
                                                                 Peer Review of
                                                       EPA Refinery Cost Model:
                                                      Gasoline Sulfur Reduction
                                                                               
                                                                   Final Report
                                                               December 19 2011
                                                                               
                                                                   Prepared for
                                                                 Lester Wyborny
                                                                     Russ Smith
                                           U.S. Environmental Protection Agency
                                Office of Transportation and Air Quality (OTAQ)
                                                     1200 Pennsylvania Ave., NW
                                                           Washington, DC 20460
                                                                               
                                                EPA Contract Number EP-C-08-008
                                          RTI Sub-contract Number 4-312-0211577
                                                                               
                                                                    Prepared by
                                       Martin R. Tallett & Daniel N. Dunbar
                                               EnSys Energy & Systems, Inc.
                                                      1775 Massachusetts Avenue
                                                            Lexington, MA 02420

 

Table of Contents
1	Introduction	1
2	Summary	2
2.1	Adequacy of Current Model Formulation Data & Results	2
2.2	Proposed Recommendations for Changes and Their Impacts on Estimated Costs	2
3	Regulatory Context	3
4	Commentary & Findings	3
4.1	Model Version Provided to EnSys	3
4.2	EPA Methodology/Modeling Approach as a Means to Assess Costs of Gasoline Sulfur Reduction	4
4.2.1	Summary of EPA Methodology	4
4.2.2	Estimation of Blendstock Volumes by Refinery	4
4.2.3	Sulfur Levels Assigned to Blendstocks	5
4.2.4	FCC Gasoline Sulfur Level Back-calculation	6
4.3	Key Determinants of Costs in Model	6
4.3.1	FCC Gasoline Volume & Sulfur Level	6
4.3.2	Assumed Target Sulfur Levels	6
4.3.3	Vendor Cost Data & Transposition into Model	7
4.3.3.1	Cost Bases Requested by EPA	7
4.3.3.2	Vendor Processing Schemes	7
4.3.3.3	Vendor 1	8
4.3.3.4	Vendor 2	9
4.3.3.5	Vendor 3	10
4.3.3.6	Other Vendors & Average Costs	10
4.3.3.7	All Vendor Capital Costs	11
4.3.3.8	Capital Costs (ISBL and Total)	11
4.3.3.9	Octane Loss	11
4.3.4	Other Key Cost Parameters	11
4.3.4.1	Capital Cost Power Factor	11
4.3.4.2	Natural Gas, Electricity & Steam Prices	12
4.3.4.2.1	Natural Gas	12
4.3.4.2.2	Electricity	13
4.3.4.2.3	Steam	13
4.3.4.3	Hydrogen Cost	13
4.3.4.4	Cost of Octane Loss	14
4.3.4.5	E10/E15	15
4.4	Additional Factors Influencing Costs	16
4.4.1	LSR Sulfur Reduction	16
4.4.2	Other	16
4.5	Model Results, Proposed Premise Adjustments and Their Impacts	16
4.5.1	Results from Current Cost Model	16
4.6	Detailed Model Integrity Findings/Issues	17
4.6.1	Approach Employed by EnSys	18
4.6.2	General Comments on Model Structure	18
4.6.3	Specific Data & Equations Integrity	19
4.6.3.1	Capital, Fixed & Variable Operating Cost Formulae	19
4.7	Potential for Model Improvement	19
4.8	Utility for Forward Looking Regulatory Analyses	19








Introduction

The EPA Office of Transportation and Air Quality (OTAQ) contracted EnSys Energy to undertake one of two "peer reviews" of the EPA's Refinery-by-Refinery Cost Model (the EPA Refinery Cost Model).    The EPA Model was originally developed by EPA OTAQ for benzene level analysis using extensive MathPro fuels blending and LP data.   The version of the Model delivered to EnSys for review had been extended and "re-purposed" by EPA OTAQ to estimate costs for reducing sulfur in U.S. gasoline from the present 30 ppm limit to 10 and 5 ppm.   As for the original benzene-oriented model, the current Model calculates costs for each U.S. refinery that produces finished gasoline.   
EnSys' review focused on the Model methodology, the associated data sources and data development process and the specific data and equations applied in the Model.  Our remit was to consider solely the adequacy of the model to capture costs for U.S. gasoline sulfur reduction to 10-5 ppm from current levels.   
The Sections below comprise our review.  
   * Section ‎2 Summary summarizes our review and findings
   * Section ‎3 Regulatory Context provides an outline of the regulatory context for the use of the EPA Model, and hence for the peer review
   * Section ‎4 Commentary & Findings comprises the core of our review
         o Section 4.1 summarizes the specific Model version supplied to EnSys for review
         o Section 4.2 first summarizes EnSys' understanding of the EPA methodology as understood by EnSys.  Our intent here was to "track" the key data items and decisions in the methodology as a key first step in reviewing it.   The balance of Section 4.2 then reviews the key elements in the EPA Model methodology
         o Section 4.3 focuses in on the factors which we considered as key to EPA's determination of sulfur reduction costs.   The section thus covers volume and sulfur level for FCC gasoline, the (adequacy of the) vendor data obtained and used by EPA OTAQ, the transposition of that data into the Model and the effect of other key assumptions affecting costs.  
         o Section 4.4 comments on additional elements in the potential costs of meeting 10-5 ppm standards, notably LSR sulfur reduction
         o Section 4.5 sets out EnSys' recommendations for adjustments to the EPA cost assumptions and our assessment of the resulting impacts on estimated costs
         o  Section 4.6 sets out EnSys' review of the details of Model data and equation integrity and lists items that we see needing to be changed  
         o Section 4.7 summarizes the need and potential we see for Model improvement
         o Section 4.8 comments on the utility of the EPA Refinery Cost Model for regulatory use and describes some of the factors EnSys sees impacting U.S. refining to 2017  -  and which therefore need to be taken into account in any rigorous analysis of Tier 3 gasoline costs.          
            
