Overview of Regrow's Q3 2024 platform data enhancements
Introduction
In August 2024, Regrow Ag introduced significant updates to the Sustainability Insights (SI) dataset for the Continental United States (CONUS) and advanced our machine learning capabilities. These enhancements offer our customers greater precision in making informed Scope 3 investment decisions and reporting on Scope 3 emissions outcomes.
Over the past two years, Regrow has invested heavily in ongoing research and development, resulting in more accurate and dependable outputs. We've also expanded our ground-truth data coverage, incorporating a larger number of fields to ensure that Regrow’s solutions can reliably predict outcomes and monitor changes on the field.
What’s New:
The latest updates include the addition of 2023 data, along with updates to the 2015-2022 datasets:
- Enhanced Regrow remote sensing capabilities by improving models, better data processing capabilities, and expanded use of Sentinel-2 sensors
- New in-house crop detection model, MonitorML
- New field-boundary model, Parcel ID
- Improved cover crop detection accuracy
- Reclassified tillage practices to better align with global standards
- Made improvements to Regrow’s DNDC model that better incorporate soil nitrogen and carbon cycling to enhance emissions modeling accuracy
- Introduced 6 new intervention scenarios in the “Plan” module of Sustainability Insights
Why Are We Introducing These Changes Now:
Regrow is continually updating our science and tech to give customers industry leading on-farm data they can use to guide high ROI investments, report, and certify the outcomes of those investments to achieve their regen ag and climate goals. Our goal is to deliver the highest quality and most accurate data, enhancing scope 3 inventory reporting and regenerative agriculture planning processes. With advancements in our science capabilities and improvements to our machine learning models, we produce more accurate data at scale while continuing alignment with accounting standards such as GHGp.
Remote Sensing and Monitor Model Updates
Regrow is improving our capabilities to detect additional crops and field boundaries through our models, MonitorML and Parcel ID.
Previously, Regrow's crop detection was based on cropland data layer (CDL) data from the US Department of Agriculture (USDA). Our new crop detection model, MonitorML, is built on Regrow’s remote sensing technology.
Crop Detection
Previous Crop Detection Approach (CDL):
Historically, much of regenerative agriculture crop detection was built solely on USDA’s CDL data. While CDL data is widely available and commonly used across the industry, it has limitations, including a spatial resolution of 30 meters, which may miss smaller fields or within-year land use changes. Despite CDL’s prominent use and value for monitoring land use/land cover, there are concerns around CDLs reliability for national scale applications, particularly regarding accuracy and potential misclassifications. CDL remains a valuable dataset, but on its own does not provide a holistic overview of crop detection.
Introducing MonitorML:
MonitorML is Regrow’s in-house crop detection model, built on remote sensing technology. MonitorML is trained and validated using a combination of ground truth data collected from field surveys, grower reported data, and the USDA CDL. In certain circumstances, Regrow also uses CDL as a primary data source when satellite imagery is not applicable or available. Using this mix of public and high quality data for model training and evaluation, we are able to improve classification relative to CDL, especially in the crops most relevant to our customers. This enhancement provides better insights into practice adoption KPIs, such as cover crop and tillage. By leveraging machine learning for crop detection and advanced algorithms, Regrow can now provide:
- Increased cropland coverage
- Enhanced identification of fields with hay, improving programs involving livestock or silage
- Improved detection of additional cover crops
- Better insights into field boundaries
- Identification of non-agricultural land through Regrow’s Parcel ID model
MonitorML has been trained to better delineate haylands from permanent grasslands and pasture, providing improved coverage in the Great Plains and the South, where these cropping patterns are prevalent. Unlike CDL, which often struggles to distinguish between haylands, temporary grasslands, and permanent grasslands— MonitorML is much more consistent in tracking these acres year over year, offering stable classifications for these land cover types. This stability enables more accurate tracking of grassland conversion into row crop production, which is vital for accurate land use monitoring and tracking changes in land use over time.
As we incorporate more data inputs, we continuously refine our ML models to provide targeted insights and emissions impact data, ensuring our customers receive the most relevant and up-to-date information.
