Maintaining the confidentiality of plot locations by exploiting the low sensitivity of forest structure models to different spectral extraction kernels

Abstract

The United States Forest Service Forest Inventory and Analysis (FIA) unit maintains a large national network of inventory plots. While the consistency and extent of this network make FIA data attractive for ecological modelling, the FIA is charged by statute not to publicly reveal inventory plot locations. However, use of FIA plot data by the remote sensing community often requires that plot measurements be matched with spatially correspondent values from spectral or geographic data layers. Extracting spatial data in a known way and associating it with plot information leaves open the possibility that a user might use extracted spatial characteristics and a moving window filter to directly infer the plot's location. Direct inference of plot location in this way would be impossible, however, if the original method of sampling the geographic data was unknown. Tests using five Landsat scenes covering a wide range of ecological types showed that varying the weights of pixels within approximately 50 m of the plot centre has little effect on the quality of subsequent models predicting basal area. This finding may support the development of automated extraction routines that vary (perhaps randomly) the geographic data extraction process and therefore increase the security of FIA plot locations.