Summary

Section 4 of this report contains specific findings and commentary, including to the Model cell level.  Below, we summarize our main findings.  

Adequacy of Current Model Formulation Data & Results

The EPA Refinery Costs Model comprises 1 main and 7 linked workbooks.  EnSys was only able to examine the single main workbook although we understand the role of the secondary workbooks was essentially limited to passing in data.   Based on our review, the EPA Model contains a small number of errors that were identified by EnSys plus a number of key data values that could or should be reviewed prior to any further Model use by the EPA.   A Model version was supplied back to the EPA which already embodied a number of changes made by EnSys and which could form a starting point for further review by the EPA.  

Proposed Recommendations for Changes and Their Impacts on Estimated Costs

The body of this report contains a number of recommendations for Model improvements.   Most center on corrections and adjustments to key formulae and data which have a direct impact on computed costs of sulfur reduction.  Making these adjustments raises assessed costs although not dramatically.  
In addition, EnSys has proposed a series of changes that would reduce the risk of errors in the Model and make it easier to understand, to trace calculations from source data to results, to adjust and therefore audit.  These center not so much on specific values or formulae structures as on applying available Excel modeling and documentation techniques such as using explicit named variables and ranges in formulae, using cell comments and other documentation etc.      


Regulatory Context

EPA OTAQ plans to use the Refinery Costs Model as part of its review in connection with potential Tier 3 Motor Vehicle and Emission Standards.  This proposed rule was initiated in February 2010.  According to EPA, "the standards in this rule will lead to reductions in ozone, particulate matter, nitrogen dioxide, and mobile air source toxics".  With respect to gasoline, EnSys understands that the rule could set new standards for volatility (Reid Vapor Pressure), sulfur and possibly other properties.   According to the EPA website, the Notice of Proposed Rule Making for Tier 3 standards is projected for publication in February 2012.  EnSys understands that EPA intends to use the Refinery Cost Model in developing the NPRM - but only with respect to the estimation of sulfur reduction costs, not for RVP reduction or any other potential aspects of the rule.        

Commentary & Findings

The Sections below provide an overview of the EPA OTAQ methodology, a detailed assessment of the parameters that, in our view, will most impact gasoline sulfur reduction costs, resulting recommended changes to costs premises in the Model and calculation of their effects, commentary on other detailed data and formula issues within the Model and, finally, summary recommendations for Model changes.
 
Model Version Provided to EnSys 
 
The Model version EnSys reviewed was titled as "Rev3" and dated December 1st 2011. Specifically one multi-page Excel workbook was provided for our review.   We note, however, that the Rev3.xls workbook linked to 7 external Excel workbooks.  None of those was available or was included in our review.    Our understanding from EPA was that the external files were not essential to our purposes since their primary purpose was to pass data in to the main Model workbook.  We did not see any errors relating to these external links but were, of course, not able to cross-check and validate them.    
 
EPA Methodology/Modeling Approach as a Means to Assess Costs of Gasoline Sulfur Reduction


Summary of EPA Methodology

The EPA Model supplied to EnSys uses a mix of public and proprietary refinery-by-refinery data to first estimate gasoline blendstock volumes, total gasoline production and reported pool gasoline sulfur level.  It then uses cross-checks of the estimated gasoline production from blendstocks to compare against total reported gasoline production from official figures.   Next, it uses estimates for all gasoline blend components other than FCC gasoline, together with officially reported refinery gasoline product pool sulfur to back calculate FCC gasoline sulfurs by refinery.  The model does not use that information directly but average levels were applied by the EPA as the basis for seeking FCC gasoline hydro-treating data from vendors whose processes are employed in U.S. refineries for desulfurizing FCC gasoline.  The Model then applies capital cost, hydrogen consumption and utilities data for FCC gasoline desulfurization in the Model to each refinery.   The general approach is to assume each refinery stays with its current technology supplier and to apply data relevant to that technology for revamping the existing installed FCC hydro-treater to a more severe duty in line with reduced gasoline sulfur target.  For refineries that have no current FCC gasoline hydro-treater, (generally they have an FCC feed hydro-treater), construction of a new FCC gasoline hydro-treater is assumed.    Potential need to desulfurize LSR is also assessed on a refinery-by-refinery basis and the associated costs estimated.   Capital investment plus capital recovery, fixed and variable costs are estimated on a refinery-by-refinery basis and then summed to create total national level investment and c/gal fully built up costs.   

Estimation of Blendstock Volumes by Refinery

We undertook a brief review of yields from units other than the FCC, comparing with those in the EnSys WORLD Model.  The only unit where yields warranted further comment was the hydro-cracker.  (See below.)  We note that the estimation method used data from a third party refinery LP model (MathPro), with variations by PADD, that has been calibrated and applied to a significant number of U.S. regional and national studies.  
As noted, the hydrocrackate yields in the EPA Model are taken from MathPro data.  We were not able to identify exactly which values were used for estimated HCR gasoline yield but we note that (a) the stated hydrocrackate gasoline yields vary substantially from PADD to PADD and (b) that the hydrocrackate yields are high.  Given today's growing emphasis on maximizing distillate yields, (which can mean a hydrocrackate gasoline yield of less than 10%), we would advise that EPA double-check the yields used. We would expect them increasingly today and in the future to correspond to maximum distillate production. 
   
Sulfur Levels Assigned to Blendstocks

We reviewed the sulfur levels used by EPA in the Model as tabulated below.   This included a close cross-check of values against those in the EnSys WORLD Model.  

