Yield Monitor Data for Management Decisions
Crop Insights written by Mark Jeschke, Ph.D., DuPont Pioneer Agronomy Information Manager
- An increasing reliance upon yield monitor data to evaluate crop performance and inform management decisions has placed greater importance on ensuring yield data quality.
- Yield monitors are capable of providing very accurate estimates of corn yield; however, real-world performance can fall well short of this potential due to lack of proper calibration and other sources of error.
- Yield monitor accuracy for estimating yields and comparing products in on-farm trials was evaluated using yield data from 286 DuPont Pioneer on-farm strip trials conducted from 2013 to 2016.
- Among the 286 trial locations, the average yield monitor error rate compared to calibrated weigh wagons was within +/-3% in 59% of locations, with the yield monitor overestimating yield in 12% of locations and underestimating yield in 27% of locations.
- Yield monitors accurately ranked the performance of trial entries in 41% of locations and correctly selected the top yielding entry in 50% of locations.
- Yield monitor estimates at 28% of locations provided both an accurate location-level yield estimate and an accurate ranking of trial entries.
- Results from the largest trials (>10 acres) suggest that over 1/3 of field-scale yield monitor data is likely inaccurate, which has important implications for management decisions based on yield monitor data.
The widespread adoption of yield monitor systems over the past 20 years has facilitated data-driven decision making in corn production in a way that was not possible before. Yield monitors do not directly measure yield, rather they estimate relative yield based on mass flow rate. Coupled with a GPS receiver, yield monitors provide an assessment of spatial variability in relative yield across a landscape.
Historically, calibrated weigh wagons and portable moisture meters have been used in the seed industry to measure grain weight and estimate grain moisture of strip plot entries. Performance data for Pioneer® brand corn products in on-farm strip trials is still almost entirely based on weigh wagon measurements. However, yield monitors are commonly used to measure performance in agronomic trials, which often include both genetic and management components and are often larger in size. Out of over 6,000 Pioneer on-farm agronomic trials conducted from 2013 to 2016, 57% recorded weigh wagon data, 39% yield monitor data, and 4% both.
Recent advances in transfer and aggregation of spatial farm data have allowed yield monitor data to be increasingly leveraged to assess performance of genetics and management practices over large scales. Given the increasing reliance upon yield monitor data to evaluate crop performance and provide a basis for management decisions, it is important to determine the accuracy of the yield monitor data being collected. In the mid 1990’s, DuPont Pioneer researchers conducted an evaluation of yield monitor accuracy compared to weigh wagons in on-farm strip trials (Doerge, 1997), the scope of which was limited by the relative scarcity of yield monitors during the first few seasons following their commercial introduction. The purpose of this Crop Insights is to revisit this topic and assess the current state of yield monitor accuracy based on a more recent and much larger dataset of on-farm strip trials.
Yield Monitor Accuracy – Potential vs. Reality
Research has shown that yield monitors are capable of providing very accurate estimates of corn yield. A 3-yr study conducted across six locations in South Dakota found very close agreement between yields measured using a weigh wagon and a well-calibrated yield monitor (r2 = 0.967) (Nelson et al., 2015). Likewise, Professor Robert Nielsen at Purdue University has reported that yield estimates in field-scale research trials from yield monitors are typically within 1% of corn yield as measured by a weigh wagon or farm scale in his research (Nielsen, 2017).
However, despite this high level of achievable accuracy, there is reason to suspect that much of the yield data being collected by growers using yield monitors falls well short of this potential. Proper calibration is critical for producing accurate yield estimates with yield monitors. If a yield monitor is not calibrated to the characteristics of the grain being harvested, or not calibrated at all, yield estimates can be skewed. Error rates of 7-10% are not uncommon for corn harvested late in the season if the yield monitor was calibrated only at the beginning of the harvest season due to changes in grain moisture content (Nielsen, 2017). A 2015 University of Wisconsin study sought to determine real-world yield monitor accuracy by conducting random spot-checks of combines during harvest (Luck, 2017). Of the four combines tested, two had error rates of 1-3% and two had errors rates in the 6-9% range.
