5

I have some 2014 MrSID files from the National Agricultural Imagery Program (NAIP) that I've converted to TIFF files. I want to use them to find the area of tomato farms in a watershed. I'm using ArcGIS 10.1.

Would an Enhanced Vegetation Index (EVI) or Normalized Difference Vegetation Index (NDVI) be useful for this? Would I even be able to find tomato farms using orthophotos? If there is other data out there that would be better for this task, please let me know.

Improve this question
2
  • are there other crops in your study area ?
    – radouxju
    Commented Oct 21, 2014 at 20:05
  • Yes. The other crops I know of are corn, soybeans, tobacco, and cotton. These are the ones I've seen while doing field work.
    – Hgotwald
    Commented Oct 22, 2014 at 20:06

3 Answers 3

Reset to default
4

There are several pro's and con's for using NAIP in land classification.

Pro's

  • High spatial resolution
  • nIR band is useful for discriminating major vegetation classes (e.g. a crop circle surrounded by arid land)
  • Acquired usually during peak growing season

Con's

  • Low spectral resolution

  • The image acquisition during peak growth season can be counterproductive when trying to discriminate between some land classes (e.g. deciduous trees and grasslands often get confused)

NAIP simply does not have the spectral resolution necessary to discriminate between most classes of vegetation. On the other hand, utilizing EVI or NDVI from NAIP in your classification is a great way to extract healthy green vegetation of any type from the image. Imagery such as Worldview-2 is much better suited for discriminating between types of vegetation--even types of crops!

Unless the tomato farms are growing in the desert surrounded by arid vegetation, you will likely need to do some sort of object oriented image analaysis (OBIA), such as image segmentation and classification. This method segments the image into image objects based on spectral characteristics. You can then classify those image objects by a variety of metrics such as shape, size, texture etc. This essentially adds flexibility to your analysis that you could not get from spectral bands alone. Common programs for image segmentation include: eCognition, SPRING, and some of the add-ons in QGIS.

Improve this answer
3
  • 1
    +1 (and I only have four left today!) for the direct knowledge of the dataset.
    – WhiteboxDev
    Commented Oct 21, 2014 at 20:51
  • So in the research I've done so far I've found out that imagery such as Worldview-2 is only available if you're willing to buy it or work for a federal company. I'm doing this project for my thesis at a state funded university. Do you have any suggestions for infrared data that is free to the public? Or if you've purchased Worldview-2 before is it expensive?
    – Hgotwald
    Commented Oct 23, 2014 at 17:52
  • 1
    If you are looking for free data, I would recommend Landsat-8. I am not sure how large the tomato farms are, but the 30m resolution is likely suitable for your analysis. What Landsat lacks for spatial resolution, it makes up for with spectral resolution. Additionally, there is a wide body of literature to draw on for using landsat for ag detection.
    – Aaron ♦
    Commented Oct 23, 2014 at 18:20
3

I think what you're really looking to do is an image classification. You are specifically looking for one land class, i.e. tomato farms. How many bands of imagery do you have and in which parts of the spectrum do they lie? Hopefully at least one is within the infrared, which is critical for vegetation type classification. If you have very fine resolution imagery then you may be able to improve the classification with texture based indices as well as the brightness data itself, particularly since you are looking at an agricultural land class.

Both supervised (when you specify the spectral characteristics of the land types that you're interested in at the start) and unsupervised (when you let the classifying determine the best clusters) image classification can be carried out in ArcGIS. See here for more details on how to classify multispectral imagery using ArcGIS: http://resources.arcgis.com/en/help/main/10.1/index.html#//00nv00000002000000

Improve this answer
2
  • +1 Texture could be a great way to extract tomato farms surrounded by forests, for example.
    – Aaron ♦
    Commented Oct 21, 2014 at 20:43
  • The NAIP imagery I have is just 3-banded in natural color. Some states have started taking photos with 4-bands with the 4th being near infrared, but in TN they have not done so yet. I will begin looking for some data sets with infrared.
    – Hgotwald
    Commented Oct 22, 2014 at 20:27
1

You can also try Cropscape (you can upload an area of interest and get percentages in that area per crop type). They have data for 2014. http://nassgeodata.gmu.edu/CropScape/

Improve this answer

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.