An Early Warning System for Flood Detection Using Critical Slowing Down

Abstract

The theory of critical slowing down (CSD) suggests an increasing pattern in the time series of CSD indicators near catastrophic events. This theory has been successfully used as a generic indicator of early warning signals in various fields, including climate research. In this paper, we present an application of CSD on water level data with the aim of producing an early warning signal for floods. To achieve this, we inspect the trend of CSD indicators using quantile estimation instead of using the standard method of Kendall’s tau rank correlation, which we found is inconsistent for our data set. For our flood early warning system (FLEWS), quantile estimation is used to provide thresholds to extract the dates associated with significant increases on the time series of the CSD indicators. We apply CSD theory on water level data of Kelantan River and found that it is a reliable technique to produce a FLEWS as it demonstrates an increasing pattern near the flood events. We then apply quantile estimation on the time series of CSD indicators and we manage to establish an early warning signal for ten of the twelve flood events. The other two events are detected on the first day of the flood.

Using persistent homology as preprocessing of early warning signals for critical transition in flood

“…In this research, we set and because our experience indicates that these values give good analytical results. Other studies using these values have also obtained good results 25 , 29 . Thus, we obtain two-dimensional point cloud data from the reconstruction of the phase space.…”

“…Table 4 shows the results of FLEWS from the norm of the zero-dimensional persistence landscapes at the optimum quantile of 12%; the results of FLEWS arising directly from the water level data at the optimum quantile 15% reported in Syed Musa et al 25 , are also provided for comparison. The FLEWS from the norm of the zero-dimensional persistence landscapes at the optimum quantile with threshold values of 4.8991 for the time series of variance and 2.8203 for the time series of the average spectral density at low frequencies succeeds in producing ten early warning signals of the twelve actual flood events.…”

“…Our previous study using water level data of Kelantan River suggested that CSD could produce an early warning signal for flood detection 25 because ten of twelve flood events were preceded by an early warning signal while the two other events were detected on the first day of the flood. Note that, throughout this study, the term of an early warning signal is meant by any signal that is determined before the day of the flood event.…”

“… 17 – 24 . In our previous study, we showed that CSD theory is able to provide early warning signals of flood on the basis of the water level data of Kelantan River 25 . The need to develop a better FLEWS remains an urgent concern on account of the increasing intensity of floods.…”

Practical guide to using Kendall’s τ in the context of forecasting critical transitions

“…A positive Kendall's τ typically means a monotonic increase in the data. This metric has been used to quantify the trend of early warning signals and risk of impending transitions in a variety of systems, particularly in computational [7,8,14] and empirical [17][18][19] ecological and climate systems. However, interpreting the values of Kendall's τ is challenging and requires careful considerations.…”