Computing Change

To compute the change from 1992 to 2016 I used the ‘combine raster’ function in ArcGIS. The resulting layer stores land cover values from 1992 and 2016 for each cell. Since the resolution of the two data sets was different, the resulting combined layer used the coarser of the two resolutions, which was 1km. While there were many interesting transitions in the new combine layer, for this project I was interested in three: cropland lost, cropland gained, and urban land gained.

I used an SQL to isolate the transitions of interest. For example, the SQL for cropland lost was ‘select where 1992 is equal to cropland AND 2016 is not equal to cropland’. This returns all of the transitions for cropland to other land cover classes while excluding the cropland that remained cropland. I then reclassified the combined data to remove all transitions except those highlighted in the SQL. The resulting data set showed all cells where, for example, cropland was in 1992, but not in 2016 as well what class the cropland became.

Transition Matrix

The combine raster provided visual, qualitative analysis of land cover change, but I also wanted to quantify how much the land was changing. To do this I generated a transition matrix using a Python notebook in ArcGIS. A transition matrix, also called a pivot table, simplifies a complex table by eliminating redundancy (ArcMap tools pivot table, esri). It quantifies how classes change between two related data sets. Regarding land cover, the transition shows what percentage of land cover remained in the same class from 1992 to 2016 and what percentage changed classes.