9:30 AM - 9:45 AM
[SSS13-03] Reproduction of water-level during a flood event using the streamflow noise recorded by Hi-net
Keywords:river discharge, upstream, streamflow noise, Hi-net
Discharge is considered to be the most important physical quantity in river hydrology. For the flood forecasting, it is necessary to monitor the river discharge by continuous and real-time manner. Conventional discharge observation has been carried out through measuring flow speed at various points in a cross section of a river, and by integrating the flow speed along the cross section. As this observation method takes a lot of time and effort, it is not suitable for continuous real-time monitoring. Therefore, we usually establish the H-Q (water-level-discharge) curve at each gauging station assuming a power-law relationship. Since water-level can be monitored continuously in real-time, we can estimate discharge indirectly by applying the H-Q curve to the observed water-level.
However, the frequency of high water-level is generally low, and going into the river at such high water-level situation is very dangerous. Accordingly, the discharge estimated by extrapolating the H-Q curve becomes less accurate as the water level becomes higher. Moreover, as a large number of tributaries branch intricately in the upstream, it is almost impossible to establish the H-Q curves for most of the tributaries.
On the other hand, seismometers installed in the upstream area sometimes detect vibrations excited by rivers, which is called “streamflow noise” in this study. As power of the streamflow noise reflects whole vibrational energy propagates from multiple nearby rivers, it will be available for monitoring discharge of multiple rivers. However, at this stage, it is not understood well under what conditions the streamflow noise can reproduce the water-level and discharge.
Based on the above background, we surveyed conditions to reproduce the water-level using streamflow noise recorded by NIED Hi-net. We employed seismograms of in total 362 Hi-net stations distributed in eastern Japan and the MLIT water-level records that are closest to each Hi-net station. The target period is October 10 to 17 in 2019, within which the Typhoon Hagibis struck eastern Japan.
First, we applied the FFT algorithm to the Hi-net vertical motion records every two minutes and calculated power spectral density every 0.2 Hz. Then we clipped the minimum power spectral density within the range of 10 minutes and 1 Hz to exclude transient signals such as earthquakes and monochromatic signals specific to each Hi-net station. The clipped power spectral density was resampled every 1 hour for frequencies from 1 to 32 Hz with the octave bandwidth. We considered a power-law relationship between the power spectral density and the water-level, and optimized the best-fit parameters through the least square method by maximizing the Nash-Sutcliffe model Efficiency coefficient (NSE: Nash and Sutcliffe, 1970).
As a result, the maximum NSE-values were high in mountainous areas such as Yamanashi, Nagano, Gunma, and west of Fukushima prefectures for all frequency ranges, indicating that the streamflow noise can reproduce the water level well in these areas. On the contrary, the maximum NSE-values were low in the Kanto-plain. We considered two reasons that explain this result: 1. upstream area with steeper river slope causes faster flow speed and excites the streamflow noise more effectively. 2. excitation of seismic noise induced by other than streamflow is lower in the upstream due to lower anthropogenic activity. To test these hypotheses, we plotted a correlograms between the maximum NSE-value, the average slope of nearby rivers, and the seismic noise power before the flood. As a result, we found a positive and negative correlations between the maximum NSE-value and the average slope of nearby rivers, and the seismic noise power before the flood, respectively. This result demonstrates that these factors are important to evaluate reproductivity of water-level using the streamflow noise.
Acknowledgements: This study is funded by The Taisei Foundation 2020 research grant.
However, the frequency of high water-level is generally low, and going into the river at such high water-level situation is very dangerous. Accordingly, the discharge estimated by extrapolating the H-Q curve becomes less accurate as the water level becomes higher. Moreover, as a large number of tributaries branch intricately in the upstream, it is almost impossible to establish the H-Q curves for most of the tributaries.
On the other hand, seismometers installed in the upstream area sometimes detect vibrations excited by rivers, which is called “streamflow noise” in this study. As power of the streamflow noise reflects whole vibrational energy propagates from multiple nearby rivers, it will be available for monitoring discharge of multiple rivers. However, at this stage, it is not understood well under what conditions the streamflow noise can reproduce the water-level and discharge.
Based on the above background, we surveyed conditions to reproduce the water-level using streamflow noise recorded by NIED Hi-net. We employed seismograms of in total 362 Hi-net stations distributed in eastern Japan and the MLIT water-level records that are closest to each Hi-net station. The target period is October 10 to 17 in 2019, within which the Typhoon Hagibis struck eastern Japan.
First, we applied the FFT algorithm to the Hi-net vertical motion records every two minutes and calculated power spectral density every 0.2 Hz. Then we clipped the minimum power spectral density within the range of 10 minutes and 1 Hz to exclude transient signals such as earthquakes and monochromatic signals specific to each Hi-net station. The clipped power spectral density was resampled every 1 hour for frequencies from 1 to 32 Hz with the octave bandwidth. We considered a power-law relationship between the power spectral density and the water-level, and optimized the best-fit parameters through the least square method by maximizing the Nash-Sutcliffe model Efficiency coefficient (NSE: Nash and Sutcliffe, 1970).
As a result, the maximum NSE-values were high in mountainous areas such as Yamanashi, Nagano, Gunma, and west of Fukushima prefectures for all frequency ranges, indicating that the streamflow noise can reproduce the water level well in these areas. On the contrary, the maximum NSE-values were low in the Kanto-plain. We considered two reasons that explain this result: 1. upstream area with steeper river slope causes faster flow speed and excites the streamflow noise more effectively. 2. excitation of seismic noise induced by other than streamflow is lower in the upstream due to lower anthropogenic activity. To test these hypotheses, we plotted a correlograms between the maximum NSE-value, the average slope of nearby rivers, and the seismic noise power before the flood. As a result, we found a positive and negative correlations between the maximum NSE-value and the average slope of nearby rivers, and the seismic noise power before the flood, respectively. This result demonstrates that these factors are important to evaluate reproductivity of water-level using the streamflow noise.
Acknowledgements: This study is funded by The Taisei Foundation 2020 research grant.