Yield monitor accuracy for estimating yields and comparing products in on-farm trials was evaluated using yield data from DuPont Pioneer on-farm strip trials conducted from 2013 to 2016 in which yield data were collected using both a weigh wagon and a yield monitor. These trials included a total of 3,923 entries across 286 locations in 15 states and one Canadian province. The brand of yield monitors used in the trials was not recorded. Likewise, the calibration status of the yield monitors at the time the trials were harvested is not known, but can be inferred to some extent based on the accuracy of yield estimates compared to weigh wagon measurements. Given the large number of trial locations, the dataset is likely a reasonably representative sample of yield data being collected by growers. If anything, the accuracy of the yield monitor data might be better than average given that it comes from on-farm trials, where yield data accuracy is presumably prioritized.
The size of the on-farm trials included in the analysis varied widely. The number of entries per trial ranged from 1 to 49, with the majority of trials including between 8 and 16 entries. Strip length ranged from 200 to 4700 ft and strip width ranged from 10 to 50 ft. With proper calibration, a yield estimate within 1-3% of the total grain harvested in a field is generally considered achievable (Darr, 2016). For the purposes of this analysis, a yield monitor estimate within 3% of the weigh wagon yield was considered “accurate.”
A linear regression of yield monitor estimates vs. their corresponding weigh wagon measurements for all 3,923 individual comparisons produced an r2 of 0.8453 (Figure 1), indicating that overall yield monitor accuracy fell well short of potential accuracy as shown by Nelson, et al. (r2 = 0.967). The slope of the regression was 0.94, indicating a slight tendency to overestimate yield at low yield levels and underestimate yield at high yield levels. Yield monitor estimates for 55% of individual entries were within 3% of weigh wagon measurements.
Figure 1. Relationship of yield monitor and weigh wagon corn yields from 3,923 comparisons at 286 locations, 2013-2016.
Among the 286 trial locations, the average yield monitor error rate was within +/-3% in 59% of locations, with the yield monitor overestimating yield in 12% of locations and underestimating yield in 27% of locations (Figure 2). Average yield estimates were more than 10% off in 7% of locations.
Figure 2. Average yield monitor error rate based on comparison to weigh wagon measurements at 286 locations, 2013-2016.
Accuracy of Entry Ranking
With the increasing reliance on yield monitor data as a basis for evaluating hybrids and agronomic practices in on-farm trials, it is important to understand the effectiveness of this technology for making accurate comparisons. In a 1996 study comparing yield monitor estimates to weigh wagon measurements in DuPont Pioneer on-farm strip trials, Spearman Rank Correlation was used to evaluate the accuracy of yield monitors in ranking the performance of entries in the trials (Doerge, 1997). The correlation coefficient (r) in this test can range from 1, indicating perfect correlation between the yield monitor and weigh wagon rankings, to -1, indicating inverse correlation. A correlation coefficient of zero indicates no correlation. A minimum threshold of 0.93 was used for designating the ranking of entries in a trial as “accurate.”
In the 1996 study, across 19 study locations with an average of around 16 entries per location, the average correlation between yield monitor estimates and weigh wagon measurements for ranking entries was 0.78. Yield monitor rankings at 6 out of 19 locations (32%) qualified as accurate. The yield monitor correctly selected the top yielding entry in 8 of 19 locations (42%).
This same methodology was applied to a subset of 150 locations from the current study with a similar number of entries per location as those in the 1996 study (12 to 20 entries). Yield monitor accuracy at ranking entries was slightly better in the current study than in the 1996 study. The average correlation coefficient across locations was 0.80, with accurate yield monitor ranking of entries at 41% of locations. The yield monitor correctly selected the top yielding entry in 75 of 150 locations (50%). Examples of individual trials from the current study with differing levels of yield monitor accuracy in estimating overall yields and ranking entries are shown in Figure 3, Figure 4 and Figure 5.
