyages. The remaining database included 19 load and 6 discharge ports involving 122 vessels moving 426 Mbbls representing 20 crude types. Crude oils carried were representative of North Sea production including 37 sampled voyages with heavy crude (API 2030), 392 with medium crude (API 3040) and 149 with light crude (API >40) and similar RVP frequencies, with 208 o8.00, 195 from 8.00 to 10.04, and 175 > 10.05. Most vessels (98%) had SBTs, but 13% included both SBT and CBT. Double hull vessels predominated (51%) followed by single hulls (31%), double bottom (12%) and a few with double sides. With only 2 VLCCs, the trend is towards employing relatively new, SBT, double hull, long-range vessels. Load ports were less concentrated with one accounting for 19% of loads, two for 10% and a further 4 exceeding 5% compared with two discharge ports accounting for 30% each of voyages. 4.3. Statistical considerations Although several key variables approximated normal distributions, descriptive summaries of both the

4. Methodology and data 4.1. Hypotheses This study aimed to identify the loss patterns and factors inuencing them when using large vessels to transport crude oil in North Sea operations. Prior hypotheses were tested that NSL% is positively correlated with WL%, varying with load and discharge port and crude type shipped (Smith, 1988) and is greatest in pre-1985 VLCCs, is not inuenced by SBT and reduced in double hull vessels (Harrison, 1999). TCL% should also vary with load and discharge port and crude type shipped. It is hypothesised that LD% varies with crude type and load port (Smith, 1988) and

ARTICLE IN PRESS
R. Bhatia, J. Dinwoodie / Energy Policy 32 (2004) 811822 817

dependent and independent variables (Table 2) revealed examples of both positive and negative skewness exceeding unity. Subsequent analysis employed mainly non-parametric statistics affording reduced inferential power, including independent multi-sample Kruskal Wallis Chi-square (KW Chi.sq) and bi-sample Mann Witney (MW) U test comparisons of central tendency and Spearmans rho correlation coefcients. Where appropriate, standard deviations (Stdev), mean ranks (MR), sample size (n), degrees of freedom (df) and the signicance (Sig.) of inferences are reported. Predictive regression equations report simple Pearson productmoment multiple correlation coefcients (r), the intercept coefcient (Bo), with t values recorded under each signicant variable coefcient. Vessel load and discharge gures generally exceed terminal gures, with a positive mean LD% (0.08%) and negative mean DD% (0.29%) indicating a VEF>1. A neutral SL% (0.003%) indicates consistent vessel measurements at the load and discharge port and with negligible evaporation during short North Sea voyages, is attributable to measurement errors. A low mean OR% (0.01%) reects clingage on tank bulkheads increasing OBQ at the load port and mean WL% (0.03%) represents undetected or undeclared water either from the cargo or the vessels tanks. Statistical analysis began by comparing intercorrelations between loss measures to test for possible redundancy. Next, the effects on loss measures of known hypotheses relating to how crude oil type, load and discharge port, and vessel characteristics affected each measure were tested. Finally, factors affecting loss rates on all voyages between particular load and discharge ports for particular vessel hull types were analysed and calibrated to develop specic predictive models of loss rates. Statistical comparisons of losses by vessel-age were limited to broad vessel classes because of insufcient voyages by pre-1985 VLCCs to allow meaningful comparison with newer tonnage.

