MBERs methodology test

Marginal Build Emissions Rates (MBERs) methodology overview

The first known harmonized MBERs dataset with global coverage was developed by the UNFCCC using a methodology that follows the GHGP Guidelines. (That UNFCCC dataset is available here.) This dataset extends the same methodology to higher spatial and temporal granularity.

The core idea of the algorithm is to use the weighted average emissions rate of the last 20% of power plants that a power grid actually did build as an observable proxy for what type of power plants a grid is most likely build next in response to any increases in net demand.

Specifically that algorithm is to:

Assemble a dataset of all power plants in a selected grid for a given operating period (hourly or annual). This dataset must include unit start year, capacity, generation or capacity factor, and CO₂ emissions or emissions factor.
For each cohort of units with the same operation and start year in the grid and time period: sum the generation and CO₂ emissions.
Ordering units from newest to oldest start year, determine the minimum number of consecutive of years that represents at least 20% of the total grid generation and during which at least 5 units began operating.
For these years, sum the generation and the emissions and calculate the MBER as the emissions divided by generation.

For greater detail of this process, see the two examples below, which demonstrate the full calculation of both annual and hourly MBERs for the CISO grid in the USA.

Key data sources for each model input:

The unit inventory of all power plants is created by combining data from Global Energy Monitor’s Global Integrated Power Tracker with the US Energy Information Administration's (EIA) Preliminary Monthly Electric Generator Inventory.
Annual estimates of country demand and fuel specific annual capacity and generation by fuel type were derived from the EIA's International Electricity dataset and EMBER’s Yearly Electricity Data. Additional plant specific annual generation for the US was sourced from EIA’s EIA-923.
Data to disaggregate the annual generation to hourly globally was predicted with internal demand and fuel specific generation models that use the above estimates combined with European Centre for Medium-Range Weather Forecasts’ ERA5 weather datasets as inputs. Power grid specific hourly generation by fuel type from EIA’s Hourly Electric Grid Monitor were used where available for US grids hourly disaggregation.
The unit specific carbon intensity was predicted with the Climate TRACE Power sector’s carbon intensity model.

The balancing authority country 3 letter alpha codes are taken from Climate TRACE and are based on ISO 3166. In the US, the balancing authority codes are a “USA-” prefix followed by the code described in the EIA Hourly Electric Grid Monitor’s List of Balancing Authorities Note: Calculations currently exclude energy storage and pumped hydro.

Example calculations

Example 1. Annual MBERs for CISO balancing authority

Following the steps above, one can calculate the annual MBER value for the CISO BA, in the United States, which covers most of California:

1. Assemble a dataset of the operating units of power plants in a selected grid for a given operating period (hourly or annual).

Table 2. For subsequent calculations, one needs the following for each unit in the grid: operation start year (start_year in table above), capacity (capacity_mw), capacity factor (cf), and carbon intensity (ci), from which one can calculate the unit generation (mwh) and unit co2 emissions (co2).

2. For each cohort of units with the same operation and start year in the grid and time period: count the units and sum the generation and CO₂ emissions.

Table 3. For each start year, cumulative count (cnt), cumulative sum of generation (mw) and cumulative sum of co2 emissions (co2) are calculated.

3. Ordering units from newest to oldest start year, determine the minimum number of consecutive of years that represents at least 20% of the total grid generation and during which at least 5 units began operating.

Figure 1. Overall CISO grid generation, by unit type and year. The bars represent the fraction of power generation relative to the entire CISO grid for units that began operation during that year (left y-axis), colored based on the fuel type. The black line represents the cumulative fraction of grid generation represented by units that began for each year beginning with the most recent year (right y-axis). Finally, the red horizontal line represents the point in which 20% of cumulative grid generation was reached moving backward in time from the present, which corresponds roughly to 2015. Therefore, the MBERs would be calculated using units that began during the years 2015 to present.

For this particular example, the MBER is 86.6 kg CO₂ / MWh (see below for full calculation). Expanding this time series backward for the same grid, we observe the transition of from non-renewables toward renewables in recent years:

Figure 2. As Figure 1, with expanded time series to demonstrate the grid evolution.

We can also demonstrate this calculation in tabular form:

Table 4. In alignment with the previous two figures, the cutoff (red line) displays the operation start year for which the cumulative generation (cum_mwh_frac) reaches >20%, and includes at least five units.

