September 25, 2013
Dear Adapt-N Users,
It has been an interesting season for nitrogen management with the unusual and variable weather conditions in most regions. We hope that our Adapt-N tool was useful in managing those challenges. We are now moving into the harvest season, and some of you may be comparing our tool’s recommendations with other N rate recommendation systems. We want to provide some guidance on this process.
When end-of-season comparisons are made between Adapt-N and other recommendation systems through retrospective runs, an appropriate expected yield should be used. Adapt-N is used as a predictive tool in the early growing season, and this means that the expected yield should be based on an achievable yield, as determined at sidedress time, based on real-world scenarios. Please enter this expected yield in the tool, rather than the actual yield achieved (if different) at the end of the season. We recommend that the achievable yield is targeted as approximately the second highest yield achieved out of the last five years, unless there are clear early indications of reduced or increased yield potential. (For example, if yields in 2008-2012 were 200, 215, 190, 185, 120, then an appropriate expected yield for 2013 would be 200 bu/ac).
The post-season evaluation should not be based on the eventual final yield of that growing season, as this may have been impacted by late-season processes like drought (such as 120 bu/ac in 2012 example above), weed competition, hail, or pest pressure, that cannot be predicted at sidedress time. This would not be representative of the use of Adapt-N or other recommendation systems in a real-world scenario, where an optimum N rate must be predicted in late spring or early summer.
Season End Date for retrospective runs
For a retrospective analysis of what rate Adapt-N would have recommended, you will need to use an appropriate “Season End Date” in the tool. This input field is found in the top left corner in interface. When you are ready to run the simulation for a location, enter the date on which sidedressing was done, or a reasonable date on which sidedressing would have been done (if no sidedressing was performed), then submit the simulation. The model will simulate through that date, allowing it to provide a representative recommendation. Note that while simulation information in the graphs of the pdf are representative through the end of the season, the “sidedress recommendation” is not usable beyond the reasonable sidedress/rescue application window (no later than tasseling), because of the way the recommendation is calculated.
If the Adapt-N tool is evaluated based on results from long-term N rate experiments, please be aware of possible confounding factors:
- First, if these plots were used to develop the comparison recommendations, then they do not provide an independent evaluation. I.e., depending on conditions, the plots may naturally bias towards the recommendation system that was developed from these plots.
- Second, be aware that long-term experiments sometimes involve the same N rate treatments on the same plots for many years in a row. This implies that carryover effects may occur, especially over multiple years. For example, residual nitrogen may carry over to the next growing season, and biomass returns to the soil will differ. Thus, repeated corn yield differences in plots of low versus high N rates generate differences in organic matter cycling and soil quality. This in turn may impact the calculation of the optimum N rate.
For these reasons, we believe that the best comparisons are made under real-world management scenarios and conditions and locations that are independent from past experimental influences.
If you perform your own statistical analysis of the trial results, be aware that limited statistical power and relatively subtle yield differences for a single replicated experiment may reduce the ability to find statistically significant differences among N rates. This does not necessarily imply that an agronomic response was not present, but that it is difficult to prove its statistical significance if the yield difference was not pronounced, or the field was very variable. This becomes less of a problem with the combined analysis of multiple field trials, allowing for a more robust evaluation of the treatment effects. Also, we have evaluated past trials by incorporating both input and output effects of the treatments, yielding a partial profit analysis.
We hope this is helpful. We very much appreciate the efforts of many field collaborators who are helping evaluate and improve the Adapt-N tool. As always, we are very eager to receive your trial results and answer your questions.
The Adapt-N Team