5. Factors affecting loss measures 5.1. Volume loss measures The water content of crude affected net losses, with a positive correlation between WL% and NSL% (Table 3). Further high positive correlation coefcients linking NSL% with TCL% and DD% with LD% were anticipated and imply some redundancy between measures, although all are reported here. Crude type inuenced NSL% and TCL%, with both negatively correlated with RVP, but TCL% alone correlated positively with viscosity. Neither measure correlated with API. Negative correlation with RVP implies higher losses in less volatile crude, probably reecting errors in dynamic ow meter measurements. Both NSL% and TCL% revealed signicant differences amongst both individual load and discharge ports (Table 4), with load ports associated with medium crude showing greater mean losses. Offshore load ports generated higher TCV losses probably attributable to different tank and pipeline design and measurement techniques. Contrary to expectations (Harrison, 1999), net losses did not vary signicantly with vessel type. NSL% differences between pre- and post-1985 vessels were statistically insignicant, as were reduced losses in the few non-SBT vessels sampled and skin type variations for both measures reecting few voyages by double-sided vessels. 5.2. Load and discharge differences Crude oil type inuenced both LD% and DD%, given their inverse relationship. LD% correlated negatively and DD% positively with RVP, a relationship reversed for viscosity. API inuenced LD% negatively, but not DD%. Reduced LD% losses in lighter crude may signify evaporation during loading operations. Greater LD% losses at onshore load ports infer greater variations between land and ship tank transfers than offshore ports often involving ship to ship transfers. As expected, DD% varied across discharge ports, reecting variable terminal designs for tanks and pipelines and volume and temperature measurement techniques. Both measures recorded increased losses on pre-1985 vessels although cargo capacity inuenced LD%, increasing on smaller ships. Double hull vessels recorded signicantly lower losses than single hulls for both measures. Overall, LD% was inuenced by crude type and vessel age, hull and size indicating increased losses in older ships with over calibrated tanks and difculties in measuring OBQ prior to loading (Van Horne, 1985). Increased DD% losses arose in older single skin vessels where more viscous crude increases ROB attributable to clingage, reduced in double hulls with ush tanks where fewer structural supports reduce ROB and obstruction

Table 2 Descriptive statistics of key variables Variable NSL% TCL% LD% SL% DD% OR% WL% LW% API RVP Viscosity Capacity bbls Ship age Valid 530 530 546 524 503 541 528 570 578 578 578 544 578 Mean 0.274 0.237 0.077 0.003 0.293 0.010 0.026 0.177 38.23 8.39 8.50 783.7 12.79 Stdev 0.33 0.31 0.39 0.17 0.35 0.03 0.11 0.16 3.95 2.44 4.85 155.6 7.89 Skewness 0.35 0.07 0.67 0.08 1.07 2.38 0.39 1.77 1.26 1.32 1.82 0.69 0.37 Kurtosis 4.23 4.24 1.51 2.51 2.23 15.33 3.29 3.66 3.91 1.53 2.88 5.60 0.86

ARTICLE IN PRESS
818 R. Bhatia, J. Dinwoodie / Energy Policy 32 (2004) 811822 Table 3 Spearmans rho correlations between loss measures and crude oil type TCL% LD% SL% DD% OR% WL% LW% Measure API RPI Viscosity Measure
a b

Rho n Rho n Rho n Rho n Rho n Rho n Rho n n n Rho Rho Rho

0.87b 530 0.31b 502 0.04 520 0.22b 503 0.19b 497 0.25b 525 0.07 523 NSL% 530 0.01 0.18b 0.00 NSL%

0.37b 502 0.03 520 0.25b 503 0.22b 497 0.14b 525 0.02 523 TCL% 530 0.05 0.26b 0.11b TCL%

0.26b 500 0.63b 499 0.25b 532 0.06 497 0.14b 538 LD% 546 0.09a 0.27b 0.21b LD%

0.12b 498 0.00 496 0.04 518 0.10a 517 SL% 524 0.12b 0.07 0.04 SL%

0.09a 487 0.05 498 0.18b 496 DD% 503 0.04 0.16b 0.23b DD%

0.02 494 0.03 533 OR% 541 0.07 0.14b 0.21b OR%

0.24b 523 WL% 528 0.00 0.12b 0.16b WL%

Correlation is signicant at the 0.05 level (2-tailed). Correlation is signicant at the 0.01 level (2-tailed).

to oil ow on tank bottoms. Unmeasured vapour emissions during discharge, crude oil washing, evaporative loss during the ballast voyage and next loading and terminal design effects may also inuence DD%.

