How do we attribute emissions from transportation? #39
Replies: 19 comments 21 replies
-
|
What are the differences in current practices by the two databases? |
Beta Was this translation helpful? Give feedback.
-
|
Yes its on the list for us to document the differences more clearly. |
Beta Was this translation helpful? Give feedback.
-
CEDA has a Two-Layer Process that attributes emissions from transportation (fuels) to sectorsLayer 1: Deriving Fuel Allocation (derived.py)This layer figures out how much fuel each economic sector uses.
Step 2: Map Vehicles to Economic Sectors
Step 3: Handle Complex Allocations
Step 4: Combine All Fuel Types
Output: A table showing each sector paired with how much of each fuel type it uses. Layer 2: Converting to Emissions Percentages & CO2 (fuel_usage.py)Once we know fuel amounts by sector, this layer converts them to carbon emissions by sector.
Step 2: Load EPA Emissions Data
Step 3: Allocate Emissions by Sector
Output: A list of economic sectors with their allocated CO2 emissions from a specific fuel type (e.g., aviation gasoline). |
Beta Was this translation helpful? Give feedback.
-
|
After chatting with @catherinebirney , I understand that FLOWSA uses data from GHGI Table 3-13 where CO2 emissions are already allocated to fuel and vehicle types which are directly mappable to NAICS then to BEA. 
CEDA uses a two-layer process to first determine the amount of fuel use by vehicle/transportation types (which are directly mapped to BEA), then allocated national total CO2 emissions from transportation to each vehicle/transportation type based on the amount of fuel use.
|
Beta Was this translation helpful? Give feedback.
-
|
Conceptually, I think it's reasonable for Cornerstone to adopt FLOWSA's approach for allocating transportation CO2 emissions. |
Beta Was this translation helpful? Give feedback.
-
|
I looked back and realized we made this swap in flowsa back in 2023 (internal issue; commit). I realized that the output should be the same with this simpler approach. |
Beta Was this translation helpful? Give feedback.
-
|
Thanks @bl-young , I can't open the issue link. Did you conduct a comprehensive analysis to compare before and after the swap? |
Beta Was this translation helpful? Give feedback.
-
I do not recall. If I did, I do not have documentation of it. But I noted in that issue at the time: "My recollection was that this should have not real impact on the FBS, but simplifies the method by requiring one fewer steps in attribution." |
Beta Was this translation helpful? Give feedback.
-
|
@catherinebirney and I added scripts to allocate transportation emissions in the FLOWSA way and CEDA way. |
Beta Was this translation helpful? Give feedback.
-
Aviation CO2 emissions (NAICS 481***, BEA 481000)FLOWSA's flow-by-sector has the following results, where, in
CEDA's results are, where
In total, FLOWSA attributes 168.9 MMT CO2 to On the other hand, FLOWSA allocates 11.5 MMT CO2 to Military Aircraft in Federal gov (defense), 0.8 MMT more than CEDA's 10.7 MMT. Since Table 3-13 has clear categorization of fuel and vehicle type that, I think we should use it to apply direct attribution to both air transportation and federal gov (defense). |
Beta Was this translation helpful? Give feedback.
-
|
Thanks @catherinebirney for generating the FBS from FLOWSA code, but I think we could put up a PR like this one to include the code that we ran to get to these FBS and allocated emissions tables. I will commit the code I use to get to CEDA results as well as the results in CSV. @catherinebirney could you follow up there to commit the code from your side? |
Beta Was this translation helpful? Give feedback.
-
Rail transportation CO2 emissions (NAICS 482***, BEA 482000)Very simple direct mapping to Rail sector for both FLOWSA and CEDA, except small rounding difference. FLOWSA
CEDA
|
Beta Was this translation helpful? Give feedback.
-
Water transportation CO2 emissions (NAICS 483***, BEA 483000)FLOWSA again directly allocates the emission, whereas CEDA applies some allocation but the sum is close to FLOWSA's allocated number 24.3 MMT. FLOWSA: 24.3 MMT
CEDA: 16.6 MMT + 7.6696 MMT = 24.3 MMT
It's worth noting that CEDA kept the fuel type separated, i.e. Residual Fuel and Distillate Fuel Oil, which is helpful to emission source analysis. I think we can and should keep them separated as the source data (GHGI T3-13) has them ready. |
Beta Was this translation helpful? Give feedback.
-
|
Fair point. Occasionally we consolidate activities to reduce the number of activities we have to map. See this dictionary entry in the method file. Very easy to break that back out again. |
Beta Was this translation helpful? Give feedback.
-
Truck transportation CO2 emissions (NAICS 484***, BEA 484000)FLOWSA and CEDA have similar results for emissions from Gasoline and LPG
But the two models have quite different results for Distillate Fuel Oil:
FLOWSA:
CEDA
|
Beta Was this translation helpful? Give feedback.
