find land cover differences

[johnpolo]

I am using National Land Cover Data from Multi-Resolution Land Characteristics Consortium​. I am looking at central states for changes in forest versus grassland cover classes between 2012 and 2015. The data are rasters with land cover classes that have names and values for the classes and area values for the class coverages. How do I find the change from, for example, mixed forest (#143) to shrubland (#152), during the years mentioned above? I am hoping to get a new raster that will show where the changes occur and that includes a table with some kind of count or indication for the amount of land changed. This is probably a basic type of analysis, but something totally new to me. Would I do this with Raster Calculator or another function in the Spatial Analyst toolbox? I'm not even sure how ask the calculator for a specific class in a field, in a layer. For example:

Layer_2012["Class_Name"="Shrubland"]

or some such.

[DanPatterson_Retired · Accepted solution]

The conditional operator ... Con ... is the tool to use, if you are looking for a particular change combination

Con—Help | ArcGIS for Desktop

I like ... Combine ... Combine—Help | ArcGIS for Desktop

because it 'combines' all combinations of two input classes and produces a table and a raster showing where things were the same, where the differences occured and what those differences were. Reading the table is the key to understanding so take some time to understand the visuals in the help topic.  Once that table and resultant raster are analysed, you can use the Con tool to produce rasters of different types.

I should point out that simple useage in the Raster Calculator can check for equality and other operators.

Raster Calculator—Help | ArcGIS for Desktop

There will rarely be one tool that does everything you need, and sometimes it is easier to breakdown an analysis into steps so that you are clear on the outputs as you go along... in that vein, there are basic 'math' style tools with the math toolset of the spatial analyst tools

[johnpolo]

Dan, I played around with Con and Combine. I think the both of those will do what I would like. I appreciate the help.

[JeffreyEvans]

If you would like to put this analysis into a quantitative framework I would recommend stepping out to the free software Map Comparison Kit. For this problem, the two models to investigate are the Fuzzy Kappa and Fuzzy Weighted Kappa. The Kappa statistic will evaluate the proportion of change corrected for chance agreement and fuzzy approaches account for inherent uncertainty or error in the discrete process. There is just nothing like this available in the common suite of GIS software.

The Weighted Kappa is of particular interest. The model allows you to define a transitional weights matrix that provides, equal, less or more, emphasis for certain change trajectories. An example of where this would be relevant is weighting transition from a certain vegetation class to urban higher than a transition to a different vegetation class.

When I review manuscripts that evaluate change direction using classified data (eg., NLCD) one thing that I look for is if the authors evaluated the significance of their results or accounted for uncertainty. If not, I often find the results suspect because model error and random agreement are not accounted for. Another approach would be a t-test of the change but, you certainly need something providing statistical support. If this is a qualitative assessment, then I would not worry.