                Gasoline Blendstock Sulfur Levels in EPA Model
Blendstock
                             Value in Model ppm S
Butane
                                      10
LSR
                                       1
Coker naphtha
                                       1
NGL
                                       1
Poly gasoline
                                       1
Isomerate
                                       1
Dimate
                                       3
Hydrocrackate
                                       8
Alkylate
                                       5
Reformate
                                       1
Ethanol
                                      10

                                       
FCC naphtha (back calculated)
                                   40 - 130

                                       

All of the values being used by EPA fall within plausible ranges on the basis that all streams will have been either Merox sweetened or hydro-treated before blending into gasoline products.  Absence of such treating will lead to higher values on a case by case basis.  We note, however, that using conservatively low sulfur values leads to higher back-calculated FCC naphtha estimated sulfur levels.  As a test, we dropped the sulfur level on all components (FCC naphtha aside) to 0 ppm.  The effect was to raise average estimated FCC naphtha sulfur by around 1.7 ppm and maximum refiner-specific sulfur by around 7 ppm.   These effects are limited, indicating the calculation of sulfur reduction costs is relatively insensitive to variations in assumed typical sulfur levels of non FCC naphtha blend stocks.  

  
 
FCC Gasoline Sulfur Level Back-calculation

We were not able to cross-check EPA's formulae in detail for back-calculating the sulfur level of FCC naphtha refinery by refinery.  The values appear to be sound.   For the different Tier 2 scenarios calculated in the Data for 2009 Model page, we note that the average (non-weighted) FCC naphtha sulfur level is estimated to lie in the range of 64-68 ppm with maxima around 120-130 and minima around 40 ppm.   Given that EPA assumed an average level of 75 ppm to assess costs for sulfur reduction, these estimates of actual 2009 FCC naphtha sulfur levels imply a modest degree of conservatism in the EPA calculations, i.e. the use of 75 ppm when the actual average appears closer to 65 ppm.    
The EPA Model uses a regression to estimate raw VGO sulfur as a function of crude sulfur.  The regression is based on data from 8 crudes and produces a near 99% R squared.    We believe this is a reasonable approach with acceptable range of uncertainty given the overall uncertainty level in this type of modeling approach. 
   
Key Determinants of Costs in Model

FCC Gasoline Volume & Sulfur Level

Costs for sulfur reduction of U.S. gasoline are clearly primarily dependent on the volume and sulfur level of FCC naphtha.   We did not research external literature to cross-check EPA's method for FCC naphtha volume calculation but their ability to use actual CBI refinery by refinery 2009 FCC charge rates is a key factor that in turn leaves open only the question of FCC yields.   Broadly, we found the yield data reasonable, especially given the back-calculation against national gasoline production contained in the Model.

Assumed Target Sulfur Levels

The extent of any "giveaway" assumed for actual blend standards against nominal standards can affect costs.   For the 10 and 5 ppm standards being evaluated, EPA OTAQ had allowed for only 1 ppm sulfur "giveaway" i.e. effective ex refinery gate sulfur targets of 9 and 4 ppm.   EPA's underlying rationale for the small levels of giveaway used was that, under Tier 3, an emissions Averaging, Banking and Trading (ABT) program would apply and that, under this, across a whole calendar year, refiners and blenders would be able to move close to the mandatory limit, for instance edging blended gasoline sulfur levels up toward the end of the year if they had produced below spec early in the year.   
As a cross-check on the validity of this assumption, EnSys obtained from the EPA 2009 data for gasoline sulfur by refinery.  The chart below summarizes 2009 data for some 92 refineries producing Tier 2 gasoline.  The graph clearly shows the effect of the ABA program which is also in effect under Tier 2.  Produced gasoline sulfur levels range from 2 to 84 ppm but we understand from EPA that the weighted average is right on 30 ppm.   The implication is that individual refinery sulfur reduction programs required to meet a 10 or 5 ppm standard will vary widely, with some refineries already producing gasoline below 10 ppm, but on average the reduction needed is going to be from 30 to either 10 or 5 ppm.  Thus, EPA are correct to have not allowed for any significant "giveaway" in the Model.











Vendor Cost Data & Transposition into Model

Cost Bases Requested by EPA

Based on the data and gasoline sulfur targets requested of vendors, the EPA Model contained costs for gasoline sulfur reduction from an estimated 25 ppm actual (30 ppm nominal) to 10 and 5 ppm nominal/actual.      
Vendor Processing Schemes

