Overview of Regrow's platform data enhancements for 2024 crop-year data
In April 2025, Regrow will release updated 2024 crop-year data alongside a refresh of prior year data, using improvements to the underlying models that we use for remote sensing. These model changes deliver greater accuracy and precision in analyzing satellite imagery and machine learning, including planting and harvest date estimates, crop type, presence of cover crops, and extent of tillage.
This year, we are pleased to report that in each of these categories, we have made significant accuracy improvements relative to our database of ground observations. This includes:
- 7% YoY improvement in correct identification of cover crops
- 8% YoY improvement in determining tillage extent (conservation vs. conventional)
- 25% improvement in estimation accuracy for planting and harvest dates for Summer crops, and 19% improvement in this metric for Winter crops
Below, we review each of the underlying model changes and updates/bug fixes, and provide context on our previous approach and the impacts of each of the updates. If you have any questions, please reach out to a member of your Regrow account team.
Cultivation Cycles
Previously, Regrow quantified emissions based on calendar year, ascribing all activities on a field in that year to a crop. Emissions quantification, including emissions factors, are now associated with a specific ‘cultivation cycle,’ as opposed to a full ‘crop year.’ This means that emissions for a given crop are ascribed to the period between the previous crop’s harvest and the harvest of the crop that is being measured (see visual diagram of change).
The ‘cultivation cycle’ approach is a more accurate agronomic representation of the emissions associated with rearing a given cash crop. This update aligns Regrow’s emissions factors more closely to LCA methodology specifications.
Specifically, we are now:
- More accurately identifying winter crops as cash crops as opposed to cover crops, which substantially reduces incidental over counting of cover crops
- More accurately calculating emissions on double cropping systems.
Cover Crop
The following updates were introduced to our cover crop methodology:
Previously, the observation window for identifying a cover crop had been 4 weeks. That is now being extended to a minimum of 8 weeks. This change improves the accuracy for Regrow remote sensing to distinguish between cover crops and extraneous green cover, such as weeds, volunteers, or short-lived commodity re-growth commodity re-growth.
In addition, we had previously provided limited reporting of cover crops on perennial fields and we no longer we report cover crops on perennial fields. We deprecated this feature to clearly differentiate between cover crops planted between commodities and similar species that are grown on a perennial basis for forage.
Lastly, we used Living Root Quality Score (LRQS), an internal scoring system developed by Regrow data science team, whereas the algorithm used to evaluate cover crops is now based on the Normalized Difference Vegetation Index (NDVI). NDVI is a standard industry measure of field greenness. This will create metric parity with other external data sources that use NDVI, making it simpler for Sustainability Insights users to compare “apples to apples.”
Tillage
The tillage observation window had previously been 12 weeks. We had previously been reporting the lowest weekly value of tillage, as opposed to median weekly value. We are shortening the observation window for which we use remote sensing to detect tillage events to 8 weeks. Also, we are reporting the median residue value during the observation period. The longer you look at a field with remote sensing, the more likely you are to identify conventional tillage, since residue is degrading the field is covered by a cover crop, or the residue has been removed. Therefore, moving to a shorter window is a more accurate way of ascribing tillage events, and is less conservative than our previous approach.
When it comes to distinguishing tillage practices across different commodities, previously, we used a single threshold across all crop types (fragile and non-fragile) to determine if there was a tillage event. We are implementing crop type-specific residue thresholds. This means that the threshold of residue needed to classify tillage practices as conventional, reduced, or no till will reflect the quantity of biomass produced by different crop types. The thresholds for fragile crops (soy, winter wheat, cotton, cereal grains, potatoes, canola) are:
- Conventional Till = 0-15% residue
- Reduced Till = 15-30% residue
- No Till = 30+% residue
- Conventional Till = 0-30% residue
- Reduced Till = 30-60% residue
- No Till = 60+% residue
Different crops produce different amounts of biomass that break down at different rates. By differentiating thresholds for each crop type, and detecting for different levels of tillage accordingly, we are able to more accurately determine when tillage events are occurring on a given field.
Crop Detection
We made several updates to our crop detection logic.
Previously, we used NDVI trendlines as a predictor of plant and harvest dates. We updated our methodology to estimate planting and harvest dates, which is an important determinant for running the DNDC model for emissions quantification. The new methodology uses advanced machine learning algorithms on top of our remote sensing data to estimate planting and harvest dates more precisely. The ML based approach increases the accuracy of our planting and harvest date determinations, which, in turn, increases the accuracy of our GHG estimates.
We also now able to detect more crops:
- Flax
- Peanut
- Dry beans
- Chickpeas
- Durum wheat
- Lentils
Green Cover
With this data update, we introduced the Green Cover metrics which will replace the Living Root metric and will provide a better understanding of cover crop potential in the supply shed. The goal of this KPI is to enable a view of greenness intensity during non-commodity periods within a supply shed over time.
Regrow uses an established remote sensing index, the Normalized Difference Vegetation Index, to estimate the greenness of a field during the non-commodity period. The NDVI increases from 0 to 1 as the chlorophyll or photosynthetic potential (aka greenness) increases in the vegetation being observed. For every field we calculate the monthly maximum NDVI during the non-commodity period (from the harvest of the previous crop to the planting of the next crop). The average of those monthly maximum values across the non-commodity period is then used to place that field into one of the four categories.
The NDVI bins applied here are:
|
NDVI |
Category |
|
0.0 - 0.3 |
No green cover |
|
0.3 - 0.4 |
Low green cover |
|
0.4 - 0.5 |
Moderate green cover |
|
0.5 - 1.0 |
High green cover |
These categories are calculated using the same NDVI values that inform our cover crop classification process. The categories also share similar NDVI thresholds so the acres of cover crop and no cover crop will correlate with acres in the high green cover and no green cover categories. Whereas all acres in the high green cover category will count as acres with a cover crop and all acres in the no green cover category will count as no cover crop acres, acres in the low or medium cover categories may be counted as cover crop, but only if they meet regional NDVI thresholds that vary from year to year. For more detail, take a look at the details related to our Cover Crop metric.
Yield
While NASS no longer provides county-level yield data for each year as of 2024, Regrow 2024 yield data is modeled from NASS state-level yield data in counties where NASS previously reported county-level yield data. By modeling 2024 yield data at a county level we are able to continue reporting county level yield where data was historically available. Predicted county-level yield is more granular and reliable than using state-level yields for all counties.
Bug fixes & infrastructure updates
Two bug fixes and updates were introduced to the Sustainability Insights application, on top of the MonitorML updates.
1. When aggregating certain KPIs across subregions up to the supply-shed level using a weighted average approach, Sustainability Insights was using an area-based weight, which is now updated to a yield-based weight, where appropriate.
2. For programs with overlapping subregions, which was limited in scope, some double-counting of acres (crop acres, cover crop acres, tillage acres) was occurring, which is now fixed.