The methodology behind our latest global data expansion
This week we’re excited to announce a major, global geographic expansion of our flagship Marginal Operating Emissions Rate (MOER) carbon data signal — growing from 40 to 210 total countries and territories. You can read the full press release here. With this expansion to ~170 new geographies, WattTime now offers actionable marginal emissions data — available with 5-minute granularity and a combination of historical, real-time, and rolling 3-day forecast perspectives — for 99% of the world's electricity consumption.
Accurate, location-specific, timely, and granular MOER signals are central to a trio of solutions — carbon-aware load shifting, emissionality-based renewables siting and procurement, and supply chain decarbonization — that can save more than 9 gigatons of emissions every year. That’s equal to nearly 20% of total global carbon emissions. These solutions also accelerate the reduction of harmful air pollution that disproportionately affects the developing countries that often are last to gain access to technological solutions that improve their lives.
Adopting those three solutions at scale (and unlocking those 9+ gigatons of emissions reductions) depends on a truly global MOER signal. Which is why this week’s announcement represents an enormous step-change for what’s possible.
In this article, we’ll take a closer look at the foundational methodology driving our MOER models thus far, as well as the new methodologies that allowed our team to expand to substantially global data coverage.
The science behind WattTime’s current MOERs
When electricity demand rises or falls, or new wind or solar capacity gets built, which generators respond varies. Namely, a certain generator (or generators) ramp up or down — or turn on or off — in response to changes in load or new renewables added to the grid mix. These power plants on the edge of the dispatch stack order are what’s known as marginal generators, and their associated emissions are what’s described by our MOER signal. Sometimes polluting, fossil-fueled peaker plants might set the marginal emissions rate; at other times, surplus renewables being curtailed might be on the margin.
Scientists agree that such a marginal signal is the best way to guide (and measure the impact of) interventions such as cleaner EV charging to reduce emissions or the avoided emissions of building a new wind farm on a coal-heavy grid.
Our current-best MOER — with foundations in academic literature and iterated on by WattTime for over ten years — is based on causal, empirical modeling. In the grid regions of North America and Europe, we use robust, detailed, generator-level data inputs to refine and train our algorithms. For example, we incorporate generation and emissions data from individual power plants via the US EPA’s Continuous Emissions Monitoring System (CEMS). We get demand, interchange, and generation by fuel type data from the US Energy Information Administration (EIA) and European Network of Transmission System Operators for Electricity (ENTSO-E). We also integrate myriad other data sources into our overall modeling. For regions where these data are available, we feed it all into a binned regression model. The result is today considered WattTime's highest-quality methodology for the MOER signal type.
In regions where we have access to partial real-time and forecasted information about the status of the grid (such as demand or energy prices), but are lacking specific ground truth time series data such as emissions or generation by fuel type required to train a binned regression model, we use a proxy regression model where we use machine learning to identify a data-rich grid with similar characteristics to use as a proxy for the grid with partial data. The power plants for the region of interest are characterized by Climate TRACE and assembled into a supply curve that resembles economic dispatch for the proxy region, which is then used to estimate the MOER based on the real-time demand in the region. MOER data from this “Proxy Regression” model is available in countries such as Brazil, India, Chile, and Turkey. (Longtime WattTime partner Microsoft helped fund development of this method.)
But after deploying these higher-quality MOER signals for the grids of all countries with the necessary data, we still had not covered about 170 countries of the world. Yet many of these are the same countries where emissions are not yet falling, and it’s not good enough to just ignore them. Our latest MOER expansion fixes that in big ways, thanks to a novel modeling approach from our team.
Inside the methodology powering our global MOER expansion
With the goal of making MOER data truly global, we’ve employed a synthetic demand model for countries where we lack historical and real-time grid information. It extends and improves upon the 2021 work of Mattsson et al.
In lieu of actual real-time demand data (like those used in our proxy regression models), the Mattsson framework uses atmospheric estimations derived from ERA5 climate and weather data, along with demographic information, to model synthetic power demand with a mean absolute percentage error of only 8% when averaged by month and hour.
To model the MOER, we expanded upon Mattsson’s academic model to account for more geographic and economic diversity. This enabled us to produce estimates of real-time and forecasted demand. To productionize this, we needed to build a sophisticated weather data modeling pipeline, ingesting gigabytes of global weather data four times a day, thus producing timely synthetic demand estimates. Currently, we’re using these synthetic demand estimates as inputs to our existing proxy regression models in order to produce MOERs using a model we are calling the “Synthetic Demand Proxy Regression” model, or “Synthetic Proxy” for short.
While the MOER signals resulting from this method are inherently less accurate than those that we can derive using the binned regression model, they’re nevertheless useful for both carbon-aware load shifting and renewables siting — which unlocks significant emissions-reduction potential. Of course, data transparency will only increase in regions across the globe. As it does, we will vigilantly be upgrading our MOER signals according to higher-quality binned regression methodologies.
Putting global MOER data into action
The MOER signals generated via our synthetic proxy regression models unlock potential to enable much more strategic emissions reductions decisions and automations, and in regions that have never before enjoyed access to such actionable data. Historically, granular and accessible emissions data have been far more accessible for countries in the Global North. This release is a major step in closing this gap, and opens up opportunities to drive emissions reductions and new solution possibilities across the Global South.
All WattTime data can be accessed through our API. Basic access is available for free to all users. Partners on our Pro data plan get full access to historical and forecast data, including premium support. Our partners with a global data license will automatically gain access to these new grid regions. Contact our team to learn more.