5.3. Other measures Variations in SL%, correlated positively with crude API, were probably attributable to measurement errors, but were unrelated to RVP or viscosity implying that evaporation at sea is minimal on short North Sea voyages. Statistical differences across load ports, although not signicant here, indicated greater losses on the few heavy crude voyages sampled. At offshore ports, dynamic vessel movements usually increase measurement errors but differences were not signicant here. Losses were signicantly reduced in double hull rather than single skin vessels where improved designs facilitate accurate measurement. Pre-1985 vessels recorded signicantly higher OR% losses, as did single skin vessels compared with double hull vessels. Losses inversely correlated with vessel cargo capacity imply higher losses in smaller vessels. Differences in OR% varied negatively with crude RVP and positively with viscosity, but not API. Evidence associating higher OBQ with less volatile and more viscous crude oils supports clingage theories whereby viscous crude tends to stick to structural girders, longitudinal supports and tanks found in older single skin or smaller vessels. Random measurement errors of OBQ and ROB, trim and weather conditions especially at offshore ports

may explain low correlation coefcients and account for unexplained variations. In these studies, WL% correlated positively with RVP and negatively with viscosity, indicating higher water losses in less viscous, more volatile crude. Although mean WL% was reduced in double hull, SBT and post1985 vessels, differences were not statistically signicant, in broad agreement with but insufcient to verify Harrisons (1999) ndings. Losses probably reect difculties in measuring free-water in high viscosity crude and its tendency to from suspensions and emulsions. Terminal design data potentially affecting tank design including access for measurement and sampling, size, shape, calibration and water determination techniques was unavailable.

6. Model building to predict loss rates Aggregate models to predict losses, taken over multiple routes and hull-types, failed to generate signicant results. However, factors affecting loss rates on voyages between ve sets of load and discharge ports for single skin (AE) and two for double-hulled vessels (F, G) were successfully modelled and calibrated. Although this strategy eliminated these effects, it included only 16 20 voyages between each port pair. Initial correlation matrices revealed API as the best descriptor of crude type, and stepwise regression procedures generated identical models even when RVP and viscosity were excluded. Of the vessel descriptors, limited samples precluded analysis of ballast tank congurations, but

ARTICLE IN PRESS
R. Bhatia, J. Dinwoodie / Energy Policy 32 (2004) 811822 Table 4 Variations in loss measures by port and vessel type Measure Load port NSL% Variable Crude Heavy Medium Light Heavy Medium Light Heavy Medium Light Onshore Offshore Onshore Offshore Onshore Offshore 6 ports 6 ports 6 ports Statistic KW Chi-sq 12.3a df 2 n 36 353 141 36 353 141 35 351 138 182 347 222 323 179 344 df 5 5 5 530 530 503 (SS), double sides (DS) 65 279 156 6 DH SS DH Mean rank 345 254 275 395 247 277 217 263 273 332 230 293 259 255 265 Mean loss% 0.16 0.29 0.26 0.05 0.27 0.22 0.08 0.00 0.01 0.14 0.29 0.13 0.04 0.01 0.00 0.27 0.24 0.29 819

TCL%

31.7a

SL%

3.9

Load port TCL% LD% SL% Discharge port NSL% TCL% DD%

MWU 19,349a 56,011b 29,592 KW Chi-sq 42.1a 89.0a 54.6a

Hull: double bottom (DB), double hull (DH), single skin NSL% DB KW Chi-sq df DH 6.28 3 SS DS MWU DB/SS 4371 DH/SS 19,734 TCL% DH/SS 19,740 LD% DH/SS 10,466a SL% DH/SS 18,378b DD% DH/SS 8152a OR% DH/SS 16,203a WL% DH/SS 21,866 Vessel age pre-1985 O; post-1985 N NSL% O=N 30,051 LD% O=N 18,854a DD% O=N 15,992a OR% O=N 24,953a WL% O=N 29,946 Ballast tanks. Non-SBT (N), SBT (S) NSL% N=S 1720 WL% N=S 1661 Cargo capacity bbls LD% OR% WL%
a b

