Moving forward with the disturbance centroids, I ran two other analysis tools in ArcMap that identify not only spatial clustering, but clustering of similar numeric (not necessarily continuous) variables attributed to the centroids. These tools are Cluster and Outlier Analysis (Anselin Local Moran’s I) and Hot Spot Analysis (Getis-Ord Gi*). While these tools essentially answer the same question (where are clusters of similar values?) they calculate this in slightly different ways. See this link for more information. For both tools I used year of disturbance as the input value to be evaluated:

The map on the left shows the results of the Local Moran’s I tool. High-High and Low-Low Clusters represent statistically significant clusters of high and low values, whereas High-Low and Low-High clusters are outliers representing either high values surrounded by low values, or vice versa. The results of the Hot Spot Analysis show similar patterns of significant clustering by year. These patterns are indicative of geographic shifts in timber harvest concentration, which could reflect management decisions. Using “Select by Location”, the land use designations associated with significant Moran’s I clusters were tabulated. Low-Low Moran’s I clusters are on the left (1985-1989) and High-High clusters (1996-2012) on the right:

Here we see an expected increase in clustering of clear-cuts and partial harvests on Non-Forest Service (private) and Matrix lands, yet it is important to keep in mind that these land-use designations did not exist until 1994.

Following this analysis of clustering by year, I was interested in attaching continuous variables to the actual disturbance patches, rather than their centroids. Toward this end, I brought in an another output from the LandTrendr algorithm: disturbance magnitude, which is a function of the slope of change in spectral values for each pixel over a yearly time step. Since magnitude is measured on a per-pixel basis, I used the Zonal Statistics tool in ArcMap to calculate the mean disturbance magnitude within each patch, and attach it to the patch attributes. I then ran Hot Spot Analysis again, with mean disturbance magnitude as the input value.

The distribution of high and low magnitude disturbances is interesting, especially when overlaid on the land use designations. As shown in the map on the right, a cluster of high magnitude disturbance is associated with a section of adjacent private lands (in grey) and a cluster of low magnitude disturbance is associated with Matrix land (in orange). This may indicate a higher proportion of clear-cutting (high magnitude disturbance) on private lands, and more partial harvesting (lower magnitude disturbance) on Matrix lands.