I would like to create a map of vegetation based on vegetation surveys made in the field. I have 340 points spread throughout the studied area where I recorded vegetation by strata ( a % for each strata) and a multispectral satellite image of the area.
I would like to use those percentages and link them with an equation or algorithm to the pixels of the multispectral image in ArcGIS. Thus, a combination of bands will have a specific combination of percentages from the strata. I then want to use those combinations to map vegetation everywhere on my satellite image.
I am looking for way to do this on ArcGIS if possible.
Not a solution for ArcGIS but the MaxEnt algorithm allows you to spatially predict class distributions based on collected presence data using predictive input rasters (multi-spectral image in this case, but topographic parameters can also be included).
The package is free (and in the meanwhile open source) and takes separate ASCII files as inputs. They need to have the same pixel size and resolution which can easily produced in ArcGIS.
You can download it here:
https://biodiversityinformatics.amnh.org/open_source/maxent/
The tutorial helps you to understand how to use it:
https://biodiversityinformatics.amnh.org/open_source/maxent/Maxent_tutorial2017.pdf
I think you can use Regression Analysis in ArcGIS to do this task.
You can download tutorial from the ArcGIS help.
https://www.arcgis.com/home/item.html?id=71a65d35688a4502b123cbdfc99afdee
May be it will help you.
If I understand correctly, you are trying to build a spatial relationship between your field data and satellite image.
It can be done with 3 steps:
1. You need to derive the NDVI from the multisperctral image. Since NDVI is proxy parameter of vegetation health the linking will be easy.
2. You need to develop correlation equation for the field data
3.The derived equation you can execute in Arcgis Raster calculator. And reclassify the output raster based on vegetaion strata percentage to obtain the vegetation map.
HTH!
Vijay
I also thought regression analysis could be used as Rao and Vijaysekaran have suggested, provided there is a linear relationship. But I think Samuel, you did not use any optical ground sensor to measure vegetation, did you? I'm not an ecologist, so I am trying to understand what exactly is contained within the 'recorded vegetation by strata'?
What exactly do you mean by "recorded vegetation by strata (a % by strata)"? By vegetation, do you mean species or habit (grass, brush, scrub, forest)? By strata, do you mean vertical stratification? Do you have any end-members (pure pixels)? Not sure how you'd do it in ArcGIS, rather than with an image processing software. You can follow this methodology:
Article Linear Spectral Mixture Analysis of TM Data for Land-Use and...
- Badges
- Science topic
- Similar topics
- Computer Science
- Images