Which Method is Most Accurate to Measure Soybean Maturity?
Measuring soybean maturity dynamics is crucial information for researchers as they analyze new plant hybrids. Accurately classifying the soybean maturity group determines the environment in which the plant will best perform, making it one of the most important factors in the product’s ultimate success.
The challenge lies in gathering that information efficiently and with statistical certainty in the research phase.
Currently, measuring soybean maturity requires manual evaluation of all plots in a field by an entire team of agronomists. Several times a week, for multiple weeks, the team walks the field to record data on every single plot.
Carrying out this process is a costly and labor-intensive exercise, regarded as a necessary part of the trial. The high degree of manual data collection means mistakes and subjective measurements are likely.
UAVs equipped with remote sensors can now collect these measurements significantly faster and more accurately than with manual methods.
Faster collection times and fewer people required to gather data also enable more time to analyze information — leading to fewer trials and lower research and development costs across the board.
Of course, faster measurements and lower costs mean nothing if the measurements aren’t accurate. In this piece, we compare the results of SlantRange measurements against manual methods and third-party analytics in measuring soybean maturity.
You’ll see from the following examples that SlantRange measurements are the most statistically accurate method of measuring soybean maturity.
Learn More: SlantRange Solutions for Research and Breeding
Measuring Soybean Maturity in 3 Ways
In this comparative soybean maturity trial, a seed producer collected and analyzed data using three distinct methods:
- RGB (color) imagery collected with a UAV and processed in the SlantRange aerial phenotyping platform
- RGB imagery also collected with a UAV and processed in a third-party analytics platform
- Ground truth data, manually gathered by agronomists
The seed producer surveyed the same trials in the same North American location to compare the differences in data collected with each approach.
Ground truth data served as the control in this experiment, with SlantRange and third-party data tested against the manual measurements. Both SlantRange and the third-party analytics platform analyzed the same raw images taken by the same RGB sensor.
At the end of the trial, results showed SlantRange systems were most effective in accurately predicting maturity.
Our data was closely correlated with ground truth data. In areas where there were discrepancies between the aerial data and manual ground measurements, the aerial data provided a record that uncovered critical errors in manual measurements. The following sections detail the findings.
SlantRange Data Shows Highest Correlation With Manual Measurements
For most agronomists and researchers, ground-truth measurements are the standard of accurate measurements.
The downside of manual measurements, of course, is the time and labor costs of collecting the data and subjective results. Measurements need to be taken several times and trials need to be replicated under a broad range of conditions to get statistically accurate results accounting for subjectivity and error.
We looked at the correlation between each type of measurement to compare the accuracy of each method.
When comparing the data across the entire trial, SlantRange data has the highest correlation with data collected by the team of agronomists.
The series of three graphs show the correlation between each of the measurement methods, with the corresponding coefficient showing the degree to which the two methods are similar.
Third-party measurements had the lowest correlation to manual methods, with a coefficient of 0.75. SlantRange and third-party measurements had a correlation of 0.80. SlantRange and manual measurements had the highest correlation of 0.90 — which indicates little discrepancy.
We then broke the data apart even further, to understand the degree of correlation between measurements taken across the entire trial. Each axis represents days to maturity for each method compared below.
In each graph, days to maturity measured by ground-truth data are on the X-axis, and third-party figures are on the Y-axis.
The long lines on the left graph indicate a larger discrepancy between data collected by the system and ground-truth data. A strong correlation would show the dots in a straight line, with less discrepancy between the systems.
Third-party data and manual measurements have a correlation of 0.75, which is the weakest of the three methods, and less than this platform’s correlation with SlantRange data. But before we could declare SlantRange the most accurate, we compared it to the manual measurements.
When we compared SlantRange and manual measurements, you can see from the graphs above that there’s a high degree of correlation — 0.90. The high correlation between UAV data analyzed with SlantRange and manual measurements indicates that SlantRange’s automated approach to data collection and analysis is as statistically accurate as manual methods.
The advantage for researchers is clear. Flying a drone over a field takes a lot less time to collect information than a team of agronomists walking each row.
SlantRange measurements are highly repeatable, meaning you can fly the fields to collect new data as often as you want. They’re also highly accurate, meaning you can feel confident that your measurements are statistically accurate and objective.
The SlantRange platform also provides a visual record that can be easily reviewed to understand anomalies in the data and what may have caused those errors. Manual observations do not provide this type of record.
As applied to soybean maturity, this means collecting data can be something one person does at regular intervals, instead of an all-hands-on-deck effort for up to six weeks during the peak growing season.
While discrepancies do exist, the following sections highlight areas SlantRange analysis can find mistakes in the manual measurements.
Learn More: New Technologies for Agricultural Intelligence
SlantRange Finds Data Errors Where Manual Methods Can’t
While SlantRange was most accurate, two particular plots stood out as clear examples of areas where our measurements uncovered data errors in manual measurements.
In both plots, measurements taken from manual sources did not match the visual record created from the aerial imagery. This historic record matters, because it gives the agronomist a chance to compare data against the imagery to either validate or confirm a mistake.
In the following examples, there’s a clear disconnect between the recorded maturity date and the plot-level imagery, which provides a clear demonstration of a false prediction.
Look at these two images of the same plot that were taken one week apart.
As you can clearly see from the visual record, the plot is close to maturity in the first image and brown in the second.
Compare this to the recorded maturity date of each method:
- SlantRange: Four days after the first image
- Manual: 10 days after the first image, and four days after images show the plot is past maturity
The second example shows a similar error from a different plot in the same trial. Again, the images were taken one week apart. In the first image, the plot is mostly mature and completely brown in the second.
Compare this to the recorded maturity date of each method:
- SlantRange: Two days after the first image
- Manual: Four days after the second image
In both examples, SlantRange and the third-party system give results within a few days of each other. The two analytics engines processed the same imagery to get slightly different results. Manual methods show a result that’s a far outlier.
When comparing the data to the visual record, the difference becomes even more apparent. You can also see the manual measurements are incorrect in both examples, likely the result of an incorrect date marked in the record.
Visual records show researchers precisely what was measured and what the field looked like on a specific day.
In each case, researchers get an opportunity to go back and re-evaluate any unexpected results.
Viewing images not only showed which data was accurate but also highlighted errors. When there’s a visual image to corroborate measurements, you can be sure that your measurements are accurate, and make corrections when they’re not.
Without that imagery, there is no record of the plot for you to review any anomalies after the fact.
More Efficient and Accurate Trial Results
Ground truth measurements have long been the standard for accurate data because agronomists have been using manual methods for centuries. But, of course, gathering data using these methods is tedious, costly, and often subjective.
As you can see when comparing manual measurements to aerial phenotyping data, not only is the automated method more efficient and accurate, but the visual record also creates a crucial history to double-check any discrepancies — creating greater confidence in trial results.
When you can feel confident that your data is accurate the first time, you can run fewer trials. In the long-term, fewer trials mean you can bring stronger products to market sooner — reducing research and development costs, and decreasing the time to profit.
For more information on how we measure soybean maturity, with additional examples, read this recent blog.
To learn more about SlantRange aerial phenotyping solutions and available data products, fill out the form at the bottom of this page for our product guide. To learn more about how we may be able to help your organization collect better data and improve your research trials, click here to get in touch with our team.