How EOSDA Monitors Sequestrated Carbon With AI And ML

How Does The Model Actually Work?

Oleksii Neskorovnyi: We developed a solution for soil organic carbon assessment with two machine-learning models to improve it over time. The first model evaluates the spatial distribution of soil organic carbon in terms of land cover using physical soil samples as a basis, and it is used as the baseline needed to claim carbon credits. The second one predicts changes in soil carbon content over time. It is used to let agribusinesses assess their future perspectives in terms of carbon sequestration.

Our approach is based on RothC, which is a model that characterizes the decomposition of organic carbon in topsoils not subject to waterlogging, considering the influences of soil type, temperature, moisture levels, and vegetation cover on this processprocess ^{Rothhamsted Carbon Model (RothC): Understanding Soil Carbon Dynamics.}. To ensure more precision and accuracy, we complement this model with a number of supplementary parameters, including topography characteristics, ecological factors, and soil management techniques. EOSDA also improved the RothC model by enabling carbon predictions for any time span within the typical 20-year period, unlike the open-source version limited to 20-year forecasts.

How Does The Model Benefit Our Clients?

Regina Urazmanova: The main benefit is that agribusinesses can generate a new source of income — carbon credits — while saving up to 90% on soil samples starting with the second sampling. Another advantage is that the model simplifies the client’s work. The lab analyzes the soil samples, and the client sends us the results. We then process them to calculate current and forecasted carbon levels. At the end of the process, the client gets a report or a raster with the carbon content numbers.

Oleksii Neskorovnyi: We use high-resolution images and open-source data from Landsat, Sentinel, and several other satellites for our analytics. We can use a ten-meter per pixel resolution for modeling. The data, such as the terrain peculiarities and input data, significantly impacts the outcomes since land cover and relief influence the carbon level and can change the model requirements. We also provide important data for precise farming management, starting with zoning and weather reports and finishing with soil moisture analytics. Some of that information is also needed for the SOC model. EOS Data Analytics prides itself on accurate and timely reports with quality information for our clients.

Regina Urazmanova: As for the benefits in terms of reducing the effects of climate change, our model allows us to evaluate whether regenerative agriculture practices are working properly. Soil conservation and other diverse practices are aimed at restoring ecosystems and mitigating the effects of greenhouse gases on climate. We should not forget about this initial benefit.

What Data Do We Need To Calculate Carbon Sequestration?

Regina Urazmanova: For the model to work correctly, we use a comprehensive dataset and up to 140 other predictors that we can use with it. Those include climatic data, satellite optical and SAR data, DEM (digital elevation model) and its derived parameters, and soil information.

The dataset should encompass weather conditions, soil and relief characteristics, vegetation indices, and various satellite-derived data, including land cover information. Since we are a company that works with satellite imagery, we rely heavily on our own findings and reports, and most of the data comes from us. Client data is also very important for modeling, as actual soil data makes the model more accurate.

And If An Agribusiness Wants To Start The Process Of Getting Carbon Credits, Where Do They Begin?

Oleksii Neskorovnyi: Some agribusinesses take upon the role of carbon project developers as they begin the process of carbon sequestration, following entering the carbon credit market. There are also companies with the same name, carbon project developers, that specialize in carbon sequestration practices. Sampling the soil on the land for several years should be the first step, and historical data, such as samples from previous years, works just fine for that purpose. It’s important for a carbon project developer to understand that this process can be quite lengthy, so preparation for the time commitment involved is essential. The soil samples are then sent to the lab, which needs to analyze them, as well as the amount of carbon and, as Regina previously mentioned, the amount of clay. Next, the data must be collected from companies like EOSDA, which provide SOC monitoring services.

Our team then starts the process of soil carbon modeling. We need to search and download or process initial datasets for the weather data, soil type and composition, terrain model, lаnd cover or land use data, crop data, soil samples, etc. In this process, we use open data sources, government data, EOSDA Crop Monitoring platform records of the client in question, and information provided by them. We can start with the proof of concept from 1,000 hectares or 2,470 acres of land size. The next step is to process the data, adjust the model for the territory where proof of concept takes place, and calculate soil organic carbon.

Regina Urazmanova: Then we move on to calculating the spatial distribution of SOC and validating the results against the soil samples. This stage is exciting because it’s where some of the most critical work happens, including model calibrations to ensure our outcomes are as accurate as possible. Optionally, for the clients that require predictions and want to save up to 90% on the next soil samples, we can refine our analysis by adjusting the content based on predictors and spatial distribution using the RothC model. That means we can compare the initial soil sample analysis to future predictions of soil organic carbon to, say, estimate potential carbon credits for the client or to assess whether the carbon sequestration practices are working.

It’s this phase of tailoring and validating our predictions that brings the project to life, offering tangible data that can impact real-world decisions. We then need to validate the results in the selected period. The final step is to prepare the report.

You Mentioned Soil Samples. How Exactly Does EOSDA Analyze Soil, And Was The Model Ever Tested?

Oleksii Neskorovnyi: Firstly, the local lab, which has much more localized knowledge on both how and when to take the samples and how to assess them properly, will complete the analysis. I would like to emphasize that agribusinesses do not need to send the samples to EOS Data Analytics. We are working with ready-to-use reports from such institutions.