Vendor 1

Vendor 1 supplied data for a range of processing schemes including covering 75-25 and 75-10 ppm extensions of existing FCC naphtha hydrotreaters using their technology and starting from differing initial FCC gasoline sulfur levels. The cases include both Minimum Investment and Minimum Octane Loss scenarios.   In summary, we have some difficulties with both the initial vendor data and with the way EPA has applied these in the Model.   The vendor cases for unit revamps to add 75-25 and 75-10 ppm capability presented different levels of investment, octane loss, delta olefins, hydrogen and utilities consumptions depending on the sulfur level of the original raw FCC gasoline.   EPA then applied these variances in the Model as a function of their estimated raw FCC gasoline sulfur for those refineries which were assessed to already have a vendor FCC gasoline hydrotreater.  
We understand that the EPA took very specific vendor cases and applied them in the Model. We note this creates a  level of precision that goes beyond the limited differentiation supplied by all the other vendors, i.e. single data points or averages had to be employed for all but one other vendor.    The vendor data show incremental hydrogen consumptions for 75-25 ppm revamps running from 45 down to 31 and then back up to 45 scf/bbl as original FCC gasoline sulfur drops; then 66 to 60 to 100 for the corresponding 75-10 ppm cases.   Given the necessity to consider average/typical situations in this analysis, we have difficulty with the inflections in these numbers; it is difficult to consider them as credible.   Similarly, EPA's interpretation of vendor capital cost data for the revamp cases lead to base ISBL costs of $157.50/bbl for a 75-25 ppm revamp where the raw FCC feed had been 200 ppm, $50/bbl for raw 800 ppm and $548/bbl for raw 2000 ppm.   The numbers are inconsistent with EPA's corresponding values for 75-10 ppm cases, namely $450, $600 and $700/bbl respectively.     We understand that the EPA was interpreting highly specific vendor cases (which incorporated variations in process configuration and capex/opex trade-offs) but we have difficulty with the inflections in the above data for 75-25 ppm cases and also the extent to which 75-25 ppm costs varied so much based on original raw FCC gasoline sulfur level.    We recommend that EPA revisit the capital costs, and also the hydrogen consumptions across these cases.  
The EPA also used data from the same vendor for construction of a new 100-10 ppm unit and applied this both directly for situations where a refinery has no existing FCC gasoline treater (but would generally have an FCC feed hydrotreater to bring FCC gasoline sulfur down to broadly the 100 ppm level) and, with adjustments, to generate corresponding data for a new 100-25 ppm unit.   The stated ISBL cost is $1500/bbl/sd for the 100-10 ppm unit.  (It is broadly in line with the $1,830/bbl/sd used in the recent MathPro study given that the latter included both ISBL and OSBL costs.) It is stated by the vendor as applying to Minimum Investment scenarios.  We believe, as discussed elsewhere, that EPA should consider applying a contingency factor to this and potentially all capital costs.  We also note that the $1500/bbl/sd figure appears to omit catalyst cost, stated by the vendor as 20 to 35 percent of ISBL capital cost for the revamp cases. It is recommended that the vendor be contacted to provide clarification concerning this point.  
Further, EPA applied the $1500/bbl to both the 100-10 and 100-25 units.  Data from Vendor 3 (below) would suggest a moderate difference of around 5% should be applied.   (Arguably the main cost that would differ would be that of the reactor.  This can often comprise around 15% of total ISBL cost and so with a one third increase in reactor cost, 5% in total ISBL appears plausible for the shift in duty from 100-25 to 100-10 ppm.)     
The $1500/bbl figure puts the revamp capital costs into some perspective.  For the 75-10 ppm unit revamps, these range from 30  -  47% of new unit cost, depending on original FCC gasoline value.   These factors are reasonable as "industry typical" estimates.  In contrast, EPA's original capital costs for 75-25 ppm unit revamps were 10%, 3% and 36% of the $1500/bbl figure.  We aware of instances where revamp costs can be low as little by way of hardware changes would be required and we also understand that EPA was reflecting very specific vendor cases.  We still have difficulty though with these values as stated and would suggest EPA review them with the vendor to confirm whether the vendor truly believes they are realistic.  (One point of comparison would be the same vendor's fractions of new unit capital cost for the 75-10 ppm cases.)    
We have similar concerns with the EPA's derived numbers for hydrogen consumptions.  Firstly, with figures as low 14 and 6 scf/bbl, we see some of these levels as unrealistically low.  Secondly, they appear inconsistent between 75-25 and 75-10 ppm cases.  We recommend the EPA review these numbers with the vendor and at least ensure consistency.  Note that Handwerk recommends applying a 2 to 10 multiplier to stoichiometric hydrogen consumption to reflect makeup hydrogen loss.  EnSys applied some very rough and ready adjustments in the Model spreadsheet to both hydrogen and capital costs to gauge their effect.
      
Vendor 2

This vendor supplied limited data around one case for construction of a unit that would take pre-desulfurized FCC gasoline at 75 ppm down to 10 ppm.  The case assumed that 30,000 bpd of whole FCC gasoline would be split into LCN and HCN and only the HCN further processed in order to minimize olefins/octane reduction.   The capital cost quoted for the unit related to a processing capacity for the HDS unit of somewhat over 23,000 bpsd.  EPA interpreted this as a 30,000 bpsd unit in its calculation in the Model of capital cost per b/sd.  Given the vendor stated that the new equipment centered solely on a new polishing reactor with ancillary facilities (no FCC naphtha splitter or LCN HDS unit), we took the view that the stated capital cost should be applied only to the 23,000 bpsd rate; also that that approach appears to be more consistent with the costs developed for other vendors.   In the Model, we adjusted the throughput basis to 23,000 bpsd from 30,000 bpsd.    We understand, however, that the vendor, subsequent to our assessment, confirmed that the capacity is 30,000 bpsd.  

Vendor 3

Vendor 3 supplied data for single stage units running 800-75 and 800-10 ppm sulfur and for two-stage units wherein the first stage ran 800-75 ppm and the second stage either 75-25 or 75-10 ppm.   To estimate capital costs for 75-25 and 75-10 ppm reductions, EPA subtracted the cost for the 800-75 ppm single stage unit from that for the two stage 800-25 or 800-10 ppm units.  This may be plausible but see further discussion below.    Using this method, EPA made what we believe was an error (Vendor Cost Info, cell D111) in computing the capital cost for the 75-10ppm unit.  They assumed it was the same as for the 75-25 ppm unit.  EnSys substituted the vendor-provided capital cost for the two-stage 800-10 ppm unit. Doing this and then subtracting off the capital cost for the single stage 800-75 ppm unit led to a more realistic estimated cost for 75-10 ppm, a cost that is now 20% higher than that for 75-25 ppm.  
A second issue relates to EPA's method for computing hydrogen consumption.   They used the same method as for capital cost, this even though the vendor supplied data for hydrogen consumptions for each stage in the two-stage units.   For the 75-10 ppm unit, EPA's estimated hydrogen is 56.4 scf/bbl.  This number cross-checks very closely with the hydrogen consumption given by the vendor for the second stage in two-stage 800-10 ppm unit.  However, EPA's value of 18 scf/bbl for 75-25 ppm by difference compares with vendor data of 45 scf/bbl based on the second stage of the two-stage 800-25 ppm unit.  In our view, 18 scf/bbl appears low and 45 scf/bbl more reasonable.   For the other utilities, and for octane loss and olefins change, EPA had no option but to take the delta between the single stage 800-75 and the two-stage 800-25 and 800-10 ppm cases. 
We understand that the vendor was asked to provide these cases as if the refiner would decide to build one or other of the units from scratch.  In reality, the situation is generally one where the, say, 800-75 ppm unit already exists and the question is - what is the cost of extending the duty of the existing unit down to 25/10 ppm via revamp?   We suggest that the EPA go back to the vendor to seek their opinion on whether considering the costs in this light would make a difference, including on how EPA has to date developed its 75-25 or 75-10 ppm costs by difference.  We also suggest that the EPA review with the vendor the differences on hydrogen consumption to obtain the vendor's view on which approach gives the more realistic results.         

