AI for Carbon Farming: Climate Change Mitigation

Carbon farming is an emerging trend in agriculture. Hence, if you know what it is - skip the next two paragraphs and go straight to the case studies about estimation of carbon soil organic carbon content. And today's hero is Machine Learning (ML).

What is Carbon Farming?

Carbon farming involves using specific agricultural practices to capture and store carbon dioxide from the atmosphere in soil and plants. This helps improve soil health, increase crop yields, and combat climate change by reducing greenhouse gases. As you know, globally, humanity is producing more carbon dioxide than is recommended, which is one of the reasons why we have global warming. Carbon farming is one way farmers can help to slow the rising temperature on the planet, preserve soil health, and get better outcomes from their lands. Examples of carbon farming include planting cover crops (e.g., barley, cereal rye, crimson clover, oilseed radish, winter wheat and others), practicing no-till farming (growing crops or pasture without disturbing the soil through tillage), and managing livestock grazing (allowing livestock to directly consume the vegetation).

How to Make Money from Carbon Farming?

You will be paid by selling carbon credits. Carbon credits are like earning points for cleaning the environment. You follow regenerative agriculture practices and earn points, which you can then sell to companies that need them because they produce a lot of CO2, and your land is sequestering (in other words, absorbing) it.

The initial step to start carbon farming practices is a baseline assessment of your soil's organic carbon content (SOC, a well-known abbreviation among carbon farmers).

This is where we would like to demonstrate the power of artificial intelligence to help farmers.

ML-based Soil Carbon Stock Assessment Protocol

Recent research at Wageningen University & Research ????, is focused on importance of increasing soil organic matter to mitigate climate change by sequestering atmospheric carbon. However, this requires robust, affordable, and scalable methods for soil carbon stock assessment to support carbon farming initiatives all around the world.

The offered innovation - Wageningen Soil Carbon STOck pRotocol (SoilCASTOR) integrates satellite data, direct proximal sensing-based soil measurements, and machine learning to estimate soil carbon stocks. Key steps of this method include: (1) the selection of spatial covariates from (satellite) data sources, (2) the selection of sampling locations, (3) the soil spectroscopy measurement of SOC in the field, (4) the training of a model on the available data, and (5) the calculation of the carbon stock with uncertainty estimates.

The brilliance of this method is that it achieves high precision in carbon stock estimates, with errors below 5%, enabling the detection of SOC changes desired for the 4 per 1000 initiative. It requires as few as 0.5 samples per hectare for farms ranging from 20 to 150 hectares. Also, it has been tested on different soil types in the USA, showing its versatility and applicability to various agricultural settings. The use of machine learning makes this method robust and scalable.

Finally, the method is compatible with carbon credit certification requirements, emphasizing the importance of quantifying errors and uncertainties in carbon stock estimates to meet certification protocols.

ML-powered and MIR-predicted SOC values vs. laboratory-measured SOC values

Another interesting research was conducted by Australian-led ???? consortium to explore the application and advantages of integrating remote sensing, machine learning, and mid-infrared spectroscopy (MIR) for estimating soil organic carbon. It highlights the practical benefits of this combined approach for various applications, such as comprehensive soil health mapping and carbon credit assessment. These advanced technologies promise to reduce costs and resource use while enhancing the precision of SOC estimation.

A comparative analysis between MIR-predicted SOC values and laboratory-measured SOC values using 36 soil samples demonstrated a strong fit (R2 = 0.83), indicating the potential effectiveness of this integrated method. Despite the limited sample size, these initial findings are promising and form a foundation for future research.

The paper also discusses the potential commercialization of these technologies in Australia to help farmers benefit from carbon markets. Finally, provided comparison underscores the potential accuracy and reliability of using MIR combined with machine learning for SOC estimation.

AI against traditions in predicting soil organic carbon

As a cherry on top, we invite you to read the research titled "Learning vs. understanding: When does artificial intelligence outperform process-based modeling in soil organic carbon prediction?" published by Austrian ???? scientists. It compares the effectiveness of machine learning algorithms (Random Forest, Support Vector Machines (SVM) with linear and polynomial kernels, and Gaussian Process regression with a linear kernel) and traditional process-based models (RothC, ICBM, AMG.v2, and C-TOOL) in predicting soil organic carbon.

Results vary:

• Process-based models outperformed ML algorithms, highlighting the latter's dependency on comprehensive datasets.
• ML models, especially Random Forest and SVM with polynomial kernel, showed increased errors (RMSE) with reduced data availability.
• A combination of ML and process-based models provides the most accurate and robust SOC predictions.

What's next for AI in carbon farming?

As you can see from the case studies, we are slowly but confidently moving towards using machine learning more frequently. In the next edition, we will look closer at how farmers can monitor their soil, which AI-powered tools are suitable for this purpose, and what other innovations in soil analysis are emerging to support this new field of carbon offsetting through regenerative agriculture.

Wishes of higher SOC,

Maryna Kuzmenko, Ph.D ????, Chief Inspiration Officer at Petiole Pro Community

#carbonoffsets