Secondly, the soil samples our model requires are with clay content. Clay has a high surface area and is negatively charged, allowing it to attract and hold water and nutrients. This property influences soil moisture levels, affecting microbial activity and organic matter decomposition, both of which are central to SOC dynamics. Also, clay plays a role in protecting organic matter from decomposition through processes such as physical protection and chemical stabilization. This stability affects the turnover rate of organic matter and ultimately impacts SOC levels. For this reason, we closely look at the clay content in soil samples. The minimum required amount of soil samples has to be one sample per four to five hectares (or nine to twelve acres) for the utmost accuracy. The samples must include the Soil Organic Carbon, clay content, bulk density data, and georeferencing.

Regina Urazmanova: And here’s how EOSDA uses the soil data in the SOC model. We model the organic matter in the soil sample based on the soil analysis. The output from the SOC model provides a broader SOC calculation than a soil sample laboratory report. Our analysis incorporates additional factors such as climate, weather, and other variables, enhancing the accuracy of mapping carbon distribution.

Oleksii Neskorovnyi: So far, we have tested the model in Australia with AgriProve, the fastest growing carbon soil tech developer in Australia and our partner since 2022. We collaborate on a two-year project funded by the Australian Government with a $9.2 million grant. Under this project, we are developing a technology to harness various indices calculated from optical and SAR satellite images. We also use a comprehensive array of predictors, i.e. variables derived from various sources, including digital elevation maps and their derivatives, as well as bioclimatic indicators and soil characteristics.

The labs’ organic carbon analysis is used for direct soil organic content calculation or integrated into the RothC model.

The project aims to establish an agri-led commercial design of soil organic farming, regenerating agriculture, and sequestering carbon at scale. So, the Carbon Project enables agribusinesses to generate income from carbon abatement activities and secure sustainable returns in the Australian market.

What Kinds of Soil Can Sequester Carbon?

Regina Urazmanova: Various soil types sequester carbon differently. Recent studies suggest that brown earth and meadow soils, for instance, have significantly higher carbon storage capabilitiescapabilities ^{Balasubramanian, D., Zhou, W.-J., Ji, H.-L., Grace, J., Bai, X.-L., Song, Q.-H., Liu, Y.-T., Sha, L.-Q., Fei, X.-H., Zhang, X., Zhao, J.-B., Zhao, J.-F., Tan, Z.-H., & Zhang, Y.-P. (2020). Environmental and management controls of soil carbon storage in grasslands of southwestern China. Journal of Environmental Management, 254, 109810. https://doi.org/10.1016/j.jenvman.2019.109810.}. Clay in the soil is what matters most for SOC modeling. That being said, soil type is only one of the factors impacting CO2 capture. The typographical setting, climate, and weather also play significant roles. The most important here is definitely the work done in the field.

What Other Services Do You Provide That Help Agribusinesses With Their Carbon Projects?

Oleksii Neskorovnyi: Organic carbon requires several conditions to be stored in the soil. One of those conditions is that perennial crops sequester more CO2 than annual ones. Cover crops and abandonment of tillage practices are also among those conditions. Agribusinesses take care of their soil in numerous ways, and having carbon credits as a result is a nice bonusbonus ^{Bedord, L. (2023, April 20). The truth about soil’s ability to sequester carbon. Successful Farming.}. EOS Data Analytics is here to provide additional services that can enhance soil health and, as a result, enrich it with carbon.

We offer two custom solutions to enhance the SOC modeling services we provide to our clients, field boundary detection and crop type classification. Let’s dwell a bit more on each. Field boundaries help us define the field size and determine its boundaries, which saves our clients tremendous time when providing us with data on the carbon monitoring areas. That custom solution can be used if the client cannot provide the boundaries in a digital format or when the area is so large it is easier to classify it from space. Crop classification is needed as different crops have diverse carbon sequestration abilities. When the territory is large enough, manually classifying crops can take a long time, and this is where EOSDA’s custom solution helps to reduce said time. Crop rotation is a sequestering practice, and historical crops matter, too. Our solution can also identify the presence of cover crops. Our custom solutions simplify and automate the significant amount of work that the carbon project developer is docking with throughout the project.

Except for those two major requirements, we can help agribusinesses with additional services that can ensure their goals of carbon sequestration are met. Yield (biomass) estimation, as a part of the yield prediction custom solution, is one of them. This solution allows us to predict the estimated yield on the client’s field, as the estimation is grounded in historical data trends. Another tool is tillage detection. The no-tillage method saves the earth from erosion and, consequently, prevents carbon from escaping. Our remote sensing capabilities help carbon project developers save time by allowing them to remotely verify that soil tillage standards are maintained, eliminating the need for frequent field visits.

The final carbon-related feature in Soil Organic Carbon modeling that is probably of most interest to carbon project developers is reporting. The carbon project encompasses a wealth of information, comprising historical field data on crops, fertilizer usage, regenerative farming practices, SOC sequestration, including initial baseline, and more. Reporting serves as the most straightforward method of transmitting project information to stakeholders. Our reports are completely customizable for the client’s needs.

Great! What Other Benefits Do You Plan To Share With The Clients? What Are The Plans For SOC Solution Development?

Regina Urazmanova: We are currently receiving numerous requests from potential clients for new projects. There is a considerable demand for soil organic carbon modeling. Request processing and communication with clients give us further and deeper insight into their needs and how we can help. We plan to refine our products and solutions according to individual clients’ needs in various markets so far.

We realized that we have a bottleneck in that not every client has soil samples or the capacity to extract them before starting the carbon project. To solve that problem, our team came up with the idea of finding local partners in diverse regions. This allows us to ensure data quality, increase production speed, and gather data in a format that perfectly suits our SOC model. For that, we invite any soil laboratory interested in cooperation to contact us at sales@eosda.com.