Other Vendors & Average Costs

The EPA also applied average costs and related data.   We suggest the EPA check the formulae to see if these best represent actual averages.  
The EPA also received data from two other vendors but had decided not to use either.  One set of data appeared to be for a specification that was unrelated to the issue at hand.  The other set of data related to a little used technology.  EPA took the conservative view to apply data averaged from Vendors 1, 2 and 3.  Hydrogen consumptions were higher than for the little used technology.   Given the data supplied to the EPA, we support their decision to not use the above two data sets.  However, we would recommend that the EPA, if feasible, go back to the vendors to seek clarifications. 

All Vendor Capital Costs

In the Model, EPA applied scale, location, offsites (OSBL) and over-design factors to the capital costs.  Our experience has been that vendor-supplied costs data, which is what EPA relied upon, tend to also be relatively optimistic.  (One vendor stated its supplied capital cost had an accuracy of +/- 50%.)   In the light of uncertainties in the costs of unit construction,  we would therefore suggest that the EPA consider applying a contingency factor to all the capital costs, potentially of around 20%.   Again, the option to apply a factor was added in the Model spreadsheet.   
Capital Costs (ISBL and Total)

The vendors reviewed above provided their capital cost estimates on an ISBL basis.  EPA applied their own OSBL factors.  These appeared reasonable for process unit revamps where infrastructure is in place. 

Octane Loss

We were not able to cross-check the vendor data for olefins reduction and octane loss.  Vendors A and C estimated octane loss at 1.0 for unit revamps, somewhat higher than the 0.6 value estimated by EPA.  Again, it would be helpful to test the impact of level of assumed octane loss on resulting costs in the Model. 

Other Key Cost Parameters

Capital Cost Power Factor

The EPA Model uses a scaling power factor of 0.65.  This appears reasonable and if anything conservative.  As noted elsewhere, however, this figure is hard-coded into all pertinent formulae.   The Model would be improved by assigning to a specific cell a name and the value and then referring to it.  That way, formulae would be more transparent and the sensitivity of results to say using 0.60 as against 0.65 could be tested.  

Natural Gas, Electricity & Steam Prices

Natural Gas
In the EPA Model, prices for natural gas and electricity were understood to be taken from the EIA AEO 2010 Reference Case.  Base costs by PADD were escalated up to 2017 costs (expressed in 2009 constant dollars) using a factor.  The resulting 2017 costs ranged from around $5.60 to $7.90 per million Btu depending on the PADD, as summarized in the table below.   To check these values, EnSys gathered data from the EIA Natural Gas Monthly to first provide current natural gas prices.  Specifically, we:
    1. downloaded data for Major Industrial User natural gas prices by state for the period January 2010 through September 2011 (the latest month available)
    2. established average prices across the Jan 2010  -  Sept 2011 period for every state
    3. identified each state within each PADD that contains refineries (for instance in PADD1 only New Jersey, Delaware and Pennsylvania are relevant, New York, Vermont etc. are not as potentially suppliers of natural gas to refiners)
    4. developed an average for each PADD based on the states that contain refineries (again for PADD1: NJ, DE, PA).
We then compared these current prices with the base prices by PADD used by the EPA.  The comparison is included in the table below.  The comparison indicates that the EPA base prices understate the differences that exist between PADD's for major industrial user prices, i.e. the cost increases in moving from PADD3 (the lowest cost PADD) to other PADD's.   The last two columns in the table compare the 2017 prices currently in the EPA Model with prices that would result from, in general, adding the $0.43/mmBtu difference between the PADD3 2017 and 2010/11 prices to the EIA 2010/11 industrial user prices to arrive at a revised set of prices for 2017.  EPA may wish to use another methodology but our overall reaction is that the 2017 prices should reflect sharper distinctions between PADDs, this unless the EIA, say, has a clear basis for these deltas changing in the future. 
 Overall, we recommend EPA review the natural gas prices used, especially outside PADD3.  Natural gas price can be expected to have an appreciable impact both as fuel and as feedstock for hydrogen production.   For both natural gas and electricity prices, we would recommend that EPA update to the AEO 2011.  We note that a recent MathPro report for the ICCT, "Refining Economics of a National Low Sulfur, Low RVP Gasoline Standard", October 25th, 2011, used AEO 2011 projections.  Exhibit A-6 indicated lower natural gas prices for their study year 2015 than for 2010. (This EIA projection seems to EnSys somewhat unexpected given natural gas prices are currently at historical lows and, reportedly, the costs of shale gas production are running above the $4/mmBtu level producer price, indicating a likelihood of increases in the future.)    
               Natural Gas Prices Comparison ($2009/million Btu)