Among these 150 locations, the average location-level yield monitor error rate was within +/-3% at 92 locations (61%) Yield monitor estimates at 28% of locations provided both an accurate location-level yield estimate and an accurate ranking of trial entries (Figure 6).
Figure 3. Example of an on-farm trial location in which both the yield estimates and ranking of entries by the yield monitor were highly accurate. (Average error = 1.5%, r = 0.93; trial located in eastern Iowa, 2015)
Figure 4. Example of an on-farm trial location in which the location average yield estimate was accurate, but the ranking of entries was only moderately accurate. (Average error = 2.0%, r = 0.70; trial located in eastern Iowa, 2014).
Figure 5. Example of an on-farm trial location in which the ranking of entries by the yield monitor was highly accurate but the yield estimates were inaccurate. (Average error = 8.7%, r = 0.96; trial located in northeast Nebraska, 2013).
Figure 6. Overview of yield monitor data accuracy in DuPont Pioneer on-farm trials, 2013-2016.
Factors Influencing Yield Monitor Accuracy
In order to further evaluate factors influencing yield monitor accuracy, additional analysis was conducted on a subset of the data that excluded locations in which the average yield monitor error rate was greater than 3%. The rationale for this approach was that locations in which the yield monitor estimates consistently trended more than 3% above or below the weigh wagon measurements likely reflected a lack of proper yield monitor calibration. This approach does not necessarily eliminate poor calibration as a source of error, but likely substantially reduces it. This subset of locations (hereafter referred to as the “calibrated subset”) included 170 of the 286 total locations (59%).
Figure 7. Relationship of yield monitor and weigh wagon corn yields from 2,346 comparisons at 170 locations where the location average error rate was less than 3%, 2013-2016.
A linear regression of yield monitor estimates vs. weigh wagon measurements for the calibrated subset produced an r2 of 0.9511 (Figure 7), a substantial improvement from the r2 of 0.8453 for the full dataset. However, there were still 21% of individual entries with yield monitor error rates greater than 3%. The accuracy of entry ranking was only slightly improved with the calibrated subset of locations. Among 92 locations with between 12 and 20 entries, the average correlation coefficient was 0.83, with 46% of locations meeting the threshold for “accurate” ranking of entries (r > 0.93). The yield monitor accurately picked the top entry in 57% of locations. The fact that there was not more of an improvement in ranking accuracy suggests that, 1) there remains error in the dataset attributable to poor calibration, 2) there are other sources of error influencing yield estimates, or 3) some combination of the two.
Previous research, including the 1996 Pioneer study, has noted greater yield monitor error in on-farm strip trials in which the strip lengths and, consequently, the load sizes were relatively small. At the time of the 1996 study, the minimum recommended load size for Pioneer on-farm strip trials was 4,000 lbs (Peterson, 1996).
Figure 8. Yield monitor error (%) as influenced by load size.
For most of the entries in the calibrated subset of the current study, load size ranged from around 2,000 to 22,000 lbs. Results show some evidence of decreasing error rate with greater load size, although outliers were still present with load sizes greater than 10,000 lbs (Figure 8).
One of the most common sources of yield monitor error is grain characteristics (moisture, test weight) that differ substantially from the grain harvested for calibration. In practice, this most commonly occurs when the yield monitor is calibrated at the start of harvest when the grain is relatively wet and not recalibrated for drier grain later in the harvest season. Analysis of data from on-farm strip trials does not provide a great deal of insight on the amount of yield monitor error attributable to lack of recalibration during the harvest season because the trials are generally harvested in a single day and typically do not include hybrids with a wide range of grain moisture. A limited number of trials in the current study did include hybrids covering a wide span of comparative relative maturity (CRM) and harvest moisture; an example of which is shown in Figure 9.
In this trial, the yield monitor estimate was very accurate for the wettest hybrid, but the error rate increased as grain moisture decreased. The driest hybrid was 9 points drier than the wettest and yield monitor error for this hybrid exceeded 12%. Experts recommend recalibration when grain moisture changes by more than 4 points.