236 246 270 350 SS DH

0.30 0.29 0.26 0.11 SS

279 290 275 277 287 278 O 167 162 142 161 166 N 8 8

156 153 154 132 253 158 N 363 384 361 380 362 S 521 519

211 182 205 242 200 219 O 267 349 184 306 264 N 219 212

231 299 233 128 257 218 N 265 242 279 256 265 S 266 265

0.26 0.08 0.00 0.18 0.01 0.02 O 0.29 0.28 0.45 0.02 0.03 N 0.08

0.22 0.32 0.03 0.53 0.03 0.03 N 0.27 0.01 0.23 0.01 0.02 S 0.03

Rho 0.11b 0.16a 0.04 513 509 496 0.01

Statistic is signicant at the 0.01 level (2-tailed). Statistic is signicant at the 0.05 level (2-tailed).

vessel age, cargo capacity and VEF were included. Each loss measure was regressed against these predictors (Table 5). Voyages involving double hull vessels generated no signicant results for route G and predictions

only for LD% and DD%, both largely explained by variations in VEF. Predictive equations for single hulled vessel voyages were also most successful for DD%, with performance on all ve routes mainly inuenced by VEF, supplemented by age

ARTICLE IN PRESS
820 R. Bhatia, J. Dinwoodie / Energy Policy 32 (2004) 811822 Table 5 Loss prediction regressions on particular load and discharge port routes Measure n r Bo VEF Cap API Age 0.0217 (2.33) 0.144 (2.49) 0.0062 (3.15) 0.113 (3.12) 0.0129 (2.24) 0.0086 (3.86) 78.247 (2.14)

Voyages n by single hull vessels by measure (e.g. NSL%) and route (e.g. B) NSL% B 14 0.57a 0.155 (2.76)b NSL% C 18 0.53a 5.932 (2.47) NSL% D 16 0.64c 4.924 (2.90) NSL% E 17 0.63c 4.818 (3.40) TCL% B 14 0.56a 0.107 (3.08) TCL% D 16 0.72c 6.942 (3.64) LD% A 20 0.64a 80.431 0.0023 (2.20) (2.79) LD% B 14 0.53a 97.468 97.533 (2.36) (2.36) LD% C 18 0.80c 61.589 61.570 (5.26) (5.26) LD% E 17 0.88c 98.37 112.652 (6.15) (6.24) SL% A 20 0.63c 2.011 (3.45) SL% D 16 0.59a 0.0189 (0.39) DD% A 20 0.62c 62.808 62.893 (3.23) (3.22) DD% B 14 0.66a 79.147 79.268 (2.77) (2.77) DD% C 18 0.78c 38.072 37.871 (2.61) (2.59) DD% D 16 0.75c 94.634 94.968 (4.26) (4.28) DD% E 17 0.92c 88.043 94.047 (7.04) (7.62) OR% A 20 0.56a 0.0195 (2.94) OR% E 17 0.62c 0.0041 (3.11) WL% A 20 0.54a 0.272 (3.85) Double-hulled vessels ROUTE F No signicant relationship (n 24) LD% G 17 0.98c 99.350 99.338 (18.11) (18.13) DD% G 17 0.92c 85.244 85.357 (7.42) (7.44)
a

0.017 (3.70) 0.0022 (3.38)

0295 (2.48)

0.0154 (2.76)

0.0367 (2.50)

0.153 3.10

0.0278 (2.94) 0.0017 (2.81) 0.0003 (3.04) 0.0164 (2.63)

Statistic is signicant at the 0.05 level (2-tailed). Values of t-statistics in parentheses. c Statistic is signicant at the 0.01 level (2-tailed).
b

and API, followed by LD%, with 4 routes involving VEF, supplemented by capacity and age. Although signicant models predicting NSL% were obtained on 4 routes, predictors were unstable including API twice, and age and capacity once each, but on routes inuenced solely by vessel age and capacity, TCL% was also determined by the same variables.

