Including the freshwater landscape in a multi-themed regionalization system to capture macroscale patterns

Background/Question/Methods: Regional to continental-scale ecology requires approaches for capturing patterns and processes operating at broad scales. Many scientists and managers do so by using an existing landscape classification system that partitions the landscape into contiguous regions based on similarities in a set of landscape features (i.e., a regionalization). Many such regionalization systems were created. However, many past efforts failed to explicitly include freshwater landscape characteristics, used subjective methodologies, and included a limited number of landscape features. With the increase in widely-available digital data, we can now include many landscape features in regionalizations, including those that characterize the oft-ignored freshwater landscape. We created three regionalizations by grouping 52 landscape features into themes that characterize the natural terrestrial landscape, the natural freshwater landscape, and an integrated freshwater and terrestrial landscape. We created regions using applied constrained spectral clustering, and compared the results among the three regionalization. We asked: 1) How do the regions partition the landscape? 2) How do the region boundaries differ? We also present a case study of >7,500 lakes distributed across 1,600,000 km2 to demonstrate how well these regions, created to maximize within-region similarity in landscape features, also minimize within-region variation among ecosystem state.

Results/Conclusions: Using principal components analysis as a variable reduction approach, we found that 17, 10, and 24 PCA axes were needed to capture 85% of the variation in the natural (a) terrestrial landscape, (b) freshwater landscape, and (c) integrated landscape, respectively. Using these PCA axes, we clustered USGS 12-digit hydrologic units to make contiguous regions for each of the three regionalizations. We did not assume an optimum number or size of region. Instead, we used an iterative process, and varied the number of regions from 5-1000 in order to maximize within-region similarity in landscape features. For each of the three regionalizations, 100  was identified as the optimal number of regions. This optimal number was based on comparing  fitted curves of the within-region sum of squares error as region number increased to that from contiguous regions made using random start seeds. The region boundaries from the natural freshwater and natural terrestrial regionalizations differed greatly, with the integrated freshwater and terrestrial landscape most closely resembling the natural freshwater landscape.This result points to the importance of including the freshwater landscape in regionalization efforts, as well as how a multi-themed regionalization system that includes the freshwater landscape can capture dominant macroscale patterns.