Updated CONUS crop list
|
Crop |
Previous SI Availability |
Updated SI Availability |
Notes |
|
Alfalfa |
Monitor (Report) Data Only |
Report |
|
|
Barley |
Report & Plan |
Report & Plan |
|
|
Corn |
Report & Plan |
Report & Plan |
|
|
Canola |
Report & Plan |
Report & Plan |
Now under “Rapeseed” for CONUS |
|
Cotton |
Report |
Report & Plan |
|
|
Fallow |
Report & Plan |
Report (Monitor) Data Only |
|
|
Hay |
N/A |
Report (Monitor) Data Only |
|
|
Oats |
Report & Plan |
Report & Plan |
|
|
Peas |
Report & Plan |
Report & Plan |
|
|
Potatoes |
Report & Plan |
Report & Plan |
|
|
Rice |
Report Data Only |
Report Data Only |
|
|
Rye |
Report & Plan |
Report Data Only |
|
|
Sorghum |
Report & Plan |
Report & Plan |
|
|
Soybeans |
Report & Plan |
Report & Plan |
|
|
Sugar Beets |
Report & Plan |
Report Data Only |
|
|
Sunflower |
Report & Plan |
Report Data Only |
|
|
Spring Wheat |
Report & Plan |
Report & Plan |
|
|
Winter Wheat |
Report & Plan |
Report & Plan |
|
|
Dry Beans |
Report Data Only |
N/A |
|
|
Flax |
Report Data Only |
N/A |
|
|
Peanuts |
Report Data Only |
N/A |
|
|
Popcorn |
Report Data Only |
N/A |
|
|
Pumpkin |
Report Data Only |
N/A |
|
|
Sweetcorn |
Report Data Only |
N/A |
|
|
Rapeseed |
Report Data Only |
Report & Plan |
Representing Canola numbers |
How Updated SI Monitor Data Compares to Other Datasets
Like the USDA Ag Census, SI Monitor data reports on the adoption of regenerative practices across the United States. Outside of SI, the most comprehensive source of agricultural data across CONUS is the Census of Agriculture (Ag Census) conducted every five years by the USDA National Agricultural Statistics Service (NASS). The Ag Census attempts to survey all farmers, asking about their intention to implement regenerative practices within their operations. Regrow takes a different approach, using remote sensing models to identify fields that have successfully implemented cover cropping, reduced till, or no-till practices.
Many things can influence the success of a practice in a given year, including weather, pest and weed pressure, field workability, etc. Often a grower who intends to plant a cover crop may have limited success due to poor growing conditions post planting. In the Ag Census these acres would be counted as “cover crop”, while Regrow may not if the crop fails to thrive.
Due to this difference in the goals of each program, direct comparison of Ag Census data and Regrow data can be misleading. In general, we advise customers to use the data together, with the Ag Census and Regrow reporting on different elements of regenerative practice adoption:
- The Ag Census reflects growers’ intentions and future plans in relation to regen ag programs (ex. Planning to plant cover crops), and may indicate willingness in adopting certain regen ag practices. A higher number in Ag Census indicates growers are more comfortable with implementing practices, and are proactively planning accordingly.
- Regrow leverages remote sensing to estimate the actual emergence of cover crops. In other words, it reveals whether the growers actually followed through on their intentions by planting seeds that successfully resulted in cover crops. The success of cover crop establishment often requires several growing seasons to refine and develop, reflecting a grower’s growing expertise in regenerative agriculture.
- Regrow’s outputs will generally be lower than Ag Census, given not all intended practices will likely be successfully implemented, and may be impacted by more or less favorable weather conditions.
Comparison of Ag Census 2022 and Regrow SI cover crop data
|
State |
Regrow Cover Crop (%) |
USDA Cover Crop (%) |
Difference (USDA % - Regrow %) |
|
Illinois |
2.8 |
3.8 |
1.0 |
|
Indiana |
4.2 |
7.9 |
3.7 |
|
Iowa |
2.2 |
5.0 |
2.8 |
|
Michigan |
9.7 |
9.1 |
-0.6 |
|
Minnesota |
4.3 |
3.5 |
-0.8 |
|
Missouri |
9.1 |
6.2 |
-2.9 |
|
Ohio |
3.1 |
7.1 |
4.0 |
|
Wisconsin |
9.6 |
7.9 |
-1.7 |
Reporting on tillage practices in the Ag Census differs from Regrow’s methodology as well, with the largest differences coming from how Reduced and No Till are defined. Regrow reports on three tillage classes: conventional till, reduced till, and no till. For each of these classes, we use a definition based on the amount of residue coverage, requiring a minimum percentage of the soil surface to be covered by crop residue. Regrow definitions are different from the Ag Census (which has additional tillage classes), leading to variations in how the data are reported in the two datasets.
