Modelled vs Measured: Proof of Work for Soil Carbon Credits

By Sam Duncan

As the world shifts to a carbon-neutral future, carbon credits are becoming increasingly important in helping companies offset their emissions. Agriculture offers a major opportunity to produce these credits, especially through what I’m most excited about: soil carbon sequestration. There are some fundamental challenges when it comes to the generation of credits though, including whether we use a modelled or a measured approach. Here I discuss the advantages and disadvantage of each, specifically in terms of the ‘proof of work’ required to validate and authenticate credits produced under each approach.

Carbon Credits 101

Firstly, some basics. A carbon credit is usually the equivalent of one tonne of CO2 that is sequestered or avoided. They are calculated depending on the mechanism through which they are produced. For example, in agriculture a carbon credit could be produced by reducing methane emissions from cattle by introducing a feed supplement like Red Algae (like the Algae being grown by the team at Sea Forest). In the context of this article, we’ll be talking about the sequestration of carbon in soil. If you’re after an excellent overall summary of the Carbon Credit market, I highly recommend this Forbes article.

Generating Soil Carbon Credits

Tradeable soil carbon credits, sold to companies as offsets, are usually generated through one of two ways;

1. Using a statistical model to predict the amount sequestered (modelling)

2. By physically measuring the amount of carbon sequestered (measuring).

Modelled approaches require a landholder to enter some calculations, like recent and historic farm management practices, that are then used to calculate a suggested improvement in soil carbon credits, with the help of underlying research and data. If the landholder does what they said they were going to do, be it cover cropping or minimum tillage, they may be rewarded after a few years with carbon credits equal to the amount predicted by the model.

The challenge here is that they may have experienced an increase in soil carbon greater than what the model predicted. This is because most models are built of limited training sets, and so cannot factor into account all possible scenarios that might occur.

Measured approaches, which I liken to the ‘hammer’ of the soil carbon credit generation, require physical measurement of the soil across the landscape in order to calculate the baseline level of soil carbon. It is expensive but it is accurate, and means that a landholder is rewarded for exactly what soil carbon they increase (instead of being based on a model).

There are new ‘hybrid’ approaches are being developed, such as the one currently under review by the Australian Clean Energy Regulator in response to the government’s goal to reduce soil carbon measurement to below $3 per hectare. These hybrid approaches combine some measurement, and some modelling to significantly reduce the total cost of measurement for soil carbon projects over the long term.

The Proof of Work

So once you’ve generated your carbon credits and wish to sell them, what is being bought? This is where we’ll introduce the concept of Proof of Work. Proof is the information underlying the credit, required to provide assurance and traceability to a buyer. Unlike blockchain and cryptocurrency, where each token is mathematically proven and the proof is tied to the token itself, carbon credits are a function of various different inputs that need to be linked back to them, that signal to the buyer those credits has legitimacy (there has been much written about actually tokenising carbon credits, which I won’t be going into today, but if you’re interested in the idea check out this article.)

Instead I want to look at what data is need to prove the credits. The following are what my team and I mapped for credits generated by the modelled approach and the measured approach.

Proof of Work: Modelled Carbon Credits

The proof of work here is cheap and scaleable. Because most of the hard work has been done to build the model in the first place, it means that landholders can quickly enter into projects to generate carbon credits without having to provide much proof themselves. The buyer will simply need to know:

  • The data behind the model being used to generate the result. Including model uncertainty against the parameters used, how the model was built and evidence around it’s success measuring a specific farming system.
  • Validation of the inputs used in the model (such as farming practices) by the landholder to feed the model.
Understanding the inputs into the model is critical in the proof of work, including what data was used to generate it and what the uncertainty is.

However the downside is that if there is a flaw in the model, the entire market and the credits generated under this model may be at risk. Similar to blockchain, the ‘proof of work’ provides scalability, but needs to have significant scrutiny applied before it is widely accepted by a market. There are some promising models emerging across the industry, such as the COMET-Farm (US) and FULLCAM (AUS), but we are still in the early phases of these markets and we lack many of the training datasets underlying these models. For example, the DAYCENT model, which provides soil organic carbon stock data to the COMET-Farm model, is still undergoing development to improve it’s uncertainty, despite already being used to generate carbon credits.

Proof of Work: Measured Carbon Credits

At the other end of the spectrum, the measurement-based proof of work requires the following data for the carbon credit validation:

  • A unique ID for each soil sample
  • The results for each soil sample
  • The person who took each soil sample
  • Location of each soil sample (Latitude/Longitude)
  • Time/Date each sample was taken
  • Time/Date each sample was analysed by the lab
  • The lab or technology that undertook the soil analysis
  • The method of testing used by the lab or technology (for example, spectrographic with information on the database).

The challenge here is the sheer amount of data required to function as the proof for the credits generated. Over the course of several years, with multiple sampling rounds, the data required for the generation of carbon credits stacks up. Unless a system is used to capture all of this information, the proof of work is burdened by labour and audit costs. The upside is that once this data is verified independently and used to generate carbon credits, those credits have a high degree of trust attached to them.

Measuring soil carbon is often a laborious and expensive task, requiring physical soil sampling in the field and testing by wet lab chemistry.

Modelled vs Measured — Who wins?

I want to take a step back for a second and look at the big picture. Having a global carbon credit market is a major milestone for humanity. Regardless of the method being used, the landholders themselves are sequestering carbon, which is a major positive for the environment.

The discussion on modelling vs measuring is really an economic one, with an economic solution. The linking of a price to each method to reflect the uncertainty and the amount of effort involved in the Proof of Work will drive uptake by each and we’ve already seen this in the Australian context. Although not directly a price subsidy, the number of credits that are generated with the FULLCAM model understates the number of soil carbon credits that most project participants believe could be produced. This has resulted in landholders who wish to invest a higher upfront cost for a potentially greater number of carbon credits naturally choosing the measurement approach.

In the future, it’s probably the hybrid approach, where continuous sample collection informs a model whilst also generating carbon credits, that offers the best solution for the proof of work problem. The feeding of a model using actual measurements can reduce future sampling requirements, whilst still allowing landholders to generate much-needed carbon credits in the short term, and do so whilst avoiding the limitations of a model.

Scroll to Top