EPA Base 2009
EIA 2010/2011
EPA 2017
Revised 2017
PADD1
4.87
9.96
7.88
10.30
PADD2
4.86
7.44
7.86
7.87
PADD3
3.49
5.21
5.64
5.64
PADD4
4.36
6.34
7.06
6.77
PADD5 ex CA
4.60
8.33
7.44
8.76
PADD5 CA
n.a.
7.04
7.44
7.49






Electricity
Electricity prices are generally likely to have less of an impact on fuels costs than do prices of natural gas.  EnSys did not check the electricity prices used by the EPA so thoroughly.  We note though that EPA does increase prices by a factor of 1.45 to arrive at 2017 prices versus the 2009 base levels EPA uses.  We note that the AEO 2010 shows U.S. industrial end use price declining in $2008 from 6.8 c/kwh in 2009 to 6.1 c/kWh in 2017.  Given EIA's projected increase in natural gas prices, we would support the factor used by the EPA.   Our main question is that the base prices used by EPA for electricity, which average 4.46 c/kWh, appear low relative to the AEO 6.8 c/kWh for industrial users, i.e. arguably the base levels should be increased leading to 2017 prices that are higher than the current 4.7  -  8.4 c/kWh in the EPA Model.            
Steam
EnSys identified two errors in the Model with respect to steam costs calculation. One error was that the EPA Model was picking up only steam variable cost by PADD, not total cost.  The error was marked and corrected. The effect was to add 0.02 c/gal to total sulfur reduction costs.   The second error was that the Model's formulae for steam generation capital cost contained a built in value of 0.04 for amortization instead of using the 0.11 value EPA used elsewhere in the model.   Making this correction nearly tripled the steam capital cost contribution and raised total steam cost by over 20%.  The effect on total gasoline sulfur reduction costs (10 ppm) was to add a further 0.03 c/gal.  

Hydrogen Cost

In the EPA Model, hydrogen costs are developed for each PADD against a 2017 timeframe using data for hydrogen production from natural gas via steam reforming.     We undertook an analysis of the hydrogen costs in the EPA Model.   We checked costs calculated by the EPA against costs calculated using the data for the hydrogen plant in the EnSys WORLD Model.   The results (PADD3 basis) came out within a few percent of the EPA figure for $/000 SCF H2.  We also checked the EPA hydrogen cost on a $/BFOE basis against natural gas price on the same basis.   The typical view is that hydrogen is around 2 times natural gas price on a BFOE basis.   The EPA factor comes out at almost exactly 2.   As a further check, we altered the EPA amortization factor from 0.11 to 0.243, a commonly used value for "hurdle rate" required capital recovery factor.  Making this change altered the BFOE basis hydrogen to natural gas price ratio to 2.37.   This is essentially identical to the 2.38 factor quoted by Baker & O'Brien in a July 2011 report for the American Petroleum Institute on the costs of U.S. gasoline and sulfur reduction.   In all in all, we believe these cross-checks reinforce the values for hydrogen cost in the EPA Model.  

Cost of Octane Loss

In the Model, the EPA gathered together a number of estimates for cost of octane.  These included:
   1. A stated historical cost of $5.29/bbl for premium  -  regular 2008-2011.  This equates to around $0.90/bbl per octane
   2. A "refiner cost" of $1.19/bbl.  We are unclear whether this was per octane or PRM-UNL delta
   3.  Run results by PADD from the EPA GRTMPS LP model.  These were, apparently for PRM-UNL, respectively: P1 $1.91, P2 $1.56, P3 $2.07, P4 $1.21, P5 ex CA $1.21, P5 CA $2.52, i.e. around $0.20 - $0.45/bbl per octane.

The LP results were lower - on a $/bbl per octane basis  -  than the historical levels. They showed highest costs for California and for PADD3 which did not seem entirely realistic.   

To cross-check these values, we looked at other sources.   We examined available Bloomberg pricing data as EIA data do not now provide sufficient grade breakdown.   The Bloomberg data show that:
   * Costs per octane averaged around $0.45/bbl from 2000 through 2004 then rose and have since remained around the $1/bbl per octane range, broadly consistent with EPA's stated historical $0.90/bbl per octane 
   * On a nationwide basis, costs have changed little between 2005-2007 and 2008-2011 but there have been changes within regions.  Octane costs in PADD1 (NYH) and PADD3 (USGC) and for PADD5 LA CARBOB have gone down whereas those for the Midwest and for PADD5 LA CG have risen.  This may relate to changes in ethanol supply, demand, distribution and also blending methods
   * In terms of regional differences, PADD1 costs have been consistently above PADD3 whereas the recent rise in PADD2 costs has moved those from below to above PADD3.  PADD5 costs have fluctuated but have roughly in line with those for PADD3.
As a further cross-check, we examined results from EnSys WORLD Model cases from our 2010 Keystone XL study for the DOE and DOS.   For the 2015 case, these showed costs ranging from a low of around $0.25/bbl per octane for PADD3 CG and $0.45/bbl for PADD3 RFG to $0.45/bbl for PADD1 RFG and around $0.40/bbl for PADD5 CG and $0.70/bbl per octane for PADD5 CARB RFG.   In the 2020 case, the octane costs were appreciably lower.  This would be expected since the cases assumed rising ethanol supply into gasoline over time in line with RFS-2.  
We were not able to test the effects of assuming different levels of octane cost in the Model.  We understand, however, the EPA used relatively low per octane costs.   We believe the above data could be useful in providing guidance to the EPA.  It looks as though:
    1. Costs will have declined by 2016/2017 because of rising ethanol production and moves to take full advantage of ethanol's high octane.  (This could be a factor in the historical versus LP results, i.e. if the real world distribution system was historically not taking full advantage of ethanol's octane but the LP model did then that would account for real world octane costs significantly above LP results.   As industry better optimizes blends over time to more fully apply ethanol's high octane then octane costs should drop and real world versus LP results narrow.)
    2. Comparison of recent real world costs with WORLD LP results would suggest that 2016/2017 octane costs could be at least 50% less than those for 2008-2011, potentially even lower, i.e. somewhere in the $0.50 - $0.25/bbl national average per octane
    3. Arguably, differences in octane costs by region would still apply, with PADD1 potentially around $0.10 - $0.20/bbl per octane above PADD3 and PADD's 2 and 5 potentially broadly the same as PADD3 for lack of any ability to clearly differentiate.      
All of the above data and projections (aside possibly from the EPA GRTMPS LP results) were based on E10 and E85 blending, not E15.   Moving to E15 would argue for octane costs at the lower end of the estimated ranges. 