Figure 9. Yield monitor error as influenced by grain moisture in an on-farm trial conducted in Indiana in 2015. Data point labels indicate the comparative relative maturity of individual hybrid entries.
Results of the current study did not provide evidence of a higher rate of yield monitor error associated with any specific corn hybrid family. The calibrated subset of locations included 26 Pioneer® brand hybrid families harvested at 30 or more locations. There were no hybrid families that read consistently high or low on yield monitors. For all of these hybrid families, the average error rate across locations was with +/-1% (Figure 10).
Figure 10. Average yield monitor error rates associated with Pioneer® brand hybrid families included in 30 or more on-farm trials*.
(* Pioneer® brand products represented in Figure 10: P0157AMX (AMX,LL,RR2), P0193AMX (AMX,LL,RR2), P0297AMXT (AMXT,LL,RR2), P0339AMXT (AMXT,LL,RR2), P0407AMXT (AMXT,LL,RR2), P0533AM1 (AM1,LL,RR2), P0570AMXT (AMXT,LL,RR2), P0589AM (AM,LL,RR2), P0589AMXT (AMXT,LL,RR2), P0636AMX (AMX,LL,RR2), P0760AMXT (AMXT,LL,RR2), P0937AM (AM,LL,RR2), P0969AM (AM,LL,RR2), P0969AMXT (AMXT,LL,RR2), P1142AMX (AMX,LL,RR2), P1197AM (AM,LL,RR2), P1257AM (AM,LL,RR2), P1311AMXT (AMXT,LL,RR2), P1417AMX (AMX,LL,RR2), P1443AM (AM,LL,RR2), P1479AM (AM,LL,RR2), P9526AMX (AMX,LL,RR2), P9526AMXT (AMXT,LL,RR2), P9538AMXT (AMXT,LL,RR2), P9644AMX (AMX,LL,RR2), P9681AMX (AMX,LL,RR2), P9703AMX (AMX,LL,RR2), P9917AMX (AMX,LL,RR2), P9929AMXT (AMXT,LL,RR2))
Implications for Field Scale Accuracy
Results of this analysis have shown that yield monitors have the capability of providing accurate yield data in on-farm strip trials but that, in reality, yield estimates and product comparisons derived from yield monitors are often inaccurate at the location level due to error in the data. However, the primary utility of yield monitors is, and always has been, assessing spatial variability in relative yield at the field scale. What insights can the results of this study provide regarding yield monitor accuracy at the field scale?
Total harvested area of trial locations in this study ranged from less than 1 to greater than 25 acres. Locations with the greatest average yield monitor error (>10%) tended to be less than 10 acres (Figure 11). However, error rates greater than 3% were still common for trials larger than 10 acres; occurring at 38% of locations.
Figure 11. Average yield monitor error rate and total harvested area of on-farm trial locations.
These results suggest that over 1/3 of field-scale yield monitor data is likely inaccurate (error rate >3%). This has important implications for management decisions based on yield monitor data, both at the farm level and beyond. As improvements in data collection and transfer make the aggregation of yield data into larger area-level datasets more seamless, the fact that a substantial portion of the data feeding into these systems are inaccurate undermines the reliability of analyses and summaries based on these data. Post-calibration using scale tickets is a fix that is often applied to align the yield monitor estimate for total yield of a field in line with the actual weight of grain harvested; however, this method applies a uniform correction across the entire field, which may not reflect the actual spatial variation of yield in the field. The oft-repeated adage regarding computer systems of “garbage in equals garbage out” is frequently, and quite fittingly, applied to yield monitor data. As the industry becomes increasingly reliant on yield monitor data for performance insights and management decisions, the “garbage out” becomes more of a concern.
Practices to Improve Yield Monitor Accuracy
The following guidelines, adapted from Yield Monitor Systems (Darr, 2016), can help maximize yield monitor accuracy in on-farm trials.
Mass Flow Sensor
The mass flow sensor must be calibrated to ensure accurate yield data. In general, the mass flow sensor should be recalibrated anytime there is a significant change in crop conditions. These include the following conditions:
- After a long period of inactivity such as at the beginning of a new season.