Predicting OR% was possible on 2 routes related primarily to vessel age, as was WL%, predictable only on 1 route. Taken overall, the most difcult loss measures to predict were WL% followed by OR%, SL% and TCL%. VEF provided a good predictor of LD% and DD% for most instances of voyages involving both single and double hulls.

ARTICLE IN PRESS
R. Bhatia, J. Dinwoodie / Energy Policy 32 (2004) 811822 821

7. Conclusion Traditionally, energy policy makers spurned interest in daily marine crude oil transportation systems and even crises gained only eeting attention. Today, oil majors incorporate specialist crude oil loss control departments to reduce losses through understanding loss patterns and trends by analysing the factors and variables affecting them. Policy makers must respond by promoting good practise. Almost universally, North Sea crude transportation operations now employ doublehull segregated ballast tank long-range vessels built post-1985, precluding meaningful performance comparisons with older vessels. However, when the vessel experience factor is calculated and applied to all loading and discharge operations, comparisons with Bill of Lading gures and applications at offshore ports where shore data is unobtainable become possible. Although the ndings reported, based on empirical operating data drawn mainly from one company, are indicative, they urgently invite more objective and reliable comparisons with other companies and routes. Oil loss measures in this study correlated with the type of crude characterised by its vapour pressure, API and viscosity. On short North Sea operations, ship loss was minimised. Vapour losses from shore tanks prompt reviews of vapour loss control procedures in particular terminals. Onboard, they necessitate thorough guidelines and crew training. Crude viscosity correlated with discharge difference and water loss, highlighting the effects of sediment and water measurement techniques. Procedures need updating in ports where they are decient. Losses varied across load and discharge ports, reecting varying port procedures and practices. A policy of management by exception is required to review extremes of terminal design including shore tank design, and measurement techniques and equipment for sediment and water, temperature, static and dynamic volume and tank calibration. Additional data related to terminal design and onboard operational procedures may help to explain variations in oil losses. Terminal design includes shore tank design incorporating access, shape, size and type of roof and tank calibration. Tank sediment and water measurement techniques comprise online or manual tank sampling. Temperature measurement issues cover the procedures employed, equipment accuracy and calibration, static and dynamic volume measurement and vapour loss control mechanisms. Onboard operational procedures include vapour loss control during voyage and cargo operations, the extent and procedures of crude oil washing and volume and temperature measurement. Using measurement system analysis will involve assessing or validating the measurement system to detect errors attributable to automated data

collection via ow meter readings and tank temperature readouts. If a substandard measurement system is identied, differences may reect inaccurate measurements rather than real losses. How can energy policy makers assist? Initially, by promoting research into crude oil loss during marine transportation, standardising loss measures and establishing more robust understanding of factors affecting losses. They will assist oil majors to promote and disseminate good practise in oil loss control procedures globally, amongst operators. Medium term, standard setting and promoting specialist operator training will ensue. Finally, as regulatory conventions are accepted and ratied, research will explore further technological innovations to eliminate other potential sources of emissions encountered in meeting world demand for oil.

Acknowledgements Thanks are due to Shell Trading and Shipping Company Limited (STASCO) for granting access to their databases and offering technical assistance.

