ABSTRACT
Characterization of hydrological time series is of paramount importance for planning and management of water resources and the spectral analysis procedures are potential techniques for performing the same. The performance of traditional Fourier-based spectral analysis are not appealing when applied for complex hydrologic time series, which are non-linear and non-stationary in characteristics. Advanced spectral analysis methods like wavelet transforms and Hilbert-Huang Transform (HHT) are two dependable computational tools to perform time-frequency characterization of hydrological signals. The creditability of these techniques is successfully demonstrated with the application on different hydrological problems of teleconnections and simulations of time series of diverse spatiotemporal scales, collected from Indian subcontinent. Wavelet transforms and HHT facilitate the characterization of hydrological signals in a multiscaling framework. Wavelet transforms and HHT are found to be effective in extracting the inherent nonlinear trend and periodicity in the time series. In this study, a novel framework involving HHT-based Time Dependent Intrinsic Correlation (TDIC) is presented for investigating the teleconnections of hydrological variables. This generic framework has the potential applications for hydroclimatic teleconnections as well as the mutual teleconnections of any two hydrological variables. The Multivariate Empirical Mode Decomposition (MEMD) enhance the domain of applications of the traditional EMD and found to be more robust in the hydroclimatic teleconnection studies, as they can perform the decomposition by identifying common scales present in different variables, performs data adaptive decomposition and simple in implementation as it performs the scale separation of all variables in a single step. The TDIC analysis showed that the nature and strength of association between hydrologic variable and climatic index differs with time scale. This information can be effectively captured and used in modeling hydrologic variables. This hypothesis is effectively implemented in the proposed MEMD-Stepwise Linear Regression (SLR) coupled model and the encouraging results showed that the technique is a suitable alternative method for improved hydrologic predictions, by accounting multiple contributory input variables. Eventhough many hybrid decomposition models were proposed in the past, many of them didn’t incorporate multiple inputs and scale-specific information during the modeling, while the proposed method successfully addressed such issues. The applicability of proposed MEMD-SLR model is demonstrated by estimating ISMR, short term drought index (SPI-3), monsoon rainfall in Kerala meteorological subdivision, monthly inflows into Hirakud reservoir and daily suspended sediment load in Indian rivers. The usefulness of MEMD for characterization of rainfall time series is further explored by using it for finding the scaling exponent required for developing hourly IDF curves from coarse resolution (such as daily) rainfall data of different cities in Kerala.
