Duke’s Kyle Bradbury, PhD, on the need for building emissions data - Climate TRACE

Can you tell me a little about the emissions sector you focus on, and why it's important to get visibility within it?

A collaboration between Duke and Dr. Jordan Malof’s team at the University of Missouri, our team focuses on emissions from the building sector. Buildings contribute anywhere from 6% to 9% of global emissions when looking at direct emissions, and far more indirectly when you factor in electricity and biomass (i.e., any solid fuels that are derived from plants and animals that are intentionally burned by humans for household energy).

We're focusing on direct emissions from both residential and non-residential buildings, including retail spaces, office spaces, hotels, warehouses, and institutional buildings like hospitals and schools. Direct building emissions primarily come from onsite fossil fuel combustion for space heating, water heating, and cooking. We're not including emissions from lighting, consumer electronics, or most air conditioning, as these are typically electric and accounted for elsewhere within Climate TRACE.

Our goal is to provide building emissions data at the neighborhood level, using 1-kilometer-by-1-kilometer grid cells. For reference, that's about 45 Manhattan city blocks. This is a much higher spatial resolution (~100x) compared to the current highest spatial resolution of emissions estimates in this sector, which is about 10-by-10-kilometer grid cells. With the lower resolution you can only get about four or five data points for all of Manhattan — compared with 450 from our model. 

Why has it been so hard to get good, high-resolution data for this sector, and what is your approach or methodology?

One of the biggest challenges is the sheer scale and diversity of buildings across the globe. If you think about building emissions, we're talking about every building across the 150 million-square-kilometers of land on Earth. Each of those buildings has different end uses, fuel types, and energy consumption patterns that make it very challenging to analyze. The only true direct measurement of combustion from fossil fuels consumed in the world’s buildings would be from the data housed in the myriad of energy companies that provide the fuel globally, so that’s a non-starter.

What we can observe, though — and only recently could assess — is the presence and size of buildings at a global scale. Our approach leverages satellite imagery products like the Global Human Settlement Layer from the Copernicus program, the Earth Observation component of the European Union’s space program. The data product provides estimates of building volume and area data from Landsat and Sentinel 2 satellite imagery, which is gathered at a 10-meter by 10-meter spatial resolution with a revisit frequency of every five to ten days.

We use these building data to estimate the energy consumed in each region of the world, and then super-resolve the 10-by-10-kilometer EDGAR v8 data product to a resolution of 1 square kilometer. We then break down the annual energy consumption for residential and non-residential buildings into quarterly estimates using heating and cooling degree day data.

Where are you in the process?

We've been able to cover the entire globe with our 1-by-1-kilometer grid analysis, by using higher spatial resolution data to disaggregate the EDGAR v8 emissions data. High spatial resolution is important so that these data can be used not only for national-level inventories, but also subnational and local emissions inventories as well. But it's important to note that the accuracy of our estimates relies heavily on available estimates of building volume and their energy use intensity value (i.e., how much energy is consumed per unit area of building). When combined, building area and energy use produce an estimate of energy consumption — the key activity data that allows us to disaggregate EDGAR emissions data spatially. Currently we rely on sparse regional statistics of energy use intensity of a given region, so while we have global coverage, the quality of our estimates may vary depending on the availability and accuracy of the underlying data for each region.

In our next steps, we're excited to estimate energy use intensity more accurately by using machine learning techniques to factor in weather data, economic factors, and regional preferences and norms.

We'll also provide quarterly emissions estimates to reflect seasonal changes in energy consumption.

What specific challenges has your team faced, and how has your approach evolved over time?

One of our main challenges has been the limited availability of ground truth data that we can trust, especially in terms of geographic coverage. What data do exist are typically at a national scale and very little information is available at subnational levels. These data limitations have led to creative problem solving for how we evaluate our data quality, and we have been exploring ways of not just evaluating the final product — our emissions estimates — but also quantifying the accuracy of each intermediate data product, including building floor area and energy. This has led us to adopt a Bayesian mindset for our work. We start with a reasonable set of prior assumptions and think iteratively about how we can improve each component of our modeling framework over time. When you have limited data, you need to make the best use of it.

Have any findings surprised you about emissions from this sector?

One striking thing I’ve noticed is the dramatic difference in emissions between the largest metropolitan areas compared to more rural regions. The top few megacities in a given country can dominate in terms of overall emissions from buildings in that region. Seeing this data visualized on maps really puts that into perspective.

In the next evolution of this work we're interested in exploring the differences between urban and rural regions in terms of energy intensity. Currently, we assume a uniform average level of use across a given region, but we're looking to factor in differences in property type, needs, and fuel choices between urban and rural areas. 

How do you envision these data informing policy, target setting, and strategies? For whom?

There are two main ways we see our data being useful. First and most immediately, our building data fills in a key component of the wider context of emissions inventories. By ensuring this sector is covered, we help Climate TRACE provide a complete emissions inventory for any country, province, county, or municipality on Earth. With buildings included, this de facto emissions inventory can shed important light on pathways for targeting the highest sources of emissions for mitigation activities. In this way, the buildings dataset has potential to assist in the mission of organizations and initiatives like C40 and WEF’s Net Zero Carbon Cities, and to support cities in developing climate action plans and city-focused emissions reduction strategies, like New York City's Local Law 97 or the EU’s Energy Performance of Buildings Directive.

In combination with data from other Climate TRACE sectors, like forestry and land use, the buildings dataset could potentially detect effects of longer-term land use changes, like rapid development of natural or agricultural land.

Down the road as we refine these approaches and novel data sources emerge, we also hope future versions of these data can be used regionally to help with monitoring progress towards emissions reduction targets and building efficiency retrofit projects. With more detailed spatial resolution data around buildings and neighborhoods, leaders around the world could identify communities that would benefit significantly from energy efficiency retrofits. We're not there yet, but I’m excited about this possibility for the future. 

After all, every step toward lowering our collective emissions helps.

Interviewed by Daisy Simmons.