I want to know how many "tall" trees (greater than 30 m) fell down between 2005 and 2015. I have LIDAR imagery from 2005 and 2015 from an area where many tall trees are known to have fallen due to ice storms.

I created 2 tree height rasters, one for 2005 and one for 2015, by subtracting the ground height from the canopy height (I originally had 4 rasters). What I want to know is the total number of pixels that declined by at least 20 m in the period 2005-2015.

I've tried a variety of approaches with this and haven't quite figured it out. Does anyone have any good hints for me?

(using ArcGIS 10.2.1)

  • 2
    Reduce to a binary raster using CON then extract by mask with the 2015 binary extracting the 2005 data. As you want to know what cells had 20m+ trees in 2005 but don't any more find the cells that don't in 2015 then use them to extract the ones that did in 2005. Be sure when you create your original rasters to specify the same cell size and use snap raster environment setting to ensure the cells align. Apr 10 '15 at 4:38
  • With some more explanations @Michael's comment can be an answer. I think the most important part is the last sentence since different cell sizes and alignments may yield inaccurate results. Still I do not quite understand why a simple Subtract operation does not give the right result. alaskasaurusrex, if the comment does not solve your problem, could you elaborate more on tried a variety of approaches with this.
    – fatih_dur
    Apr 10 '15 at 5:16
  • 1
    @MichaelMiles-Stimson Your comment would make a great answer;)
    – Aaron
    Apr 10 '15 at 18:10
  • 1
    @Aaron, I like Jeffery's answer except that it should be noted that the calculation is wrong. It's not looking for the difference between 2005 & 2015, it's not a growth measurement; I think it should be h2005 > 30 && h2015 < 30 as the conditional statement. Apr 12 '15 at 21:58

Just use the ArcGIS raster calculator with a CON statement.

CON(("h2005" - "h2015") >= 30, 1, 0)

This will result in a binary raster where [1] represents differences of >= 30m and [0] no change at this threshold. And yes, set your analysis environment for extent and snap raster.

  • 2
    Yes, I completely agree. I was applying the OP's threshold. I would imagine that it would be much more tractable to conduct the analysis on "binned" lidar returns and not rasterized data. There no way to track the error/bias when conducting an analysis on rasterized data whereas, the lidar point data would allow a much more specific analysis where error variance could be quantified. I would also be looking for increases in openings and not changes in "individual trees", which requires a specialized algorithm to identify. Apr 10 '15 at 20:17
  • Smaller cells mean better results in this case; it depends on the quality of the LiDAR - how many pulses per square metre, how many returns per pulse, I would think that a cell size that had around 10 pulses on average would be good, the result raster could be filtered later to a larger cell size if it gets unwieildy. I think though if a tree fell and is replaced by another tree 20m+ over the 10 year period it wouldn't be bais... after all you're not trying to track individual trees only the existence of a tall tree not necessarily the same one. Apr 12 '15 at 21:54

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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