4. For these years, sum the generation and the emissions and calculate the MBER as the emissions divided by generation.

Table 5. Calculations of the annual CISO grid MBER (mber_kg_per_mwh) using the cumulative generation (cum_mwh) and emissions (cum_co2).

Example 2. Hourly MBERs for CISO balancing authority

A weakness of annual MBERs is that they cannot measure structural change related to timing. This limits their utility in analyses of the effects of load shifting, load shaping, and hourly matching. Thus, Climate TRACE developed what we believe to be the first dataset applying the default GHGP Guidelines for MBERs to the hourly level. An example of this is highlighted below for the CISO comparing MBERs for a typical daytime and nighttime hour.

Daytime CISO MBERs:

Figure 3. As Figure 1, but specifically for the daytime hour of 19:00UTC (11:00 local time).

CISO has seen such large deployment of solar in the last few years that during a typical daytime hour, the last 20% of generation built is almost entirely solar. The red box captures the last 20% of generation. The model is effectively estimating that any significant increase in daytime net demand in California in 2025 would likely largely cause more solar build. Following the steps above using only the hour of 19:00UTC, this leads to a MBER of 25.0 kg CO₂ / MWh.

However, California is well known for the infamous "duck curve" reflecting the fact that the grid's increasing dependence on solar is leading to different power conditions at different times of day. So, here we contrast the hourly MBER for a typical nighttime hour in CISO:

Nighttime MBERs:

Figure 5. As Figure 3, but during the nighttime hour of 7:00UTC (23:00 local time).

In contrast to the daytime (Figures 3 & 4), the generation during the nighttime comes in large part from natural gas-powered units, many of which date back as far as 2010. While some carbon-free energy is available at night, largely from wind, the end result is a higher MBER than in the daytime. Specifically, a MBER of 228 kg CO₂ / MWh is calculated.

Conclusions:

As demonstrated above, there is significant diurnal variability in the MBER for individual grids, with roughly an order of magnitude difference in the MBER at nighttime compared to daytime. These results stress the importance of considering the causal effect of timing of electricity load, generation, and storage on inducing structural change in power grids.

Because of rapid growth in solar in many grids worldwide, many grids similarly exhibit lower hourly MBERs during daytime than nighttime hours. However, in some power grids other temporal profiles exist; for example, in grids like SPP where wind build has outpaced solar build, low MBERs are driven by patterns in wind speed rather than sun.

Current known model limitations and future plans

Biomass is currently treating as 0 emissions to measure life cycle, not direct emissions. Future versions will examine whether this is appropriate and update accordingly.

For grids totalling 2% of global electricity demand, MBERs are not yet hourly. Future work will expand this.
The GHGP Guidelines indicate one should aggregate in most cases to the area where the local authority has dispatch control (often known as the balancing authority). While for most countries there is one national grid operator, there are countries with multiple grid operators active (i.e. CHN, USA, CAN, IND, RUS, and possibly VNM and BRA). For this version we were able to model USA balancing authorities. Future work will also separate out national MBERs for these other countries.
The GHGP Guidelines also indicated that for grids with very high imports or exports as a share of native generation, it may be appropriate to combine two or more balancing authorities. (Note the GHGP Guidelines do not call for flow tracing.) This is not currently implemented in the first MBER version. Future versions will explore this and implement as appropriate. In the meantime, BAs with exceptionally high net imports or net exports have been excluded. This includes: PRY, LAO, USA-GRID, USA-SEC, USA-WWA, USA-SPA, USA-YAD, USA-GRIF, USA-CHPD, USA-AVRN, USA-GWA, USA-SEPA, USA-BPAT, USA-HGMA, and USA-DEAA.
The following BAs have significantly different fossil fuel capacities according to different sources, and have been excluded pending further analysis: AFG, CAF, ESH, GLP, GUF, IMN, KHM, LBN, MDA, MTQ, NGA, PSE, REU, SGN, SLE, URY.
There are a group of very small countries where there is not sufficient data on annual demand or generation to make MBER estimates. These are: AIA, ALA, AND, ATA, ATF, BES, BLM, BVT, CCK, CUW, CXR, ESH, FSM, GGY, GLP, GUF, HMD, IMN, IOT, JEY, LIE, MAF, MCO, MHL, MNP, MTQ, MYT, NFK, PCN, PLW, REU, SCG, SGS, SJM, SMR, SXM, TKL, TUV, UMI, VAT, and WLF