High-resolution carbon accounting: modelling better emissions estimates

High resolution carbon accounting for grid imports

This is the second blog post in a series about high-resolution carbon accounting. In our first blog post we introduced the concept and its background. This post focuses on the data processing and modelling required for high-resolution carbon accounting. We also present preliminary results from analysis on data for a full year.

In a nutshell, high-resolution carbon accounting calculates the emissions intensity of the grid over small time intervals. It then uses this high-resolution data to calculate emissions from grid imports. This is in contrast to the current practice of estimating emissions based on an annual average value for the emissions intensity of the grid. Because emissions intensities for the grid are not constant, high-resolution carbon accounting promises to be a more accurate method for estimating carbon emissions. 

Data processing flow

High-resolution carbon accounting requires a data model  combining a number of data sources and data processing steps. The diagram below outlines our data processing flow.

Figure 1: Data processing flow for high resolution carbon accounting
Figure 1: Data processing flow for high-resolution carbon accounting

In the diagram we see the following processing steps:

  1. We start out by calculating the emissions intensity per fuel technology—e.g. brown coal, black coal, rooftop solar, etc. We do this using the electricity sector emissions and generation data 2020-21 published by the Clean Energy Regulator. 
  1. We then combine these per fuel technology intensities with generation data per fuel technology from the openNEM API. This results in the emissions intensity for in-state electricity generation per 30 minute interval for each NEM state.
  1. The next processing step takes energy imports from other states into account. This results in the overall emissions intensity per 30 minute interval for each NEM state*. 
  1. We then combine this data with interval data of grid imports measured by a Wattwatchers device. This results in the emissions from electricity imports from the grid for a site, per 30 minute interval.
  1. The next step is to compare high-resolution carbon accounting with the current practice of using annual averages. To do this, we calculate the annual average emissions intensity per state from the full year of 30 minute intervals. 
  1. We then combine the annual average intensities with the Wattwatchers interval data. This results in the estimated emissions from grid imports based on annual average emission intensities.

We analysed a full year period, from July 1 2021 until June 30 2022, using this processing flow.

Emissions intensity analysis

Looking at the emissions intensities for the first 2 weeks of February 2022 (Figure 2, below), we observe clear day-night patterns in most states. The grid has a considerably higher emissions intensity at night than during the day. This is not surprising, as low intensity solar generation is absent at night.

Emissions intensity by NEM state in 30 min intervals over a 2 week period
Figure 2: Emissions intensity by NEM state in 30 min intervals over a 2 week period

When we look at the emissions intensity over the full year, using average weekly values to remove the noise from day-night patterns, we see very little seasonal variation (see figure 3 below). This was a bit of a surprise to us. As the solar day is shorter in winter we expected greater seasonal variability. 

Weekly average emissions intensity over a year by NEM state
Figure 3: Weekly average emissions intensity over a year by NEM state

When we look at data from daytime (sunlight) hours only, we can detect some seasonality in New South Wales and South Australia. The emissions intensity is lower in summer, when the sun is higher in the sky. We still see hardly any seasonality in Queensland. This is at least partly explained by its location closer to the equator. We also see quite different patterns in Tasmania and Victoria that are actually due to imports from other states.

Weekly emissions intensity over a year for daytime hours only (08:00 - 16:00)
Figure 4: Weekly emissions intensity over a year for daytime hours only (08:00 – 16:00)

Finally, figure 5 below shows the impact of imports from other states on emissions intensity. Imports have the largest negative impact on the emissions intensity in Tasmania and South Australia. This is not surprising given these two states have the cleanest in-state generation. In Victoria the emissions intensity is substantially lower at times, due to imports. Again this is to be expected as Victoria has the highest intensity in-state generation among NEM states.  

The impact of imports from other states on the emissions intensity
Figure 5: The impact of imports from other states on the emissions intensity.

Carbon emissions analysis

We applied the high-resolution carbon accounting model to 488 residential sites from the My Energy Marketplace project. We then compared the results against emissions estimates based on the annual average emissions intensity. The graphs below (Figure 6) show the percentage by which annual average accounting over- or under-estimates emissions for sites in different states **. 

Percentage by which annual average accounting over- or underestimates emissions for 488 sites across 4 NEM states.
Figure 6: Percentage by which annual average accounting over- or underestimates emissions for 488 sites across 4 NEM states.

We can see that for Victoria, Queensland and New South Wales annual average accounting roughly results in between 5% underestimation and 5% overestimation. For sites in South Australia the inaccuracies tend to be much larger, from 20% underestimation to around 80% overestimation for one site. The much larger inaccuracies in South Australia are caused by the high proportion of renewable energy generation in that state. The next blog post in this series will analyse the accounting inaccuracies in more detail.

Single site analysis

The graphs below represent a single site in South Australia. The top graph shows the site’s grid imports per 30 minute interval.

The middle graph shows the inaccuracy when calculating emissions using the annual average emissions intensity compared to high-resolution accounting. Positive (blue) values indicate intervals where annual average accounting overestimates emissions. Negative (red) values indicate intervals where annual average accounting underestimates emissions. The majority of intervals in this site are blue, therefore average annual accounting overestimates emissions for this site.

This is confirmed in the bottom graph, which shows the cumulative error of average annual accounting. For this site, the overall accounting difference over the full year amounts to 147.8 kg CO2-e. This represents a 16% overestimation of emissions when using annual average accounting.

Difference between high resolution and average annual carbon accounting for a single site in South Australia
Figure 7: Difference between high-resolution and average annual carbon accounting for a single site in South Australia.

Conclusions

We’ve developed a data processing flow to calculate emissions associated with grid imports in 30 minute intervals. The emissions intensity of the grid exhibits strong day/night patterns and to a lesser extent seasonal patterns. High-resolution carbon accounting incorporates these patterns. Emissions estimates are therefore more accurate than those obtained from calculations using an annual average emissions intensity.

We confirmed this by comparing the emissions estimates from high-resolution carbon accounting to estimates obtained using average annual emissions intensities for 488 residential sites over a one year period. Annual average accounting underestimated emissions up to 20% for some sites and overestimated emissions up to 80% for another.

We saw the largest estimation errors for sites in South Australia, the state with the highest percentage renewable energy generation of mainland NEM states. Accordingly, we expect estimation errors of annual average accounting to increase in the other NEM states over the coming years, as ever more renewables are added the the grid.

Next in this blog series

In the next blog posts in this series we will hone in on two specific aspects of high-resolution carbon accounting:

  • Accounting accuracy: Why does it matter? What will the impact of moving to high-resolution carbon accounting be?
  • Carbon efficiency: What is it? How does high-resolution carbon accounting give companies an additional lever to reduce carbon emissions?

Adriaan Stellingwerff is a Senior Software and Data Engineer at Wattwatchers.

* Our current modelling is based on net imports over each 30 minute interval (AEMO reports data in 5 minute intervals but does not include rooftop solar, which is only available in 30 minute intervals). We assume imported energy in any 30 minute interval originates from the neighbouring NEM states. Or in other words, the emissions intensity of the imported energy is set to that of the exporting state.

** Our data set didn’t include any sites in Tasmania, hence the absence of this state in the emissions analysis.

ARENA logo

ACKNOWLEDGEMENT

The My Energy Marketplace project is receiving funding from ARENA as part of ARENA’s Advancing Renewables Program. The views expressed herein are not necessarily the views of the Australian Government, and the Australian Government does not accept responsibility for any information or advice contained herein.