|
Rolla, ND | Now as far as grid size. It's a little bit of a balancing act. Generally, I like something that is one half to one third implement width. That seems to work well, and with the smoothing that occurs, you get rid of some of the abruptness of changes that large grids make happen.
Step one is to go through your yield data. Some maps take a lot of cleaning others don't. An article at http://plantsci.missouri.edu/soyx/soyfacts/yield_data_guide.pdf talks somewhat about the whys and wherefores.
I like the FieldAnalyst from South Dakota State available at: http://plantsci.sdstate.edu/precisionfarm/paper/publicationSoftware...
They also have an article on cleaning data at http://plantsci.sdstate.edu/precisionfarm/paper/Carlsonfactsheets.a...
Basically what you do with data cleaning is get rid of points that are off. For instance on a lot of yield maps, you will see a few points where the combine stopped and they show hundreds of bushels an acre on that point, you want to remove that and use the software to generate something closer to reality.
Headlands are probably the biggest area of fixing maps. Either because of lots of low readings as the combine turns or those lower than actual readings as grain starts to come in.
Even after the software is through, take a look and see if there is a swath or so that have obviously low data. Say the combine wasn't getting a full swath or something. Relatively easy to just select the points and delete.
Much of it is due to what in my mind is an incorrect way of handling point data that goes back a long ways. And it's certainly understandable as to why the software uses it. Let's say you have two points right on top of each other that together represent the yield for a spot, one is 100 bushels, the other is zero, what's the yield, well actually, it's 100 bushels per acre, but averaging will give you 50. It's because the software doesn't take an area, and add up the total grain to get yield, the software takes an area and computes yield using computed yields, and those original computed yields are based on certain things like a certain header width. In the middle of the field, it usually doesn't make much difference.
I don't think there's any real way to do it the way I would like because in order to do it, every point would really need to be time stamped so you can really use the distance to compute the bushels and you'd also use actually distances to other points and not the assumed width of the header you enter into the yield monitor. And then you would instead of assigning all the yield to that point right where the data is recorded, you would use a curve for when grain actually comes to the monitor as compared to when it is cut and that curve would spread the grain back out along the path of the combine more closely like it was when it was cut. Really in the basic information files the data is there to do that, I think.
Anyway, that's beyond what we are doing here.
So start with your data, run it through cleaning, then generate a suface. Inverse distance weighted seems to work pretty well with yield data. Then generate the application map.
Marv
Edited by BigNorsk 10/5/2007 12:41
| |
|
Something for Everyone to Shoot Holes IN !! Moved From Crop Talk.