E10/E15

As stated above, we understand that the EPA's cost projections for gasoline sulfur reduction in the Model are based on presuming E15.  The MathPro 2015 cases for the ICCT assumed E10 plus E85.  We believe it would be useful if the Model could have the flexibility to examine the impacts of sulfur reduction at 10% as well as 15% ethanol.  The effect would be to raise sulfur reduction costs as concentrations of FCC gasolines would be higher. 

Additional Factors Influencing Costs

LSR Sulfur Reduction

LSR sulfur reduction is the second main factor applied in the Model to achieve gasoline pool sulfur reduction.   We support the EPA view that this is a relatively low cost option and therefore likely to be employed.  We briefly checked the EPA rationale for assessing, refinery by refinery, what if any LSR HDT capacity would be added and whether it would be revamp or new.  Again, this logic was complex. In our view it works recognizing the limitations in our review. 
We next focused on the LSR HDT costs in the Model.    The base capital cost taken by EPA from Handwerk appears reasonable as does the hydrogen consumption.   We note, however, that EPA do not appear to apply the Overdesign factor of 1.15 they hold in the Utility & Capital Costs page for either LSR HDT revamp or new unit or the 0.85 factor present for capacity utilization.  We suggest EPA check this and apply at least the service factor if not both of these factors.   To gauge the effect, EnSys applied both factors.         

Other

The EPA Model does not appear to explicitly consider cost factors outside of FCC gasoline and LSR desulfurization units.  Arguably, there should be some small additions to acid gas and sulfur recovery plants although these might be considered part of offsites.   
  
Model Results, Proposed Premise Adjustments and Their Impacts 
 
Results from Current Cost Model

The table below breaks out the impacts in the 10 ppm case of the changes we made or proposed to the Model.   These included formula corrections and suggested changes to data values.   As is evident, the effects are minor with the potential exception of changes to vendor capital cost, hydrogen consumption and related values.   As previously discussed, the need for and extent of any such modifications is the subject, in large part, of obtaining further clarifications from the vendors regarding the data they supplied.  Actual appropriate changes may be significantly lower than those indicated below.  

Summary of Impacts on Costs of Proposed Model Changes c/gal Total U.S. Gasoline

                          individual effect (10 ppm)
                          cumulative effect (10 ppm)
Corrected steam cost lookup formula
                                     +0.02
                                     +0.02
Corrected steam amortization formula
                                     +0.03
                                     +0.05
Adjusted non PADD3 natural gas prices
                                     +0.02
                                     +0.07
Vendor capital cost, hydrogen and related adjustments
                                  up to +0.26
                                  up to +0.33
LSR HDT service and over design factor adjustments
                                     +0.03
                                     +0.36

                                       
                                       

to 2.2 c/gal at 10 ppm and 2.54 at 5 ppm.  Use of 24.3% arguably indicates the upper end of the range for estimated average costs.  
The data in the table represent  only preliminary values prior to EPA's further review and adjustments to the Model.  Note, we did not test for the effects of potential changes in octane cost. 
We would further point out that, at present, the Model only reports at the national level.  We recommend that reporting be extended to the PADD level.  This would both provide additional insights and add a form of sanity cross-checking based on the relative c/gal costs assessed by PADD.   

Detailed Model Integrity Findings/Issues

The EPA requested that EnSys assess the integrity of the refinery-by-refinery cost model by working through the equations present in the spreadsheet.  As mentioned in a prior Section, the Model version EnSys reviewed was titled as "Rev3" and dated December 1st 2011. Specifically one multi-page Excel workbook was provided for our review.   As noted elsewhere, the Rev3.xls workbook linked to 7 external Excel workbooks.  None of those was available or was included in our review.    Our understanding from EPA was that the external files were not essential to our purposes.  We did not observe any errors relating to these external links but were, of course, not able to cross-check and validate the links.    
The following describes our approach taken and specific findings. 

Approach Employed by EnSys

The time available for EnSys to review the EPA Model was limited to 5 days (with 2 people reviewing).  We were therefore not able to rigorously analyze and check every formula or parameter.   We consequently focused first on what we considered were the key parameters that would have the most significant impacts on results, i.e. applying the EPA methodology as developed.   That part of our analysis is the focus of Section 4.   
We used that analysis also as the basis for a more detailed probe.  We tracked specific key formulae, such as for calculation of capital costs, in doing so finding certain errors as noted in Section 4.  In addition, we tracked the full set of formulae for one refinery rigorously and spot-checked for others, in that regard finding no errors in the formulae we examined.  

