How Do I Get Value From My Data? (part 2)
Using data to answer these sort of questions effectively means doing things like experiment design to control for multiple variables, statistical analysis, collecting huge data sets that span multiple climatic variations, and many other complex tasks. It also means knowingly doing things that are sub-optimal.
This Part 2 of 2 blog posts on How Do I Get Value From My Data?
For example, let’s assume that you think the right thing to do this season is spray fungicide. However, you’re not really sure, so the natural thing to do is leave areas where you don’t spray fungicide and compare them with areas you did spray it. That means you are voluntarily choosing to NOT do the “best” thing in exchange for learning something. Seems reasonable for just this one thing. But what about variations due to soil type? Nitrogen? Planter population? Seed variety? All these things vary across all your fields, too. And don’t forget about those multi-year variables like cover crops and soil compaction. And isn’t it a little drastic to compare maximum application rate to zero application rate? What about half rate? Fourth rate? Double rate? Will it even matter if your yield monitor is only accurate to +/- 10%?
Even with all that, let’s assume you’ve got a good test plan down to evaluate your fungicide. But you have other decisions to evaluate each year, too. You spent extra money to variable-rate apply your nitrogen this year, so you need to run some tests on that system. Those tests are going to interfere with your fungicide tests because a plant deficient in nitrogen may have a different response to fungicide than a plant with plenty of available nitrogen.
Even if we could solve that, how are we going to figure out which variable rate nitrogen map is the “right” one for a field? It’s not like I can have two competing services create prescriptions maps for me, and plant half the field with one map and half the field with the other map: the maps themselves change the application rate based on location by definition! So who’s to say which service creates better prescription maps if I can’t really compare them?
Two things should be immediately apparent from this discussion about evaluating the revenue implications of our daily decisions:
- pooling good data from multiple farms and/or huge areas can give more definitive results than small areas alone,
- and tracking and analyzing all these overlapping trials is going to require automation and software because “experiment design” and “statistical analysis” aren’t ever going to take up the bulk of your day.
Aaron Ault is the OADA Project Lead and a Senior Research Engineer at the Open Ag Technology Group at Purdue University.