-
|
@bl-young @catherinebirney fyi ^ Distillate Fuel Oil is a big diff we need to resolve. |
Beta Was this translation helpful? Give feedback.
-
Transit and ground passenger transportation (NAICS 485*** for private sector and S00201 for government sector, BEA 485000)Both FLOWSA and CEDA apply some allocation and end up having very close total emissions and slightly different breakdown. FLOWSA: 0.3 + 4.2 + 3.1 + 14.4 = 21.9 MMT
CEDA 4.1 + 0.1 + 17.3 = 21.5 MMT
It's worth noting that CEDA allocates zero CO2 from natural gas to Buses because in Table A-69 (fuel consumption) the value is withheld therefore interpreted as zero which then led to no allocation. |
Beta Was this translation helpful? Give feedback.
-
|
And here, CEDA combines S00201 with 485 correct? So what happens here is dependent on #44 |
Beta Was this translation helpful? Give feedback.
-
|
Correct! Let's discuss and make a decision over there. |
Beta Was this translation helpful? Give feedback.
-
Pipeline transportation (NAICS 486***, BEA 486000)Very simple sector, both FLOWSA and CEDA directly allocate CO2 and end up having close results. FLOWSA: 71.3 MMT
CEDA: 71.7 MMT
Small discrepancy is because CEDA doesn't allocate Natural Gas to Bus or Truck transportation, which should be fixed if we adopt the FLOWSA approach. |
Beta Was this translation helpful? Give feedback.
-
Scenic and sightseeing transportation (NAICS 487*** and 488***, BEA 48A000)Simple allocation, both FLOWSA and CEDA end up having ~2 MMT, small discrepancy is likely due to the difference in underlying Use matrix, i.e. FLOWSA uses FLOWSA: 1.9 MMT
BEA: 2 MMT
|
Beta Was this translation helpful? Give feedback.
-
Postal service (NAICS 491***, BEA 491000)Both FLOWSA and CEDA used U matrix to allocate emissions. But FLOWSA only allocates (Gasoline?) emissions to passenger cars (0.358 MMT), whereas CEDA allocates both Gasoline emissions (0.425 MMT) and Distillate Fuel Oil (diesel) emissions, i.e. truck transportation by postal services (1.17 MMT). @bl-young @catherinebirney I think it's reasonable to allocate some diesel emissions to postal services under both "Light-Duty Trucks" and "Medium- and Heavy-Duty Trucks", given USPS does operate fleets of trucks. FLOWSA : 0.358 MMT
CEDA: 1.598 MMT
|
Beta Was this translation helpful? Give feedback.
-
|
Good spot with this discrepancy. I agree we will need to determine the path forward here. Your suggestion seems reasonable, but I think is also a good example of why the transportation approach in general needs to be improved if we can find better data, given that we have no way to distinguish between consumption of gasoline and diesel by different sectors (or even for non-transportation use) with the use table. |
Beta Was this translation helpful? Give feedback.
-
|
I dug up this report for USPS, which I think both validates your thinking that distillate is used by the fleet, but also highlights the issue, which is that the amount of diesel consumed is 20% of the amount of gasoline, and yet based on the attribution of emissions, the emissions from diesel trucks in CEDA is about double that of gasoline. |
Beta Was this translation helpful? Give feedback.
-
Couriers and messengers (NAICS 492***, BEA 492000)Very simple sector, both FLOWSA and CEDA directly allocate CO2 and end up having close results ~20 MMT. Small discrepancy is likely due to the difference in underlying Use matrix, i.e. FLOWSA uses U from SUT whereas CEDA uses U from MUT FLOWSA: 21.2 MMT
CEDA: 20.2 MMT
|
Beta Was this translation helpful? Give feedback.
-
State and local government educational services (NAICS 923110, BEA GSLGE) and hospitals and health services (NAICS 923120, BEA GSLGH)FLOWSA doesn't allocate any transportation emissions to these two sectors, whereas CEDA allocates a total of 133 MMT. This explains the under-allocation by CEDA in Truck transportation.
|
Beta Was this translation helpful? Give feedback.
-
|
@bl-young @catherinebirney let's discuss this with truck transportation |
Beta Was this translation helpful? Give feedback.
-
Remaining sectors including NAICS 922*** and 928*** that do not cleanly map to BEA GSLGO, S00102, S00203, and S00600Overall, FLOWSA allocates a total of 134 MMT to NAICS 922*** and 928***
While CEDA allocates a total of 219 MMT to BEA GSLGO, S00102, S00203, and S00600.
The difference is likely coming from CEDA's under-allocation in Truck transportation. |
Beta Was this translation helpful? Give feedback.
-
|
The nuances with these sectors would be cleared up once we decide on the core schema for GHG model. |
Beta Was this translation helpful? Give feedback.
