Application of Empirical Mode Decomposition for Identification and Characterization of Long-Run Persistence in California Reservoir Inflows

Abstract:

This research quantified long-run hydrological persistence in two important regions for the California water supply system using a relatively new approach in hydrology, Empirical Mode Decomposition (EMD). EMD is a user-adaptive method that identifies non-stationary persistence, as Intrinsic Mode Functions (IMFs), prevalent in hydrological data but often not represented in current water supply modeling. The primary objectives of this research included: 1) examination of variability, patterns of persistence, and the degree of cross-correlations among annual and monthly inflows in these two regions, 2) identification of potential sources of persistence in inflows, particularly modes of climate variability, and 3) to provide water supply managers with suggestions to incorporate these methods into water supply modeling. Results indicate that inflows are highly variable, highly cross-correlated within each region and between the two regions, and moderately correlated to long-run patterns of persistence. Average annual IMF periodicities of >10 years contributed to a considerable amount of variance in annual inflows for the two regions (27.05% and 14.92% for the Oroville and Shasta/Trinity inflows, respectively) and exhibited the strongest moderate (on average 0.50) correlations to climate IMFs of similar length. Monthly correlations of inflows to climate indices were generally weak, but also exhibited the strongest correlations in IMFs with average periodicities of >10 years. This research underscores the need to utilize uncorrelated sources of water, such as desalination, into water management plans and to incorporate long-run persistence into water supply scenario modeling to increase reliability, especially during prolonged periods of hydrologic downturns. The EMD analytical method is a powerful and approachable tool that can be adapted for water supply managers to use as a precursor to this future scenario modeling.