EIA vs. FERC Form 1 fuel cost data

[zaneselvans]

From @gschivley:

Does anyone have background knowledge of the EIA fuel cost data? I know not all plants report, but can I be reasonably confident about the $/mmbtu values in out_eia__yearly_generators? Are they all-in, including transportation? Do they agree with fuel costs from out_ferc1__yearly_steam_plants_fuel_sched402? I assume EIA has better coverage but haven't had a chance to confirm yet. We've concluded that EIA AEO's regional coal fuel prices for electric plants are generally too low. Calculating values from EIA data seems to give better results for coal dispatch.

Here's the values I get for a 26-zone region. SQL query below. The gen_region table is the attached csv, referenced using duckdb.

SELECT 
    g.region,
    CAST(SUM(og.capacity_mw) AS INTEGER) AS capacity_mw,
    ROUND(SUM(CASE WHEN og.fuel_cost_per_mmbtu IS NULL THEN og.capacity_mw ELSE 0 END) * 1.0 / SUM(og.capacity_mw), 2) AS fraction_cap_null_fuel_cost,
    ROUND(SUM(CASE WHEN og.fuel_cost_per_mmbtu IS NULL THEN og.net_generation_mwh ELSE 0 END) * 1.0 / SUM(og.net_generation_mwh), 2) AS fraction_gen_null_fuel_cost,
    ROUND(SUM(og.fuel_cost_per_mmbtu * CASE WHEN og.fuel_cost_per_mmbtu IS NULL THEN 0 ELSE og.net_generation_mwh END) / SUM(og.net_generation_mwh), 2) AS weighted_avg_fuel_cost
FROM 
    out_eia__yearly_generators og
JOIN
    gen_region g on og.plant_id_eia = g.plant_id_eia
WHERE
    og.technology_description = 'Conventional Steam Coal'
    AND
    og.report_date == '2022-01-01'
    AND
    og.operational_status_code = 'OP'
GROUP BY 
    g.region
ORDER BY
    fraction_gen_null_fuel_cost

[zaneselvans]

The FERC fuel data is quite messy. I would try and avoid that as a source if you can — they use freeform strings for the fuel types, and we have to clean them up using our best-guess. I haven’t checked how well the bulk values correspond between FERC & EIA lately.
The EIA fuel costs are based on individual deliveries from the Fuel Receipts & Costs table in EIA-923. There’s an unknown lag between the fuel being delivered and consumed. But they provide a lot of detailed information, including the energy source code, fuel heat content per unit, units delivered, where it came form what the transportation mode was, etc.

Unfortunately, a number of plants redact the price they paid for a delivery, and in those cases we fill the price from the aggregate values for the month & state & fuel-type published in the EIA API data. This still leaves some gaps, particularly for natural gas in the northeast.

I suspect that you will get higher coverage if you look at the monthly data, since some of the redacted values are seasonal.

[zaneselvans]

From @gschivley

Oh, interesting to know about the seasonal data. If I tried to assemble it on a monthly basis do you think costs would correspond to the month of generation, or would the lag potentially introduce weird effects? GenX can use a fuel price that varies over time.

[zaneselvans]

I wish I knew how to interpret the delivered vs. consumed lag. I think it depends what kind of on-site storage there, which I imagine there’s often lots of for coal, but less of for gas, and there’s some fuzzy judgement call to be made too — like if you’re receiving new fuel at price X and you had a bunch on hand already that you bought at price Y, and then you’re burning some of it… what’s the right price to associate with it? The one you paid before? Or the one you’re paying now to replace the stockpile?
If you want to see the most original data in PUDL, I think you’ll want to do an aggregation starting with out_eia923__fuel_receipts_costs

The monthly and yearly tables aggregate by time, plant, and fuel type, and so necessarily lose a bunch of the detail that varies across individual deliveries.

[zaneselvans]

From @gschivley

I didn't expect to get different values between the monthly and yearly tables. Is there an obvious reason why?

Here are both the monthly and yearly calculated costs. I also fixed my query so that the weighted avg fuel cost is calculated dividing only by generation from records where fuel cost is not NULL.

ROUND(SUM(og.fuel_cost_per_mmbtu * CASE WHEN og.fuel_cost_per_mmbtu IS NULL THEN 0 ELSE og.net_generation_mwh END) / SUM(CASE WHEN og.fuel_cost_per_mmbtu IS NULL THEN 0 ELSE og.net_generation_mwh END), 2) AS weighted_avg_fuel_cost

coal_costs_compare_yearly_monthly.csv

It might also be because I'm weighting by generation. Have to think this through to see if I should be weighting by consumed fuel instead...

[zaneselvans]

Hmm. We’re definitely not weighting by generation. For the aggregated pricing IIRC we’re weighting by heat content (since we’re primarily display $/MMBTU)