Delineation of climate regions using in-situ and remotely-sensed data for the Carolinas

Climatologically homogeneous regions in the Carolinas were delineated using a multi-step approach integrating in-situ and remotely-sensed data. We adopted a consensus clustering technique that obtains climate regions for precipitation and temperature separately. Both average linkage hierarchical and k-means non-hierarchical clustering methods were used to create weather station clusters. Using the resulting precipitation and temperature clusters as training data, we performed a machine-learning decision tree classification of remotely-sensed data (i.e., MODIS and TRMM) to map five precipitation classes and seven temperature classes for the Carolinas. These data were intersected to produce 17 consensus clusters for the Carolinas, and 16 climate regions when summarized by counties. The resultant climate regions showed rational climate regionalization reflecting controls on Carolina climate including topography, latitude, storm tracks, and proximity to the Atlantic Ocean. The use of remotely-sensed data effectively helped the delineation between weather station clusters and even detected consensus clusters that were not identified by intersecting weather station clusters grouped using only in-situ data. We compared the regions with the 15 existing National Climatic Data Center climate divisions using within- and between-cluster standard deviations for both in-situ and remotely-sensed data. Climate regions could improve the existing climate divisions in delineating climatologically homogeneous regions and in separating heterogeneous regions.close121