References
Alexanders, 2002. Oil enters oceans around US mostly as a result of human activities. Gas and Oil Connections, News & Trends, North America, Vol. 7, no. 2, June 13. Via Internet: http://www. gasandoil.com/goc/news/ntn22479.htm API, 1992. Atmospheric Hydrocarbon Emissions from Marine Vessel Transfer Operations, API 2514A. American Petroleum Institute, London. Baird, S., Hayhoe, D., 1993. Oil Spills. Energy Fact Sheet. Energy Educators of Ontario. Via Internet: http://www.iclei.org/efacts/ oilspill.htm Fearnresearch. Annual Review. Oslo, Norway. Gilje, T.T., Dinwoodie, J., Challacombe, J.A., 2002. Crude carrier consolidation and capital cost. International Journal of Maritime Economics 4 (1), 3554. Godal, O., Holtsmark, B., 2001. Greenhouse gas taxation and the distribution of costs and benets: the case of Norway. Energy Policy 29, 653662. Harrison, P., 1999. The Inuence of Vessel Characteristics on Crude Oil Measurement Losses Report No. 2. Melverley Consultants, London. Harrison, P., 2001. Marine Oil Transportation Database. Melverley Consultants, London. IP, 1999a. Petroleum Measurement Manual: Guidelines for Crude Oil Washing of Ships Tanks and the Heating of Crude Oil being Transported by Sea. Institute of Petroleum, London. IP, 1999b. Standard Methods for Analysis and Testing of Petroleum and Related Products and British Standard 2000 Parts. John Wiley, London. Intertanko, 1996. Oil Cargo Losses and Problems with Measurement 2nd Edition. International Association of Independent Tanker Owners, London. Jones, P., 1985. Oil loss, measurement and control. A paper presented to the Chinese Petroleum Society, China, July 229.

ARTICLE IN PRESS
822 R. Bhatia, J. Dinwoodie / Energy Policy 32 (2004) 811822 Smith, D.L., 1988. Crude oil loss analysis. In: Jelffs, P.A.M. (Ed.), The Second Oil Loss Control Conference, Institute of Petroleum, London, October 1987. John Wiley, Chichester, pp. 6 and 24. Stopford, M., 1997. Maritime Economics, 2nd Edition. Routledge, London. TO, 2002. Pressure cooker. Tanker Operator, May, Vol. 1, no. 7, pp. 1516. Uhlin, R.C., 1985. Physical loss of cargo from crude oil tankers. In: Inkley, F.A. (Ed.), Oil Loss Control in the Petroleum Industry, Institute of Petroleum, London, October 1984. John Wiley, Chichester, pp. 143151. UKOOA, 1999. Environment Report. Industry Issues: Sustainability. United Kingdom Offshore Operators Association. Via Internet: http://www.ukoa.co.uk/issues/sustainability/environment.htm. Van Horne, E.K., 1985. Crude oil voyage variation analysis. In: Inkley, F.A. (Ed.), Oil Loss Control in the Petroleum Industry, Institute of Petroleum, London, October 1984. John Wiley, Chichester, pp. 8485. LL, 2002. Campaign to reduce vapour emissions. Lloyds List. 5 September, Lloyds of London, p. 14. Miles, J., Jelffs, P.A.M., 1988. Computer aided loss investigation and monitoring. In: Jelffs, P.A.M. (Ed.), The Second Oil Loss Control Conference, Institute of Petroleum, London, October 1987. John Wiley, Chichester. Mokia, Z., Dinwoodie, J., 2002. Spatial aspects of tanker lay-times. Journal of Transport Geography 10 (1), 3949. Patin, S., 2002. Oil Pollution of the Sea. Via Internet: http:// www.offshore-environment.com/oilpollution.html Purvey, R., Robinson, E., 1985. Crude oil production, acquisition and transportation. In: Inkley, F.A. (Ed.), Oil Loss Control in the Petroleum Industry, Institute of Petroleum, London, October 1984. John Wiley, Chichester, p. 64. SGS, 2000. Custody Transfer of Crude Oil and Loss Control Issues. SGS Redwood Services, Ellesmere Port UK, pp. 575-595. Skea, J., 1992. Environmental issues facing the oil industry. Energy Policy 20 (10), 950958.