Non-stationary modelling framework for regionalization of extreme precipitation using non-uniform lagged teleconnections over monsoon Asia

Abstract

Global warming has increased the spatio-temporal variations of Extreme Precipitation (EP), causing floods, in turn leading to losses of life and economic damage across the globe. It is found that EP variability strongly correlates with large-scale climate teleconnection resulting from ocean–atmosphere oscillations. In this study, the Non-Stationary Generalized Extreme Value (NSGEV) framework is used to model EP for high resolution daily gridded (0.5° latitude $\times$0.5° longitude) APHRODITE dataset over Monsoon Asian Region (MAR) using climate indices as covariates. The proposed framework has three major components (i) Selection of non-uniform time-lag climate indices as covariates, (ii) Regionalization of NSGEV model parameters, and (iii) Estimation of zone-wise EP changes. According to Akaike Information Criterion (AICc), results reveal that the NSGEV model is prevalent in 92% of the grid locations across MAR compared to Stationary(S) GEV models. The Gaussian Mixture Model (GMM) clustering algorithm has identified six zones for MAR. It is observed that the derived zonal parameters of NSGEV model is able to mimic the EP characteristics. Further, zone-wise estimation of EP changes for selected return periods shows that the relative percentage change in intensity ranges between 4 and 11% across the six zones. The change in EP is significantly higher in the monsoonal windward and coastal regions when compared to the other parts of MAR. Overall, the intensities of the EP across MAR are increasing, and return periods are decreasing, which can majorly impact on planning, design and operations of the water infrastructure in the region.

Appendices

Appendix

Cross-correlation

Cross-correlation analysis determines the degree of similarity between two time-series over a time-lag range. The cross-correlation coefficient $({\mathrm{r}}_{\mathrm{k}})$ indicates the strength of the association between two different time series. A strong negative or positive association is determined by the correlation coefficient's proximity to − 1 or 1. Equation (19) defines the correlation coefficient between two-time series ${\mathrm{Y}}_{1}$ and ss ${\text{Y}}_{2}$. with a time lag (K). (Coles 2001)

$${\text{r}}_{{\text{k}}} = \frac{{\mathop \sum \nolimits_{{{\text{t}} = 1}}^{{{\text{T}} - {\text{K}}}} \left( {{\text{Y}}_{1} \left( {\text{t}} \right) - { }\overline{{{\text{Y}}_{1} }} \left( {\text{t}} \right)} \right)\left( {{\text{Y}}_{2} \left( {{\text{t}} + {\text{K}}} \right) - { }\overline{{{\text{Y}}_{2} }} \left( {\text{t}} \right)} \right)}}{{\sqrt {\mathop \sum \nolimits_{{{\text{t}} = 1}}^{{\text{T}}} \left( {{\text{Y}}_{1} \left( {\text{t}} \right) - { }\overline{{{\text{Y}}_{1} }} \left( {\text{t}} \right)} \right)^{2} } \sqrt {\mathop \sum \nolimits_{{{\text{t}} = 1}}^{{\text{T}}} \left( {{\text{Y}}_{2} \left( {\text{t}} \right) - { }\overline{{{\text{Y}}_{2} }} \left( {\text{t}} \right)} \right)^{2} } }}$$

(19)

where $\overline{{{\text{Y}}_{1} }} \left( {\text{t}} \right)$ and $\overline{{{\text{Y}}_{2} }} \left( {\text{t}} \right)$ are the mean for each time series.