-
|
My fear with attribution mobile emissions to these sectors is that they could be getting a significantly higher share of mobile emissions due to fuel consumption for stationary combustion purposes, and we don't have any way to differentiate that right now. I guess its similar to why we don't attribute mobile emissions to any industries for their own vehicle usage. |
Beta Was this translation helpful? Give feedback.
-
Lastly, final consumers (F01000)FLOWSA allocates and saves emissions to F01000, whereas CEDA doesn't. But since we've decided to account for emissions by households, I think it's reasonable to adopt the current approach FLOWSA takes with small modification if necessary. |
Beta Was this translation helpful? Give feedback.
-
|
In general, the consensus is to use the approach implemented in flowsa (based directly on 3-13). However there are a few key mapping differences to resolve: Summary of Mapping differences
|
Beta Was this translation helpful? Give feedback.
-
|
The transportation emissions allocations using the Use table will be impacted by this choice because the redefinitions include movement of output to and from the transportation sectors. |
Beta Was this translation helpful? Give feedback.
-
|
The major shortcoming of both of these methods is the arbitrary selection of sectors to which to allocate transportation emissions in general, which are limited to 48* and government sectors. This perhaps dates back to data-less choices made in the initial USEEIO v1 GHG satellite table model which later became the National GHG Industry Attribution Model. This is inconsistent both my non-scientific observation on the roads and service stations of the U.S. seeing a significant portion of vehicles, cars, light and medium-duty trucks having commercial marking and otherwise being used for work purposes to get workers and equipment to job sites. Also the Use tables shows vehicle/transportation related goods and services being purchased by commercial enterprises. For examples, in the 2017 BEA_Use_SUT 811000 "Automotive repair and maintenance (including car washes)" is being purchased by nearly every 5*, 6*, 7* and 8* industry. And of course the same for Petroleum refinery products being purchased by all these industries - which is itself used for further attribution but only to further attribute between the arbitrary transportation and government industries, when the major of this use by commercial sectors is likely for transportation fuels. The error that results from the current approach used in both USEEIO and CEDA is likely the over-attribution of the transportation GHGs to households, since the GHGs to from passenger vehicles are exclusively attributed households, and the undercounting of commercial sector GHGs. |
Beta Was this translation helpful? Give feedback.
-
I believe redefinition of Use table would re-distribute transportation fossil fuel consumption from sectors that have commercial markings but not primarily producing transportation services to sectors where the primary output is transportation services. @jvendries I remember you compared before and after redefinition IO tables - is this aligned with your findings?
Purchases of "Automotive repair and maintenance (including car washes)" does not indicate direct consumption of transportation fuels like diesel and gasoline in order to operate vehicles for transportation purpose. Sectors like this should be excluded from the GHG attribution IMO. |
Beta Was this translation helpful? Give feedback.
-
And so specifically, our question is, when we look across sectors in the after redefinitions table, do we see a decrease in fuel purchases from refinery and an increase in transportation purchases across sectors (e.g., manufacturing, gov't etc), that would indicate this shift is happening. To the extent that this is happening, it would encourage us I think to attribute more directly just to the truck transport sector. |
Beta Was this translation helpful? Give feedback.
-
We haven't been able to exactly duplicate the monetary reallocations happening between before and after redefinitions for specific sectors, but a quick look at the commodities 324110 (petroleum refining) and 484000 (truck transport) in both the before and after redefinitions tables doesn't seem to indicate a reallocation away from fuel purchases maching a reallocation to truck transport. See below for additional context, though I should emphasize this is based on net reallocations between commodities and so we should not be drawing definitive conclusions. For example: in the image below, the cell values are the use table after redefinitions divided by the use table before redefinitions (so a value of 1 means no change, and a value of 0 means a total reallocation of that value from that column), and the use table commodities (rows) have been transposed for easier inspection. Here we're looking only at the 324110 and 484000 transposed commodities, sorted from smallest to largest on 324110: 
These are the first 25 or so rows from that comparison However, because these are ratios of probably very different magnitudes between commodities and we can only see the net effect of all reallocations, we can't definitively say that some value from petroleum wasn't transferred to trucks. All we can say that alot of the petroleum consumption of 515200 was transferred away somewhere else. In general though, there are few sectors for which the value of the commodity ratio for 484000 is larger than 1, indicating a net transfer to that sector, and for those that are greater than 1, it is usually the case that the value for 324110 is also larget than 1, also suggesting no net transfer between these commodities: 
|
Beta Was this translation helpful? Give feedback.
-
|
Resolution We will use CEDA's approach and use FLOWSA to allocate to government sectors (GSLGE, GSLGH, GSLGO), but revisit in the future given the likely the over-attribution of the transportation GHGs to households |
Beta Was this translation helpful? Give feedback.
-
|
I created this issue to work on improving the transportation emissions allocation.. |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
Emissions are from 3-13 (CO2), 3-14 (CH4), 3-15 (N2O), and A-90 (HFCs)
Beta Was this translation helpful? Give feedback.
All reactions