- Regrow’s reduced tillage class is most similar to the Ag Census’ conservation tillage class, with both categories requiring a 30% residue coverage threshold.
- For No-till, Regrow requires a 60% residue coverage threshold, while the Ag Census does not have an explicit no till definition.
- In both of the above classes, Regrow does not use any tillage implement information (as the data is solely derived from remote sensing). This may lead to cases where a grower uses practices traditionally associated with reduced or no till, but due to low level of residue on the field following tillage, Regrow assigns the field a more intense tillage practice.
- In general, Regrow’s reduced and no tillage numbers will be lower than the Ag Census due to the different class definitions and residue only based classification methods.
Comparison of Ag Census 2022 and Regrow SI Conservation Tillage data
|
State |
Regrow Reduced Till (%) |
USDA Conservation Till (%) |
Difference (USDA % - Regrow %) |
|
Illinois |
23.8 |
38.8 |
15.0 |
|
Indiana |
33.1 |
34.2 |
1.1 |
|
Iowa |
16.2 |
35.9 |
19.7 |
|
Michigan |
25.8 |
30.8 |
4.9 |
|
Minnesota |
10.7 |
36.6 |
25.9 |
|
Missouri |
31.0 |
21.2 |
-9.8 |
|
Ohio |
29.7 |
29.9 |
0.2 |
|
Wisconsin |
27.0 |
28.1 |
1.1 |
Field Delineation
Delineated field boundaries that represent crop land are essential parameters for remote sensing, and serve as the basis for DNDC to quantify emissions of an area.
Field Delineation: USDA CLUs vs. Regrow Generated Field Boundaries
Regrow has developed an in-house field delineation model called Parcel ID, which generates precise field boundaries, crucial for evaluating key performance indicators, baseline emissions, and abatement potential in SI.
By using Regrow’s generated field boundaries, customers benefit from a higher level of precision compared to the less accurate legacy solutions provided by Common Land Units (CLUs), which lack comprehensive coverage, as detailed below.
Benefits of using Regrow Generated Field Boundaries:
Regrow's Parcel ID uses satellite imagery to accurately delineate agricultural field boundaries within a region. This is crucial for generating SI data, as it provides a detailed list of fields, which are processed through Regrow’s Monitor and Measure APIs. Parcel ID simplifies MRV farmer enrollment by pre-filling field boundaries, making it easier for farmers to select their fields. It also improves the accuracy of covered area identification and the exclusion of non-agricultural land. Unlike outdated CLUs, which can complicate farmer pre-fill and credit auditing in MRV programs, Regrow's Parcel ID offers precise and current field delineations, enabling more efficient credit auditing.
In many regions outside the corn belt, we found CLUs often missed many agricultural fields, providing an incomplete picture of cropping practices.
Comparison of CLUs (left) and Regrow parcels in Missouri (right)
Comparison of CLUs (left) and Regrow parcels in Wisconsin (right)
Comparison of CLUs (left) and Regrow parcels in Massachusetts (right)
Field Delineation Impacts to SI data?
With this update, Regrow has doubled the number of fields that are monitored and is able to more accurately classify crops versus grassland systems. Sustainability Insights also excludes fields with unsupported crops from emissions quantification leading to a more accurate emissions estimate. In some cases this changes the percentage of observed acres that have adopted a given practice, as we are evaluating a larger number of acres. For example, the percentage of fields in a supply shed that have adopted a ‘no-till’ practice may look different if the total number of fields/acres monitored in that supply shed increases. In other cases, with better crop classification and more selective emissions quantification of fields and crops, the acreage count for a given row crop might be reduced.
Cover Crop Reclassification
The previous methodology for assessing cover crops used a static time window (October - April), and could only evaluate cover crops following a summer commodity. This method worked well in the US Corn Belt, but was less generalizable to cover crop identification in other regions.
Regrow’s new methodology uses a dynamic time window and geographically specific thresholds that reflect climate variability, allowing for the assessment of cover crop presence at any time when no commodity crop is present on the field. This dynamic approach enables the detection of cover crops, including winter commodity systems.
