Hydroclimatic time series analysis and clustering at multiple time scales

Detailed investigations across time scales and variable types can progress our understanding of hydroclimate. In this work, we analyse temperature, precipitation and streamflow time series at nine time scales (i.e., the 1-day, 2-day, 3-day, 7-day, 0.5-month, 1-month, 2-month, 3-month and 6-month ones). The analyses are performed over the continental United States, and in terms of temporal dependence, temporal variation, “forecastability”, lumpiness, stability, nonlinearity (and linearity), trends, spikiness, curvature and seasonality, among others. Thus, they facilitate extensive characterizations of the cross-scale properties of the temperature, precipitation and streamflow variables. Based on these characterizations, various similarities and differences are identified between the examined variables regarding the evolution patterns of their features with increasing (or decreasing) time scale. Moreover, the computed features are used as inputs to unsupervised random forests to detect any meaningful clusters between the time series. The clustering is performed separately for each set {time scale, variable type}, and allows the investigation of the spatial variability of the temperature, precipitation and streamflow features across the examined continental-scale region and across time scales, with the spatial patterns emerging from it being largely analogous across time scales. Lastly, explainable machine learning is applied to compare the features with respect to their importance-usefulness in the clustering. For most of the features, this usefulness can vary to a notable degree across time scales and variable types, thereby implying the need for conducting multifaceted time series characterizations for the study of hydroclimatic similarity.