4

I have two maps that originally come from a harvester and a tractor as shapefiles: Corn Yield and Planting Speed. I would like to know what would be the best procedures to statistically answer the two questions below:

  • Is there a correlation between the Corn Yield and the Speed Planting? In other words, does the tractor speed during the planting affects the final corn yield?

  • If yes, what would be the optimal speed to reach the highest corn yields?

PolyGeo
  • 65,136
  • 29
  • 109
  • 338
Lima
  • 71
  • 5
  • Interesting problem... what format are the maps in at the moment? – Michael Stimson Sep 09 '15 at 03:20
  • They come as point shapefile, but if you think an interpolation to raster would be the best (or anything else that I don't know yet), we could head this way. – Lima Sep 09 '15 at 13:13
  • Yes, absolutely, interpolate to raster. This is because your points aren't likely to line up exactly so correlating them would be difficult... you could produce Thiessen/Voronoi Polygons https://en.wikipedia.org/wiki/Voronoi_diagram from your points (both are credited with the work independently, hence the two names) if you want to work with vector overlay. http://gis.stackexchange.com/questions/6971/is-it-possible-to-create-thiessen-polygons-around-nodes-in-qgis has instructions. – Michael Stimson Sep 09 '15 at 21:35
  • You will also have to manage edge effects and a separate effect of lag from when the planter speeds up to when the seed settles on a steady spacing. This speed issue will be more or less important for a vacuum plate / disc planter compared to an air seed distribution planter. Once you've got a plot of the data you might be able to see if there is an issue. – BillW Sep 11 '15 at 03:11
  • Hi Lima, how did you go with this? – BillW Sep 18 '15 at 02:16
  • I followed your instructions to interpolate and use the point sampling that worked pretty well, I could have a spreadsheet with both speed and corn yield for each point. Now I am having difficulty with the regression. I`ll probably have to use a specific statistic software like R for example and still trying to understand how to use it. If somebody could indicate me any good tutorial or another solution would be great. – Lima Sep 22 '15 at 19:04

1 Answers1

4

Firstly make a decision if you will be satisfied with a regression, the type of analysis that can be performed in a spreadsheet program. That regression might be linear or polynomial etc, with a calculation of R2. Then it is a matter of gathering values for each of the layers at a series of points. For repeat analysis, the best practice method is to make a grid first, maybe 10 m x 10 m which is independent of how the data is scattered about. Then use the grid sampling tools in QGIS to pull out the data.

I did this in a similar project although the data was a mix of vector and raster.

EDIT / Additional

Method with QGIS is to make raster surfaces from the vector point datasets and then sample that at points. Returns a shapefile of points from which you pull the corresponding values.

  • I ensure my shapefile(s) of point data from the harvester is in a UTM projection, e.g. EPSG:32755. Obviously have a look at which of the attributes you are interested in and that those units are sensible. The canvas of QGIS ought to be set to this projection as well (some tools are sensitive to this) with OTF on, this is on the bottom right corner of the screen.

  • Use the RASTER|INTERPOLATION tool (it says it is a plugin but I think it is standard) from the QGIS menu to make a surface. This is the step where you can inadvertently alter the data and a good interpolation will produce a surface that is faithful to the original. You will have to find your own way with the settings here. I aimed for a raster with a cell / pixel size nearing 10 m. The real variability in results comes from the type of sampling system used and those settings. See dialog interpolation

  • Make a shapefile grid over the area of interest. Projection same as before. The tool in VECTOR|RESEARCH TOOLS|REGULAR POINTS is great for this, it should result in a grid which fits the boundaries of the data, see image overview of grid

  • Install from the PLUGINS|MANAGE AND INSTALL PLUGINS dialog the POINT SAMPLING TOOL. This will use the grid shapefile and drill into any number of raster layers, you will have two. See this dialog point sample

  • You will now have produced a new shapefile which contains at each row, the values from each of the rasters. Copy the values out of the attribute table into a spreadsheet program and perform the regression.

ciao

BillW
  • 510
  • 2
  • 9
  • Yes, the regression would be a good way, probably a quadratic regression. First I need to put both data at the same spreadsheet. I think thats what you talking about the grid. If I had the same grid for both I could import the yield and speed data at the same point in the map. Do you know how to do it? – Lima Sep 09 '15 at 17:53
  • My preference is to convert the vector point data to a raster surface and extract the values from that. This is based on experience with other workflows and might not be the only way to go. To properly explain it, I'm going to edit the answer, as this comment dialog doesn't let me format etc. – BillW Sep 11 '15 at 01:26