The thresholds for cover crop presence are set dynamically by assessing the average greenness of nearby perennial grass areas during the cover crop season. The updated methodology for dynamic thresholds increases the overall accuracy of cover crop detection, especially in areas with temperate climates such as the US South including:
- Minimum of 8 weeks required between main commodity crop harvest and the following crop planting to make a green cover determination
- This results in green cover determinations that exclude volunteers, weeds, and short-lived commodity re-growth as possible cover crops
- Regional greenness thresholds are used to inform the cover crop determination. In practice this means in order for a determination of cover crop to be made, the field must have enough living vegetation on the field to cover the soil for the majority of the cover cropping period.
- This means that colder regions where there is less green cover may have lower greenness thresholds for classifying a cover crop.
How does this affect SI data?
Cover crop rates may increase in regions with colder winters as we now set a lower threshold for cover crop classification to reflect the harsher growing conditions while they may decrease in areas with milder winter conditions. This is because weed fields and volunteer crops are more accurately classified as non-cover crops.
Cover crop trends vary by year and region, with notable updates in our latest analysis. Generally, Regrow has identified an increase in cover crops from 2015 to 2019 compared to previous data. However, cover crop occurrences from 2020 to 2023 remain consistent with earlier findings. We're also gaining a deeper understanding of how weather patterns influence cover crop adoption.
Tillage Reclassification
Regrow’s new tillage classification approach makes a significant change in the definition of conservation tillage to align with USDA and NRCS references. Specifically, conservation tillage is now defined as having a minimum of 30% remaining crop residue on a field after a tillage event (as opposed to Regrow’s previous threshold of 15% minimum remaining crop residue on a field after a tillage event). This definition aligns with accepted global definitions of conservation tillage, defined by leaving at least 30% crop residue or more on the field.
Our previous approach defined conservation tillage as having a minimum of 15% remaining crop residue on a field after a tillage event. This methodology operated relatively well in the US corn belt, but had known gaps in the South, Northeast, and Western parts of the US. In all regions, it tended to overstate the amount of reduced tillage. Our updated thresholds take a more conservative stance, requiring a higher amount of residue to be present on the field post harvest to qualify for reduced and no till.
The lower residue threshold was initially used in Regrow’s tillage classification process to reflect norms in certain regions that focused primarily on corn and soybean production. By aligning our classifications to observed residue cover, the result is a more conservative account of tillage practices that removes the ambiguity in tillage intensity definition and better aligns with USDA and NRCS references. As a result, there is a decrease in the reported prevalence of conservation tillage, aligning our data more closely with regional survey statistics.
Historically, tillage practices in the Corn Belt (focused on corn and soybeans) had more variable tillage definitions, which influenced Regrow's approach. Meanwhile, in other regions like the Wheat Belt, where minimizing wind erosion is critical, farmers have historically been practicing reduced or no till in a way that targets specific residue cover outcomes. Initially Regrow employed a compromise approach between Corn Belt and Wheat Belt regions, which was similar to the Ag Census methodology from the USDA. However it has become clear that a more scalable, and consistent definition is needed.
Regrow is adopting a unified approach aligned with USDA’s NRCS Program and evolving best practices, ensuring national applicability. This strengthens the reliability and scalability of regen ag programs, crucial for CPGs aiming to meet stringent standards and gain consumer trust. The unified approach simplifies regen ag program implementation across different regions of the country, which is especially important for companies with widespread supply chains.
Comparison of Ag Census 2022 and Regrow SI Conservation Tillage data
|
USDA Ag Census |
Regrow’s Previous Classification |
Regrow’s Updated Classification (aligned to NRCS) |
|
|
Conventional till |
<15% residue cover |
<15% residue cover |
<30% residue cover |
|
Reduced Till |
15% - 30% residue cover |
15% residue cover (Reduced Till - Low Residue) |
30% residue cover |
|
Conservation Tillage |
>30% residue cover |
30% residue cover |
30% residue cover (equivalent to reduced tillage) |
|
Conservation Till / No Till |
30% residue cover |
50% residue cover |
60% residue cover |
Emissions and DNDC Model Updates
Regrow's Denitrification-Decomposition (DNDC) model is a biogeochemical model that estimates nutrient cycling in soil and GHG emissions, considering changes from new farming practices. It supports over 100 crop types and has been validated by more than 500 scientific peer reviews. DNDC is calibrated for each crop, field, and region by sourcing the best available emissions data for specific crop and geographic combinations.
