Today, when you undertake a soil carbon trading program, soil carbon quantification can be achieved through measurement, modelling or a combination of both.
The measurement method refers to direct measurement of soil organic carbon by either extracting soil cores from the ground and having a lab measure the quantity of carbon within each core, or by using sensors pushed into the ground to measure soil organic carbon and density.
Soil organic carbon modelling refers to the use of algorithms, such as machine learning algorithms, to model the relationship between soil carbon related data sources, such as farm management strategies or satellite remote sensing, and an area’s soil organic carbon content. The resulting model is then used to predict carbon content using related data, such as remote sensing, as inputs. Models can be used in a variety of applications, including generating carbon distribution maps, quantifying carbon for a certain area at farm scale or examining a soil core using spectroscopy to bypass lab analysis.
For the purposes of this post we are primarily discussing modelling for quantification of soil organic carbon at farm scale in comparison to direct measurement of soil organic carbon through lab analysis or spectroscopy.
A rising perception in the carbon farming industry is that the modelling method results in better unit economics than the measurement method, and is hence more profitable for the carbon farmer.
At first glance, soil carbon modelling does seem quite attractive; it utilises technologies such as remote sensing to model soil carbon without sampling the soil with the modelling improving over time as more measured data is collected when the model is refined with ground truth data.
What’s important to understand is that knowledge of soil organic carbon, its changes and opportunities for gains and losses under various climate and land management strategies, is currently limited and still in its early days. To add to this complexity, the variability created by farm management practices, local weather, and soil response to these factors (i.e. macroporosity) also lacks high precision data to support modelled quantification of soil carbon stock of low variance. There are also still many regions without accurate baseline data, and estimates of soil carbon stock changes are, and will remain for some time, associated with large uncertainties.
Modelling soil carbon is useful in predicting changes in soil organic carbon levels at a regional level. At farm scale however, it becomes more difficult to generalise a soil organic carbon model, and model development often needs further work to account for changes in climate conditions, farm activity, variations within that activity, as well as different soils found in new areas. Models are also prone to silent failure if they encounter data that isn't similar to the data they were developed on, since they are designed to learn relationships from observed data. Adding to this the difficulty of remote sensing penetrating the ground at depth and the complex ecosystems within soil, reliably modelling soil carbon becomes a complex undertaking.
Field work is required for both the measurement and the modelling method, not just for measurement.
What is not widely understood is that verification of soil carbon modelling does involve field work. Soil samplers charge by the day, whether you sample 1 – 3 samples in a day to 100cm or take 45 samples to 100cm, the cost for sampling is the same. Thus the physical cost of collecting samples for a 1000ha property is similar to the cost for collecting samples for model calibration and verification. The main difference between the two methods from a cost perspective is not field labour, but lab analysis, which is not a significant component of the cost of the quantification process.
Analysing soil cores in a lab has the advantage of operating within a well–calibrated, controlled environment, resulting in lower measurement errors. In a modelling approach, lab samples are used as calibration data and once the model is developed, new inputs (e.g. new observations from satellite–based remote sensing) are used to predict carbon content. Modelled soil organic carbon data is therefore more likely to differ significantly from true carbon content measurements.
Despite the fact that calibration sampling can help mitigate some of these issues, quantifying soil carbon at farm scale without well–considered and measurement–backed guardrails can result in significant variance (error), negatively impacting the integrity and confidence of any sequestered soil carbon.
Measurement vs modelling in the context of trading carbon as a commodity.
The commodity market values and trades a commodity based on its quality and mass, which are both measured. While country and regional yield predictions are often based on models, no commodity is currently sold based on modelled measurements.
Carbon data collected for farm management purposes is different from carbon data collected for trading carbon as a commodity. In the context of understanding carbon distribution across a farm for the purpose of managing it, modelling soil organic carbon is useful for understanding the soil organic carbon distribution across an area. To achieve this, a common approach is to create a soil carbon model using historical soil carbon data (typically to 150 mm). These samples are collected often randomly and sometimes composited, meaning multiple samples are combined for average values to determine soil organic carbon quantity. While this approach can generate soil carbon distribution maps, using composited samples from very different areas can have undesirable impacts when calculating population statistics. This results in larger variances when quantifying carbon.
Variance, or margin of error, refers to the percentage point that a derived value differs from the real value. It is exactly this difference between a measured/modelled value and the actual value that vastly impacts the amount of carbon you derive from your carbon farming project.
When you’re building soil carbon on a farm for the sake of trading it, the lower your variance, the higher your carbon credit returns.
When it comes to trading carbon from a carbon farming project, it’s important to understand that you can only trade what’s been measured.
Even if you’ve modelled the soil organic carbon levels of your project area, when it comes to crediting carbon sequestration, you need to measure to ensure your modelled soil organic carbon predictions are in–line with reality. When trading carbon, the quantified value of the total mass of carbon on the farm derived from the first measurement or modelled estimate (plus the upper bound error) is subtracted from the quantified value of the second measurement or modelled estimate (minus the lower bound error), as shown in Figure 1.
As mentioned before, the difference in error (variance) here can be huge. The more variance there is, the more carbon you leave behind in your calculations and crediting period. This is the key difference between measured and modelled approaches. Minimising variance in measurement or the modelled estimate is essential if you want to maximise your carbon credits from an agricultural project.
So why is the difference important when we talk about carbon market integrity?
Both modelling and measurement are covered by CER soil carbon methods. This is because when you’re running a carbon project with a big focus on carbon credits, you want to ensure you derive the most credits possible at the highest integrity from your project area. This is why all traded commodities are sold on measurement of both quality and mass.
In order to maintain integrity, methods must remain focused on measurement, with research focusing equally on reducing measurement costs and improving modelling predictability.
Carbon stock quantification is a statistical problem, and sampling point placement is an optimisation problem, both of which can be solved mathematically. In addition, as the statistical variance of carbon stocks directly affects the amount of carbon available for trading, optimising sampling points to achieve a low statistical variance maximises returns for carbon project stakeholders.
To uphold carbon market integrity and maximise returns for carbon project developers, we at Carbon Count have developed a software platform that enables users to design and manage carbon projects, minimising variance and optimising their carbon credit returns.
Our team is looking forward to speaking to you about how we can help you design better carbon projects.
Thank you! Click the button below to view the whitepaper.