Adaptation of the visibility graph algorithm for detecting time lag between rainfall and water level fluctuations in Lake Okeechobee. (December 2019)

• Record Type: Journal Article
• Authors: John, Rahul; John, Majnu
• Is Part Of: Advances in water resources. Volume 134(2019)
• Publication Date: 2019-12
• Subjects: Time series -- Visibility graph algorithm -- Cross correlogram -- Water levels -- Precipitation; Hydrology -- Periodicals; Hydrodynamics -- Periodicals; Hydraulic engineering -- Periodicals; 551.48
• DOI: 10.1016/j.advwatres.2019.103429
• Languages: English
• ISSNs: 0309-1708

Abstract

Highlights: Visibility graph algorithm (VGA) may be adapted to detect the time lag between two hydrogeological time series. Cross-correlograms may yield incorrect results if the times series are not pre-whitened; VGA based approach will identify the correct lags without the pre-whitening step. Multivariate simulations revealed that VGA-based approach detects the lags correctly even for small sample sizes if underlying models are of the same type. VGA based approach detected time lags of 4 days as the most prominent lag between rainfall and water level fluctuations in Lake Okeechobee, FL. Abstract: Identifying time-lag between two hydrogeological time series for planning and management of water resources has a long history and is of continuing research interest. Many hydrogeological studies in the past have used visual inspection and cross-correlogram techniques for quantifying the time lag. Cross-correlogram techniques, if not done under the transfer function framework, could lead to ambiguous results. In order to conduct cross-correlogram analysis under the transfer function framework, careful pre-processing steps have to be undertaken, which are often ignored in practice. In this paper, we propose a new approach to compare two sets of hydrogeological time series data using the visibility graph algorithm and show the advantages of using the new approach over the traditional approach. Application of the new approach is demonstrated by assessing the lags between rainfall and water level fluctuation in Lake Okeechobee, Florida. We also present simulation studies to better understand the performance of the method for different sample sizes, different underlying models and in the presence of missing values.