Thursday, January 13, 2022

Half-baked Research 2021: Part 4 - More Precision Cropload Management (aka precision thinning)


 I'm a bit tired of talking about precision thinning (and precision crop load management), have written up much about it in the past:

You get the idea, but, I am always looking for that "magic bullet" to make it more accurate and my life easier. To that end, I tried to follow in 2021 the exact RECIPE as described above. The details:

  • six Honeycrisp trees selected on G.11 rootstock in our 2014 NC-140 Honeycrisp planting at the UMass Orchard in Belchertown, MA
  • 14 flower clusterd tagged and whole tree bloom counted
  • measuring of fruitlets began on 18-May, and were measured 5 more times at 2-3 day intervals (average persisting fruit size on 21-May was 9.2 mm) until the last measurement on 4- June (7 days after the previous measurement, average fruit size of persisting fruits was 24.2 mm)
  • at first, measurements were voice-entered into the Malusim app, however, once it got "flakey" I switched to using the Orchard Tools app, which worked very well.
  • used the Malusim app at first to predict fruit set, however, quickly discovered there were some issues in the way erroneous (too big or too small) measurements were being handled and thrown out, switched to the Ferri spreadsheet, although I did not use the tree top and bottom distinction.

Here is the result from the Ferri spreadsheet and I think it was quite accurate. It's a no-brainer to follow this RECIPE for tall spindle trees and apples of high value such as Honeycrisp, Gala, and Fiji. I'll admit I did not follow it to guide my chemical thinner applications, because someone else chose the rates and timing and made the applications on their schedule. But it worked out.

Actual number of apples per tree at harvest 61! Pretty close to 70 predicted!

So, what's half-baked about it? Well, myself and colleagues at UMass (Dan Cooley and PhD student Paul O'Connor to be exact) are collaborating with researchers at Carnegie Melon University (CMU) on a NIFA funded project: Using Computer Vision to Improve Data Input for Precision Thinning Models in Apples. I pushed for them to use the RECIPE while doing their study of computer vision to measure fruit growth, and on two other varieties (Gala and Fuji) and another Honeycrisp block. So they did, with the exception is that all data was collected by pen and paper, but thanks to our summer help (Evan Krause), was entered and imported into the Ferri spreadsheet to predict fruit set. OK. I ran the model in the Ferri spreadsheet and came up with some very low fruit set numbers. I won't even show you that here, because although it was more-or-less field evaluated, it was by observation only. Even though I believe final fruit set was counted, the data entry source escapes us at this time. (Arghh.) Here was the problem, the student plucked spur leaves off tagged spurs to assist with the computer vision, and I believe that resulted in very poor fruit set (compared to the rest of the tree) on those hobbled spurs. (Hasn't someone a long, long time ago stressed the importance of healthy spur leaves for susbsequent fruit set and size/quality?) It's too bad because otherwise had the help to do a nice set of fruitlet measurements to run in the fruitlet growth rate model. Next year? No half-baked research? (Maybe?) But thankfully -- both to me as the writer and you the reader? -- this is the last half-baked research for 2021!

YouTube video of May 19, 2021 Twilight Meeting at UMass Orchard with CMU robot



Posted by at No comments:
Subscribe to: Posts (Atom)