Modeling of Basin Scale Hydro-meteorological association by Hybrid Wavelet-ARX approach

Modeling the interaction between meteorological and hydrological variables (hydro-meteorological association) is challenging owing to its high spatiotemporal variability. However, the reliable modeling of these hydro-meteorological association can help in ensuring future water security under changing climate, and it might reduce the cost of managing water resources. These associations are expected to evolve with time. Hence, analysis of the meteorological and hydrological variable at their constituent wavelet level might help in modeling of underlying association between them. This article reports the research findings of a recent study (Suman and Maity, 2019), in which a method based on Multi-Resolution Stationary Wavelet Transformation (MRSWT) is used for transforming the variables (target variable: hydrological variables; forcing variables: meteorological variables) to their wavelet components. The memory of the components of the target variable is modeled by a kernel-based auto-regressive (AR) model and the prediction residuals are modeled using auto-regressive model with exogenous inputs (ARX). The MRSWT components of the meteorological variables are considered as the exogenous inputs. The predicted components of the target variables are inverse-transformed to obtain its predicted value. This hybrid Wavelet-ARX approach is applied for predicting total monthly precipitation over Upper Mahanadi Basin using 16 predictor meteorological variables. The efficacy of the model (compared to other modeling frameworks, such as ARX, Vector ARX) in modeling hydro-meteorological association is observed given the poorly associated hydro meteorological variables. Additionally, a relative importance analysis (RIM) framework in the context of the model is formulated using dominance analysis (DARIM) and Birnbaum Importance Measure (BIM). These RIM frameworks help in separating a set of predictor variables, which have stronger hydro-meteorological association with total monthly precipitation compared to other meteorological variables. Under these frameworks, five most important meteorological variables with the strongest hydro-meteorological association are selected, and the model is again trained using these five inputs. The effectiveness of RIM frameworks in selecting predictors with stronger hydro-meteorological association is observed as the similar model performance is obtained with five selected predictors. Hence, hybrid wavelet-ARX model can effectively model hydro-meteorological association, and RIM frameworks can help in figuring out the predictors with the stronger hydro-meteorological association, leading less complexity and computation requirement in modeling. The developed model is suitable for extracting meteorological forcings and is desirable in a changing climate.

 

Keywords: Hydro-meteorological association; Rainfall prediction/simulation; Climate change; Hybrid Wavelet-ARX model; Relative Importance analysis.

 

Reference: Suman, M. and Maity, R., 2019. Hybrid Wavelet-ARX approach for modeling association between rainfall and meteorological forcings at river basin scale. Journal of Hydrology, 577, p.123918.