Aiming for the optimum: examining complex relationships among sampling regime, sampling density and landscape complexity to accurately model resource availability

Authors: Ira L. Parsons, Melanie R. Boudreau, Brandi B. Karisch, Amanda E. Stone, Durham Norman, Stephen L. Webb, Kristine O. Evans, Garrett M. Street

Context Obtaining accurate maps of landscape features often requires intensive spatial sampling and interpolation. The data required to generate reliable interpolated maps varies with spatial scale and landscape heterogeneity. However, there has been no rigorous examination of sampling density relative to landscape characteristics and interpolation methods. Objectives Our objective was to characterize the 3-way relationship among sampling density, interpolation method, and landscape heterogeneity on interpolation accuracy in simulated and in situ landscapes. Methods We simulated landscapes of variable heterogeneity and sampled at increasing densities using both systematic and random strategies. We applied each of three local interpolation methods: Inverse Distance Weighting, Universal Kriging, and Nearest Neighbor — to the sampled data and estimated accuracy (R²) between interpolated surfaces and the original surface. Finally, we applied these analyses to in situ data, using a normalized difference vegetation index raster collected from pasture with various resolutions. Results All interpolation methods and sampling strategies resulted in similar accuracy; however, low heterogeneity yielded the highest R² values at high sampling densities. In situ results showed that Universal Kriging performed best with systematic sampling, and inverse distance weighting with random sampling. Heterogeneity decreased with resolution, which increased accuracy of all interpolation methods. Landscape heterogeneity had the greatest effect on accuracy. Conclusions Heterogeneity of the original landscape is the most significant factor in determining the accuracy of interpolated maps. There is a need to create structured tools to aid in determining sampling design most appropriate for interpolation methods across landscapes of various heterogeneity.

Suggested Citation