General Comments on Model Structure

The above line of approach was consistent with our focus on factors which could impact cost results.  A second focus of our review was on the quality of the Model in terms of its use of sound Excel modeling practices.   In that regard, we believe the Model can be significantly improved.   The following, based on our findings, is not necessarily an exhaustive list:
   * the same data item appears to be entered in more than one place in the model
   * at times data are labeled as say (2009) "capacity" when in fact the numbers represent actual (2009) charge rate. An example is Page T3, column R. It is under a heading for 2009 capacity whereas the values entered are 2009 charge rate
   * while the spreadsheet does contain a number of defined names, this facility is generally not used in formulae; rather formulae generally contain only cell references and consequently are difficult to follow and more prone to error
   * in numerous formulae, values are hard wired when, in our view, they should not be; for example, base unit capacities relating to data supplied by vendors are hardwired into capital cost calculation formulae rather than being explicitly referenced; capital cost power factors are hard wired in.   This approach significantly increases the chance of errors, makes the Model more difficult to understand and review in terms of key premises and logic used and makes it more difficult to alter premises and test the effects of doing same 
   * there is a near total lack of internal documentation including such useful items as Excel cell comments.      
Overall, these make the Model more difficult to adjust, cross-check and keep accurate; also to hand on efficiently and reliably to a new user. 
Specific Data & Equations Integrity

Capital, Fixed & Variable Operating Cost Formulae

EnSys explicitly checked these key formulae item by item.   The formulae appear correct recognizing that the capital cost formulae contain hard wired values for vendor base unit capacity and for power factor for capital cost scaling base on actual estimated FCC gasoline treater capacity versus vendor base capacity.    
EnSys undertook its analysis using selected formulae in the COSTS 10ppm page.  We spot checked additional formulae in the same page and also in the COSTS 5ppm page.   The latter page appeared to be correctly picking up vendor-derived data for 5ppm (not 10ppm) cases.   
 
Potential for Model Improvement

As stated at various points across Section 4, the Model as it stands today does contain a small number of formula errors and a number of data values which warrant further review.  EPA's correction of these errors and review of selected data items should improve the Model. 
Specific opportunities exist in the form of adopting better practice and discipline regarding using explicit variables wherever possible, naming cells and ranges and using these in formulae, extending the use of LOOKUP and related formulae which tend to ensure correct values are produced and using cell comment and other techniques to establish and maintain better internal documentation.   

Utility for Forward Looking Regulatory Analyses 

The EPA Refinery Cost Model is fundamentally a static analytical tool, although it can be updated to new historical base year data.  It also is currently limited to addressing costs of gasoline sulfur reduction only.  Thus, while it can make a contribution to EPA's current assessment of the costs of potential Tier 3 regulations, which could include RVP as well as sulfur reduction, the Model's analytical scope is limited.   
Arguably, for rigorous analyses, the EPA has the choice of using either the Refinery Cost Model or LP modeling.  Each method has its pros and cons.  The former, as noted, is restricted in its scope and ability to look forward.  The latter (LP) has the benefit that it can assess multiple regulatory changes and their interactions  -  not just one  -  and can and should take into account the changes that are likely to occur across U.S. refining over the several years to the analytical time horizon.   That said, at least equal rigor needs to be applied to an LP as to a spreadsheet model, in part to obviate the risk in an LP to under or over optimize and/or generate extreme solutions.  Also, the spreadsheet model is able to compute (approximate) gasoline sulfur reduction costs by individual refinery whereas the LP will almost inevitably deal with aggregated groups of refineries.  
Given the EPA's analytical timeframe of (around) 2017, to be realistic and defensible, any rigorous assessment of costs needs to take into consideration a range of developments  -  and uncertainties - impacting U.S. refining.  The following is a partial list:
   * Current major projects under way at selected U.S. refineries that will add and modify capacity, including ability to process higher volumes of heavy Canadian crudes. (See below for a listing of current major projects.)  
   * Flat to declining U.S. gasoline demand in association with rising ethanol supply; sensitivity of refining situation and sulfur recovery costs to actual levels of U.S. gasoline demand and future ethanol supply achieved
   * Potential shifts over time in FCC feedstocks and operating modes, notably: shift towards more resid in FCC feed as VGO is increasingly "pulled away" to hydro-cracking units as demand for distillates grows, FCC catalyst/operating shifts to lower gasoline higher distillate (LCO) and also higher propane yields; potential resulting impacts on FCC gasoline qualities, volumes and thus concentration in the U.S. gasoline pool
   * Extent to which current trends for growth in U.S. refined products exports will continue and their potential impacts on U.S. refining
   * Potentially substantial changes in U.S. domestic oil liquids supply, both light sweet crudes from the Bakken , Eagle Ford, Niobrara, Utica shale and other plays and the potential impacts of rising NGL's supply
   * Potential for continuing growth in Canadian oil sands production and exports to the U.S.
   * Sensitivity of U.S. crude slates, by region, to the above supply developments combined with logistics developments (Keystone XL and other projects that could influence crude dispositions especially to PADD3 refineries).
 Without taking account of these and other factors, EPA analyses could either under or over state costs and/or be subject to criticism for lack of realism.          


                         Major U.S. Refinery Projects
Refinery 
                             Crude capacity 000bpd
                   Incremental heavy crude processing 000bpd
                                   Start up
WRB Refining Wood River Illinois
                                      75
                                      130
                                    2011/12
WRB Refining Borger Texas 
                                      50
                                      110
                                    2011/12
Marathon Detroit Michigan
                                      13
                                      80
                                    2H 2012
BP Whiting Indiana
                                      35
                                      260
                                     2013
Motiva Port Arthur Texas
                                      325
                                 Up to 325 (1)
                                    1H2012
Valero Sunray, Corpus Christi and Three Rivers Texas  -  minor debottlenecking to process additional Eagle Ford etc. 
                                       
                                       
                                       
   1. Whether the newly expanded Motiva refinery will process primarily heavy Western Canadian or Saudi crudes has been the subject of recent debate. 