DNDC Updates
The previous data for CONUS in SI was generated using DNDC version 11.0. For the July 2024 update, the modeling was done using version 12.2. The major changes between these versions include the following:
- Enhancing soil nitrogen cycling by improving the urea hydrolysis routines in the model
- Enhancing soil organic carbon cycling by introducing a biological mixing routine to represent influence of macrofauna on the soil system
- Amending problematic soil temperature methods which in turn improved emissions modeling
Based on the above changes, we increased precision and accuracy, and have overall seen an average increase in emissions. Additionally, we have observed a decrease in soil organic carbon sequestration values. These changes have improved emission predictions and our models ability to better represent the natural soils processes.
CO2 Conversion Factors
Regrow has changed CO2 equivalent conversion factors for N2O and CH4 to reflect values published in the 2021 IPCC Report.
The table below summarizes the changes between each version
|
Version |
N2O Multiplier |
CH4 Multiplier |
|
AR5 |
44 / 28 * 265 = 416.429 |
16 / 12 * 28 = 37.33 |
|
AR6 |
44 / 28 * 273 = 429 |
16 / 12 * 29.8 = 39.733 |
With the adoption of the new AR6 CO2e conversion factors, we anticipate a 3% increase in N2O values and a 6% increase in CH4 values, both expressed in CO2e. There are no changes to the multipliers for dSOC, so no adjustments are expected in this area. Detailed values for AR6 can be found in Table 7.15 of the IPCC Technical Report, found here.
Quality Control and Assurance
Monitor Sustainability Insights QA
The primary aim of Monitor Sustainability Insights (SI) QA is to discover errors or problems in the modeled agronomic logic, so that solutions can be implemented before finalized data production. Additionally, it sheds light on assessment of the overall quality of the data in terms of alignment with reasonable cropping practices, temporal expectations of those practices, and regional-scale reports of practice adoption. Monitor SI QA is conducted on 1) Monitor results, aggregated with SI logic, to check for data validity and 2) Monitor results, aggregated with SI logic, as compared to USDA Agricultural Census data. The process consists of metrics, reporting, and analysis aggregated over time and geographic extent to check for valid data generation results. The QA process ensures data completeness by analyzing error rates, cropping period consistencies, and practice distributions. The concepts checked for data validity include:
- Measurement of return rate of associated field data over the supported time period
- Alignment of agronomic expectations for cropping period time windows
- Assessment of spatiotemporal distribution of management practices and crop types
The concepts checked for data comparison to USDA Agricultural Census include:
- Assessment of methodological differences, in relation to how and in what manner the data is collected and aggregated
- Measurement of data reporting completeness, i.e., how much field data is retained for reporting for each approach based on inclusion criteria
- Comparison of management practice rates and crop type distributions, to find common trends in resilient agricultural practices temporally
DNDC QA
The primary goal of DNDC SI QA is to identify abnormal values or outliers in our modeled GHG results that may indicate errors or problems within the model or input generation pipeline itself. DNDC SI QA aims to validate the GHG and EF numbers that are produced using a large suite of peer reviewed studies and LCA/EF databases. When conducting QA on the DNDC produced GHG values, data is aggregated such that results can be assessed on a country, regional, and crop level. EFs will also be aggregated in a similar fashion such that we can validate them against our known data sources. Validating data includes ensuring that GHG values are within reasonable bounds for individual emissions (SOC, N2O, and CH4) defined by a collection of peer reviewed studies and their corresponding in situ measurements. Similarly, crop EFs are validated by comparing values with Quantis data. If outliers are identified at either phase of QA, DNDC inputs such as weather, soil texture, fertilizer amounts, and crop rotations will be investigated to identify a reasonable explanation for the outlier or identify a problem within the model itself.
Sustainability Insights Bug Fixes
Bug fix: in certain instances, Sustainability Insights was using a mean average to aggregate KPIs, such as emissions factors. With the recent refresh in data, that was fixed to always use an area-weighted average for relevant metrics.