- Switching between crop types.
- Significant changes in crop moisture of more than 4%.
- Significant test weight changes.
- Changes in crop conditions that cause a shift in normal operating speeds including lodged or downed crops, high moisture crops, or significant changes in ground conditions.
Calibration Procedure: Specific calibration procedures change based on the manufacturer, but several general recommendations fit all brands:
- Calibrate for at least the minimum number of loads recommended by the yield monitor manufacturer.
- Each calibration load should be at least 3,000 lbs; greater than 5,000 lbs is preferred.
- Calibration loads should be taken as single passes when possible to avoid errors associated with grain flow delay.
- Each calibration load should be conducted at a different mass flow rate. This can be controlled either by slowing down the maximum speed of the combine or by maintaining a set speed and reducing the active header width.
- The calibration flow rates should cover the entire range of flow rates that are expected in the target crop.
- After calibration, you can use “regions” or “loads” to monitor the accuracy of the calibration.
The moisture sensor should also be recalibrated periodically or when there is a significant change in crop conditions.
Calibration Procedure: Specific calibration procedures change based on the manufacturer, but several general recommendations fit all brands:
- Start a new combine “load”. This will create a new log that can be used to calibrate the grain moisture.
- Harvest an entire grain tank of grain.
- Stop the harvester and randomly sample the grain tank from several locations.
- Record the load moisture from the yield monitor.
- Calculate the actual moisture content of the grain tank sample using an accurate moisture tester. Handheld moisture meters are generally not accurate enough for this measurement unless it has been calibrated against a higher accuracy meter. To reduce errors, record three separate moisture readings from the single grain sample and use the average as the actual moisture.
- Enter the difference between the actual moisture and the yield monitor load moisture as a moisture offset.
Temperature calibration requires a similar offset adjustment. Make sure to calibrate temperature when the combine is not operating and has been in a constant shaded environment for a couple of hours.
Best Management Practices for Test Plots
While yield monitors can be excellent tools for field scale evaluation, care must be taken when using these same tools for small scale comparisons such as test plot strips. The following steps will help to improve yield monitor performance in short test strips, but well calibrated weigh wagons are still recommended for greater accuracy.
- Operate at normal combine speed. Test plots are often shorter rows which can lead to operators slowing down. The mass flow sensor is calibrated for normal crop flow, so to maintain accuracy the test plot should be conducted under the same conditions.
- Conduct rolling starts. To get the combine up to steady state grain flow as quickly as possible make sure the combine is moving at a normal speed when first engaging the crop. This is known as a rolling start.
- Be wary of significant moisture differences. If the test plot has significant grain moisture differences (more than 5% differences) then hand samples of the plots should be collected to verify the moisture content. For every 1% error in grain moisture the yield calculation will be off by 2.5 bu/acre.
- Avoid changing terrain. If the test plot field has rolling terrain you should harvest all plots in the same direction. This will reduce the impact of field slope on yield data errors.
- Maintain an accurate header width. When harvesting a test plot with a platform header be sure to maintain a consistent cutting width throughout the plot.
Darr, M. 2016. Yield Monitor Systems. Iowa State Univ.
Doerge, T. 1997. Weigh Wagon vs. Yield Monitor Comparison. DuPont Pioneer Crop Insights 7:17.
Luck, B.D. 2017. Calibrate your yield monitor for greater accuracy during harvest. Univ. of Wisconsin Extension A4146.
Nelson, B.P., R.W. Elmore, and A.W. Lenssen. 2015. Comparing yield monitors with weigh wagons for on-farm corn hybrid evaluation. Crop, Forage, and Turfgrass Management.
Nielsen, R.L. 2017.Yield Monitor Calibration: Garbage In, Garbage Out. Purdue Univ. Agronomy Extension.
Peterson, T.A. 1996. 1996 Guidelines for using yield monitors to collect Pioneer strip trial data. DuPont Pioneer Crop Insights 6:17.