POME-copula for hydrological dependence analysis

D. Liu, D. Wang, Liang Wang, Yongping Chen, Xi Chen, Simin GuProceedings of the International Association of Hydrological Sciences 2015 · DOI

Abstract

Hydrological multivariate analysis has been widely studied using copula-based modelling, in which marginal distribution inference is one of the key issues. The main object of this study is to discuss the applicability of the principle of maximum entropy (POME) in marginal distribution inference, thus to develop a POME-copula framework to analyse the dependence of hydrological variables. Marginal distributions are derived with the POME approach before bivariate copulas constructed with corresponding parameters estimated by the dependence of the derived margins. The proposed POME-copula has been employed in hydrological dependence analyses, with the annual maximum streamflow and water level collected from the Yangtze River, and the monthly streamflow from the Yellow River. Results show that the POME-copula method performs well in capturing dependence patterns of various hydrological variables.

Introduction

Assessing hydrological risks, such as catastrophic flooding, necessitates the simultaneous analysis of multiple variables, including peak discharge and water level. Traditional multivariate distributions often impose a single distribution family on all variables, which frequently fails to capture the inherent skewness and complex dependencies of hydrological data. While copula theory offers a solution by decoupling marginal distributions from the joint dependence structure, the selection of these margins has historically been a subjective process, often relying on "best-fit" guesses from a limited set of standard distributions.

This study introduces the POME-copula framework, which utilizes the Principle of Maximum Entropy (POME) to objectively derive marginal distributions directly from the data's statistical moments. This approach minimizes subjective bias and provides a more flexible modeling environment for complex environmental systems.

Methodology

The POME-copula framework follows a two-step procedure:

  1. Marginal Distribution Inference via POME: Instead of assuming a predefined distribution shape, the researchers employ POME to find the most unbiased probability density function (PDF) that satisfies the constraints of the observed data's first three moments: mean, variance, and skewness. This ensures the model captures the essential characteristics of the river data without overfitting.
  2. Dependence Modeling via Archimedean Copulas: Once the margins are established, they are linked using Archimedean copulas—specifically the Clayton, Frank, and Gumbel varieties. These functions describe the joint behavior of the variables. Parameters are estimated using rank correlation coefficients (Spearman’s $\rho$ and Kendall’s $\tau$), and the fit is validated using Kolmogorov-Smirnov tests for the margins and $Sn$ statistics for the copulas.

Results and Discussion

The framework was evaluated using two datasets: annual maximum streamflow ($S1$) and water level ($S2$) from the Yangtze River (1950–2008), and monthly streamflow data ($S3, S4$) from the Yellow River (1998–2012).

Marginal distributions of normalized hydrological variables derived using the POME approach for (a) Yangtze water level, (b) Yangtze streamflow, and (c) & (d) Yellow River streamflow stations.

The POME-based margins achieved high p-values in Kolmogorov-Smirnov tests, indicating an excellent fit to the observed data. Notably, POME proved highly effective at capturing the strong positive skewness characteristic of the Yellow River's streamflow. Among the copulas tested, the Frank copula consistently provided the best fit for both river systems, outperforming Clayton and Gumbel models in capturing the symmetric dependence structures.

For the Yangtze River data, the POME-Frank model generated simulated correlations ($r = 0.94$) that almost perfectly matched the observed data ($r = 0.95$). Visual analysis of density plots confirmed that the model accurately captures "tail dependence," which is critical for modeling extreme events like floods.

Conclusion

The POME-copula framework provides a robust, objective method for hydrological dependence analysis. By letting the data's own moments dictate the marginal distributions, it avoids the biases inherent in traditional distribution selection. Future work will explore extending this framework to higher dimensions (multivariate beyond bivariate) and incorporating higher-order moments as POME constraints to further refine model accuracy in complex datasets.

Summary updated

Abstract

Hydrological multivariate analysis has been widely studied using copula-based modelling, in which marginal distribution inference is one of the key issues. The main object of this study is to discuss the applicability of the principle of maximum entropy (POME) in marginal distribution inference, thus to develop a POME-copula framework to analyse the dependence of hydrological variables. Marginal distributions are derived with the POME approach before bivariate copulas constructed with corresponding parameters estimated by the dependence of the derived margins. The proposed POME-copula has been employed in hydrological dependence analyses, with the annual maximum streamflow and water level collected from the Yangtze River, and the monthly streamflow from the Yellow River. Results show that the POME-copula method performs well in capturing dependence patterns of various hydrological variables.
BibTeX
@inproceedings{liu2015pome,
  author = {D. Liu and D. Wang and Liang Wang and Yongping Chen and Xi Chen and Simin Gu},
  title = {{POME-copula for hydrological dependence analysis}},
  booktitle = {Proceedings of the International Association of Hydrological Sciences},
  year = {2015},
  doi = {10.5194/piahs-368-251-2015},
}
Figure