|
Missouri | It is good to know that everything else went well.
Other than the time spent performing the calibration, calibrating 6 machines (assuming all 6 are in good mechanical condition, with no defects that affect monitor accuracy) would offer no substantial difficulty. An average calibration takes 2 hours, and results in a combine that will read totals within 1%-2% of scale tickets. With multiple machines, that error might double. If one was off 1-2% high, and another 1-2% low, you've doubled the range of your error to 2%-4%, but you'll never know it by comparing scale tickets, since the errors offset. In this way, it is actually possible to have a greater degree of error in your map than the totals indicate. The same way a planter can have 3% skips and 3% doubles, but still has 100% correct population.
(standard disclaimer...my experience is with Ag Leader yield monitors, Greenstar may be similar, or different, in these regards).
The issue I would see is that errors on one machine are not necessarily linear, or even constant. These yield monitors are calibrated over a range of flow rates, and can be accurate at one flow rate, off at another. A machine with a poor calibration could be 100% on target at a feed rate of 1000bu/hr, only counting 80% at 1500bu/hr, back 100% on target at 2000bu/hr, and counting 120% of actual yield when feeding it 2500bu/hr. This introduces additional error potential if you choose to apply a linear correction based on a snapshot analysis of one pass compared to another pass of another machine of uncertain accuracy.
Usually, when a single combine is inaccurate, it's yield map still has value, because even though the number of bushels is off, it's still pretty accurate about indicating "Area A" of the field yielded better than "Area B". In fact, I have printed yield maps before and after a machine is calibrated. The totals get updated. The legend of the map gets updated. The overall layout of the map, looks nearly the same. In most of these instances, that yield map still has value. Even though the total number may not jive, it shows where the field did better, and where it did worse. Having multiple machines with mulitple unknown accuracies substantially muddies the water.
Normalizing the data from each machine so they each show the same average yield looks, on the surface, like a good idea, but it's based on the assumption that the average yield seen by every machine was the same. What if one machine spent more time in a higher/lower yielding area of the field? Little things like maybe the machine A in the field cut out all the terrace pans, or operator B preferred to harvest the lower yielding area to cover more acres while spending less time unloading. Anything that caused one machine to harvest a different total average yield is something you will be unable to account for in analysis.
Nothing you can add to the data will improve accuracy, but it is possible to eliminate errors if you can find a way to consistently identify them. A machine logging an empty swath is one of the very few examples I can think of where post processing helps. That is an error that can be identified, and therefore eliminated. I haven't used your particular software, but if it is like others, it should have an option to filter datapoints below a set yield threshold. If you set it to throw out anything showing less than 10 or 15 bu/ac (or whatever figure seems right), that ought to do it.
If it were me, I would probably scale the data to get each combine in a similar yield range, and I'd still consider that yield map *useful*, but I wouldn't call it *accurate*.
I hope mapshots has a better answer for you. Please share what you find. I'm always hoping to find better ways to clean up yield data. | |
|
Cleaning up yield data
