Delineation of Phenoregions in Geographically Diverse Regions Using k-means++ Clustering: A Case Study in the Upper Colorado River Basin

Abstract

Phenological monitoring and modeling over a geographically diverse area has long been a problem due to the spatially variable nature of phenological forcing. Phenoregions, which are phenologically and climatically self-similar clusters, have many potential benefits as a geographic unit for monitoring and modeling of vegetation dynamics. This research develops an improved method to delineate regions of similar phenological forcing in geographically diverse regions, using the Upper Colorado River Basin (UCRB) as a case study. Principal component analysis plus k-means++ clustering are adopted to delineate phenoregions in the UCRB, using variables related to elevation, temperature, precipitation, soil, and vegetation history. Raster data at 1 km spatial resolution are used to extract these variables. A series of hierarchical, non-nestable phenoregion maps is generated. The optimal phenoregion map is selected based on spatial homogeneity and spatial concordance with other phenoregion maps generated using different numbers of clusters. This series of phenoregion maps can be considered as a framework for phenological modeling and monitoring, as well as for useful potential vegetation delineation, natural resource conservation, and policymaking.